music/benchmark/EAR_SLATE_CH02_V2.json
{
"instrument": "ear_axes",
"version": "ear-axes/1.0.0",
"_honest_tier": "MEASURED RANK, NEVER A QUALITY CLAIM. The reading orders a slate against four declared composition clauses. It gates nothing, promotes nothing, and every axis prints its own measured separation beside its value -- including the three that read at or below chance.",
"law": "A cue's CHARACTER is the change it earns, the way it develops its motif, the way it breathes, and whether more than one voice is ever carrying the tune. A wall that counts band memberships is measuring a different thing and must not be read as this one.",
"what_this_does_not_say": "Nothing here gates a track up or publishes a quality claim. The reading is a RANK WITHIN THE SLATE READ: adding a track changes every other track's percentile, so no value here is an absolute score and none may be quoted as one. THE POOL IS CLASS-BLIND. A class the corpus says should sit narrow (BATTLE dynamic range p90 11.256 against EXPLORATION's 28.979) reads low on dynamic_breath for behaving correctly; a mixed-class slate is not a like-for-like ranking.",
"class_relative_normalisation": "BOARDED, NOT BUILT HERE. The bands are per-class and the composite's percentiles are not: the worked case is Highland Fury (TP_B2_HYBRID_1), which sits INSIDE its BATTLE band on both banded axes that are terms -- structural_yield 1.8889 in [0.71, 2.6133] and dynamic_range_db 6.34 in [5.452, 11.256] -- and still reads rank 7 of 19, because it is being percentiled against EXPLORATION rows whose dynamic range the corpus says should be nearly three times wider. Normalising each axis within its own class band would repair it and is a different build with its own controls; asserting it here without them would be the fitted weight this file refuses. Read a mixed-class slate as a mixed-class slate.",
"terms": {
"turn_pacing": {
"clause": "structural change EARNED per unit of texture chopping",
"cite": "MUSIC_PRACTICE_GAP_AUDIT.md S3; floor_instruments.py:28-56"
},
"motivic_dev": {
"clause": "developing variation -- how many operations, and how many kinds",
"cite": "MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dynamic_breath": {
"clause": "the cue breathes, and its silences are composed rather than dead",
"cite": "MUSIC_PRACTICE_GAP_AUDIT.md S9, S13; PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"voice_dialogue": {
"clause": "a second tune-carrying voice, and the melody traded between voices",
"cite": "MUSIC_PRACTICE_GAP_AUDIT.md S6"
}
},
"weights": "EQUAL AND DECLARED. Four terms, each the mean of its axes. Nothing was fitted.",
"refused_axes": {
"boundaries_per_min": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"loudness_range_lu": "REFUSED as deciding -- measured INVERTED on the graded set",
"syncopation_audio": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class"
},
"self_declared_degenerate_instruments": {
"layer_census": "floor_instruments.py:248 -- its estimator saturates near 8 streams where a real cue runs 24 to 40 parts, so it cannot adjudicate the layer clause",
"noise_structure": "floor_instruments.py:246-247 -- fires on 0 of 18 across both tiers with nothing clearing the multiplicity bar",
"summit_tests": "floor_instruments.py:124-128 -- its four continuous axes SEPARATE NOTHING on audio; it is a SCORE instrument that degrades honestly",
"hook_presence": "instr_hook_presence.py:200-210 -- the module declares its own floor DEGENERATE and says passes_floor should not be believed"
},
"bands_available": true,
"order": [
"BENCH_R1_ELEVENLABS_EDIT_V2_2",
"BENCH_R1_ELEVENLABS_EDIT_SERENE_V2",
"BENCH_R1_ELEVENLABS_TP_E1_v4_1",
"BENCH_R1_ELEVENLABS_CUE_MC0037_WaterfallTension_B",
"BENCH_R1_ELEVENLABS_TP_E1_v6_1",
"BENCH_R1_ELEVENLABS_CUE_MC0037v2_Waterfall_A",
"BENCH_R1_ELEVENLABS_CUE_MC0038_BossP1_A",
"BENCH_R1_ELEVENLABS_CUE_MC0036_InerieTension_A",
"BENCH_R1_ELEVENLABS_EDIT_V2_1",
"BENCH_R1_ELEVENLABS_TP_E1_v6_2",
"BENCH_R1_ELEVENLABS_CUE_MC0037v2_Waterfall_B",
"BENCH_R1_ELEVENLABS_CUE_MC0034_KelimutuTension_A",
"BENCH_R1_ELEVENLABS_CUE_MC0035v2_LiangBua_A",
"BENCH_R1_ELEVENLABS_CUE_MC0035_LiangBuaTension_B",
"BENCH_R1_ELEVENLABS_TP_B2_HYBRID_1",
"BENCH_R1_ELEVENLABS_CUE_MC0050_Motif_A",
"BENCH_R1_ELEVENLABS_CUE_MC0050_Motif_B",
"BENCH_R1_ELEVENLABS_EDIT_BATTLE_V2",
"BENCH_R1_ELEVENLABS_TP_S1_v1_2",
"BENCH_R1_ELEVENLABS_CUE_MC0034v2_Kelimutu_A",
"BENCH_R1_ELEVENLABS_EDIT_V2_3",
"BENCH_R1_ELEVENLABS_CUE_MC0039_BossP2_B",
"BENCH_R1_ELEVENLABS_EDIT_KOMODO_FULLPOWER",
"BENCH_R1_ELEVENLABS_CUE_MC0035_LiangBuaTension_A",
"BENCH_R1_ELEVENLABS_CUE_MC0036_InerieTension_B",
"BENCH_R1_ELEVENLABS_GEN_KOMODO_V3_B",
"BENCH_R1_ELEVENLABS_CUE_MC0034v2_Kelimutu_B",
"BENCH_R1_ELEVENLABS_EDIT_T2_MELODY_1",
"BENCH_R1_ELEVENLABS_CUE_MC0034_KelimutuTension_B",
"BENCH_R1_ELEVENLABS_CUE_MC0037_WaterfallTension_A",
"BENCH_R1_ELEVENLABS_TP_E1_v5_1",
"BENCH_R1_ELEVENLABS_TP_E1_v5_2",
"BENCH_R1_ELEVENLABS_CUE_MC0038_BossP1_B",
"BENCH_R1_ELEVENLABS_TP_S1_v1_1",
"BENCH_R1_ELEVENLABS_CUE_MC0036v2_Inerie_B",
"BENCH_R1_ELEVENLABS_EDIT_V2_4",
"BENCH_R1_INHOUSE_BASELINE_R4",
"BENCH_R1_ELEVENLABS_CUE_MC0036v2_Inerie_A",
"BENCH_R1_ELEVENLABS_TP_B1_v1_2",
"BENCH_R1_ELEVENLABS_GEN_KOMODO_V3_A",
"BENCH_R1_ELEVENLABS_TP_E1_v4_2",
"BENCH_R1_ELEVENLABS_TP_B1_v1_1",
"BENCH_R1_ELEVENLABS_EDIT_FLORES_P3",
"BENCH_R1_ELEVENLABS_CUE_MC0035v2_LiangBua_B",
"BENCH_R1_ELEVENLABS_CUE_MC0039_BossP2_A"
],
"tracks": {
"BENCH_R1_ELEVENLABS_CUE_MC0034_KelimutuTension_A": {
"cue_class": "EXPLORATION",
"character_reading": 0.6055,
"leave_one_term_out": {
"without_turn_pacing": 0.574,
"without_motivic_dev": 0.5782,
"without_dynamic_breath": 0.6481,
"without_voice_dialogue": 0.6218
},
"axes": {
"structural_yield": {
"value": 1.4286,
"term": "turn_pacing",
"slate_percentile": 0.7,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 5.0,
"term": null,
"slate_percentile": 0.7875,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 13.46,
"term": "dynamic_breath",
"slate_percentile": 0.4778,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 10.88,
"term": null,
"slate_percentile": 0.7444,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 4.5,
"term": "motivic_dev",
"slate_percentile": 0.75,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 2.0,
"term": "motivic_dev",
"slate_percentile": 0.625,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.7642,
"term": "voice_dialogue",
"slate_percentile": 0.9886,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.0919,
"term": "voice_dialogue",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0667,
"term": null,
"slate_percentile": 0.6667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0034_KelimutuTension_B": {
"cue_class": "EXPLORATION",
"character_reading": 0.4437,
"leave_one_term_out": {
"without_turn_pacing": 0.4437,
"without_motivic_dev": 0.3701,
"without_dynamic_breath": 0.3267,
"without_voice_dialogue": 0.6343
},
"axes": {
"structural_yield": {
"value": null,
"term": "turn_pacing",
"slate_percentile": null,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": null,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": null,
"term": null,
"slate_percentile": null,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": null,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 16.78,
"term": "dynamic_breath",
"slate_percentile": 0.6778,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 15.47,
"term": null,
"slate_percentile": 0.9,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 5.0,
"term": "motivic_dev",
"slate_percentile": 0.7841,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.1023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.0227,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 0.0,
"to_s": 12.6,
"duration_s": 12.6,
"mean_energy": 0.1134
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 12.6,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0034v2_Kelimutu_A": {
"cue_class": "EXPLORATION",
"character_reading": 0.5313,
"leave_one_term_out": {
"without_turn_pacing": 0.5793,
"without_motivic_dev": 0.5209,
"without_dynamic_breath": 0.4455,
"without_voice_dialogue": 0.5796
},
"axes": {
"structural_yield": {
"value": 0.5238,
"term": "turn_pacing",
"slate_percentile": 0.3875,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 5.5,
"term": null,
"slate_percentile": 0.8625,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 20.65,
"term": "dynamic_breath",
"slate_percentile": 0.7889,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 11.95,
"term": null,
"slate_percentile": 0.7889,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.5,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 2.0,
"term": "motivic_dev",
"slate_percentile": 0.625,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.1023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.4086,
"term": "voice_dialogue",
"slate_percentile": 0.6705,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 0.0,
"to_s": 23.8,
"duration_s": 23.8,
"mean_energy": 0.1567
},
{
"from_s": 104.8,
"to_s": 119.6,
"duration_s": 14.9,
"mean_energy": 0.1245
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 38.7,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0034v2_Kelimutu_B": {
"cue_class": "EXPLORATION",
"character_reading": 0.4576,
"leave_one_term_out": {
"without_turn_pacing": 0.3893,
"without_motivic_dev": 0.5685,
"without_dynamic_breath": 0.4064,
"without_voice_dialogue": 0.4662
},
"axes": {
"structural_yield": {
"value": 1.4,
"term": "turn_pacing",
"slate_percentile": 0.6625,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.5,
"term": null,
"slate_percentile": 0.55,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 15.39,
"term": "dynamic_breath",
"slate_percentile": 0.6111,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 12.04,
"term": null,
"slate_percentile": 0.8111,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.6177,
"term": "voice_dialogue",
"slate_percentile": 0.8068,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.0612,
"term": "voice_dialogue",
"slate_percentile": 0.0568,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0035_LiangBuaTension_A": {
"cue_class": "EXPLORATION",
"character_reading": 0.482,
"leave_one_term_out": {
"without_turn_pacing": 0.5968,
"without_motivic_dev": 0.3547,
"without_dynamic_breath": 0.4019,
"without_voice_dialogue": 0.5745
},
"axes": {
"structural_yield": {
"value": 0.4375,
"term": "turn_pacing",
"slate_percentile": 0.1375,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.5,
"term": null,
"slate_percentile": 0.55,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 18.62,
"term": "dynamic_breath",
"slate_percentile": 0.7222,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 14.77,
"term": null,
"slate_percentile": 0.8778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 7.5,
"term": "motivic_dev",
"slate_percentile": 0.8523,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 4.0,
"term": "motivic_dev",
"slate_percentile": 0.875,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.1503,
"term": "voice_dialogue",
"slate_percentile": 0.2159,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.2131,
"term": "voice_dialogue",
"slate_percentile": 0.1932,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0035_LiangBuaTension_B": {
"cue_class": "EXPLORATION",
"character_reading": 0.5762,
"leave_one_term_out": {
"without_turn_pacing": 0.5762,
"without_motivic_dev": 0.5347,
"without_dynamic_breath": 0.3921,
"without_voice_dialogue": 0.8017
},
"axes": {
"structural_yield": {
"value": null,
"term": "turn_pacing",
"slate_percentile": null,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": null,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": null,
"term": null,
"slate_percentile": null,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": null,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 32.73,
"term": "dynamic_breath",
"slate_percentile": 0.9444,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 16.43,
"term": null,
"slate_percentile": 0.9667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 1.5,
"term": "motivic_dev",
"slate_percentile": 0.5455,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 3.0,
"term": "motivic_dev",
"slate_percentile": 0.7727,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.1023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.1533,
"term": "voice_dialogue",
"slate_percentile": 0.1477,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.2667,
"term": null,
"slate_percentile": 0.9333,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 0.0,
"to_s": 20.1,
"duration_s": 20.1,
"mean_energy": 0.0966
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 20.1,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0035v2_LiangBua_A": {
"cue_class": "EXPLORATION",
"character_reading": 0.5991,
"leave_one_term_out": {
"without_turn_pacing": 0.6405,
"without_motivic_dev": 0.5356,
"without_dynamic_breath": 0.5655,
"without_voice_dialogue": 0.6549
},
"axes": {
"structural_yield": {
"value": 0.6667,
"term": "turn_pacing",
"slate_percentile": 0.475,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 5.0,
"term": null,
"slate_percentile": 0.7875,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 17.66,
"term": "dynamic_breath",
"slate_percentile": 0.7,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 12.1,
"term": null,
"slate_percentile": 0.8333,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 6.0,
"term": "motivic_dev",
"slate_percentile": 0.8068,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 3.0,
"term": "motivic_dev",
"slate_percentile": 0.7727,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.2698,
"term": "voice_dialogue",
"slate_percentile": 0.3295,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.3499,
"term": "voice_dialogue",
"slate_percentile": 0.5341,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0333,
"term": null,
"slate_percentile": 0.6111,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0035v2_LiangBua_B": {
"cue_class": "EXPLORATION",
"character_reading": 0.2036,
"leave_one_term_out": {
"without_turn_pacing": 0.2036,
"without_motivic_dev": 0.2429,
"without_dynamic_breath": 0.2443,
"without_voice_dialogue": 0.1236
},
"axes": {
"structural_yield": {
"value": null,
"term": "turn_pacing",
"slate_percentile": null,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": null,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": null,
"term": null,
"slate_percentile": null,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": null,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 6.6,
"term": "dynamic_breath",
"slate_percentile": 0.1222,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 5.84,
"term": null,
"slate_percentile": 0.4333,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.4844,
"term": "voice_dialogue",
"slate_percentile": 0.4659,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.2158,
"term": "voice_dialogue",
"slate_percentile": 0.2614,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.1333,
"term": null,
"slate_percentile": 0.7778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0036_InerieTension_A": {
"cue_class": "EXPLORATION",
"character_reading": 0.6321,
"leave_one_term_out": {
"without_turn_pacing": 0.6636,
"without_motivic_dev": 0.5587,
"without_dynamic_breath": 0.5428,
"without_voice_dialogue": 0.7632
},
"axes": {
"structural_yield": {
"value": 0.75,
"term": "turn_pacing",
"slate_percentile": 0.5375,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.0,
"term": null,
"slate_percentile": 0.4,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 25.21,
"term": "dynamic_breath",
"slate_percentile": 0.9,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 18.19,
"term": null,
"slate_percentile": 0.9889,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 6.5,
"term": "motivic_dev",
"slate_percentile": 0.8295,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 4.0,
"term": "motivic_dev",
"slate_percentile": 0.875,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.1023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.2775,
"term": "voice_dialogue",
"slate_percentile": 0.375,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 104.0,
"to_s": 119.6,
"duration_s": 15.6,
"mean_energy": 0.0854
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 15.6,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0036_InerieTension_B": {
"cue_class": "EXPLORATION",
"character_reading": 0.4725,
"leave_one_term_out": {
"without_turn_pacing": 0.6092,
"without_motivic_dev": 0.3194,
"without_dynamic_breath": 0.4337,
"without_voice_dialogue": 0.5277
},
"axes": {
"structural_yield": {
"value": 0.25,
"term": "turn_pacing",
"slate_percentile": 0.0625,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 2.0,
"term": null,
"slate_percentile": 0.225,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 14.97,
"term": "dynamic_breath",
"slate_percentile": 0.5889,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 4.48,
"term": null,
"slate_percentile": 0.2333,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 10.5,
"term": "motivic_dev",
"slate_percentile": 0.9091,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 5.0,
"term": "motivic_dev",
"slate_percentile": 0.9545,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.1023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.3476,
"term": "voice_dialogue",
"slate_percentile": 0.5114,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.2,
"term": null,
"slate_percentile": 0.8667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0036v2_Inerie_A": {
"cue_class": "EXPLORATION",
"character_reading": 0.3625,
"leave_one_term_out": {
"without_turn_pacing": 0.4042,
"without_motivic_dev": 0.4416,
"without_dynamic_breath": 0.2648,
"without_voice_dialogue": 0.3394
},
"axes": {
"structural_yield": {
"value": 0.4615,
"term": "turn_pacing",
"slate_percentile": 0.2375,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.0,
"term": null,
"slate_percentile": 0.4,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 16.26,
"term": "dynamic_breath",
"slate_percentile": 0.6556,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 9.6,
"term": null,
"slate_percentile": 0.6778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.1023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.4436,
"term": "voice_dialogue",
"slate_percentile": 0.7614,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0036v2_Inerie_B": {
"cue_class": "EXPLORATION",
"character_reading": 0.3856,
"leave_one_term_out": {
"without_turn_pacing": 0.3684,
"without_motivic_dev": 0.4725,
"without_dynamic_breath": 0.2216,
"without_voice_dialogue": 0.4801
},
"axes": {
"structural_yield": {
"value": 0.625,
"term": "turn_pacing",
"slate_percentile": 0.4375,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 5.0,
"term": null,
"slate_percentile": 0.7875,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 24.5,
"term": "dynamic_breath",
"slate_percentile": 0.8778,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 6.73,
"term": null,
"slate_percentile": 0.5,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.1023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.0757,
"term": "voice_dialogue",
"slate_percentile": 0.1023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0037_WaterfallTension_A": {
"cue_class": "EXPLORATION",
"character_reading": 0.4417,
"leave_one_term_out": {
"without_turn_pacing": 0.4306,
"without_motivic_dev": 0.4393,
"without_dynamic_breath": 0.4519,
"without_voice_dialogue": 0.445
},
"axes": {
"structural_yield": {
"value": 0.6667,
"term": "turn_pacing",
"slate_percentile": 0.475,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.0,
"term": null,
"slate_percentile": 0.4,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 10.07,
"term": "dynamic_breath",
"slate_percentile": 0.4111,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 3.97,
"term": null,
"slate_percentile": 0.1222,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.5,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.5709,
"term": "voice_dialogue",
"slate_percentile": 0.6932,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.1592,
"term": "voice_dialogue",
"slate_percentile": 0.1705,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0037_WaterfallTension_B": {
"cue_class": "EXPLORATION",
"character_reading": 0.6829,
"leave_one_term_out": {
"without_turn_pacing": 0.6063,
"without_motivic_dev": 0.8688,
"without_dynamic_breath": 0.5882,
"without_voice_dialogue": 0.6681
},
"axes": {
"structural_yield": {
"value": 3.5,
"term": "turn_pacing",
"slate_percentile": 0.9125,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.5,
"term": null,
"slate_percentile": 0.55,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 39.63,
"term": "dynamic_breath",
"slate_percentile": 0.9667,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 2.8,
"term": null,
"slate_percentile": 0.0111,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.5722,
"term": "voice_dialogue",
"slate_percentile": 0.7386,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.4378,
"term": "voice_dialogue",
"slate_percentile": 0.7159,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 84.7,
"to_s": 119.6,
"duration_s": 34.9,
"mean_energy": 0.0955
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 34.9,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0037v2_Waterfall_A": {
"cue_class": "EXPLORATION",
"character_reading": 0.6406,
"leave_one_term_out": {
"without_turn_pacing": 0.6,
"without_motivic_dev": 0.6591,
"without_dynamic_breath": 0.6727,
"without_voice_dialogue": 0.6307
},
"axes": {
"structural_yield": {
"value": 1.75,
"term": "turn_pacing",
"slate_percentile": 0.7625,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.5,
"term": null,
"slate_percentile": 0.55,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 14.59,
"term": "dynamic_breath",
"slate_percentile": 0.5444,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 8.2,
"term": null,
"slate_percentile": 0.6111,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 1.5,
"term": "motivic_dev",
"slate_percentile": 0.5455,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 2.0,
"term": "motivic_dev",
"slate_percentile": 0.625,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.6107,
"term": "voice_dialogue",
"slate_percentile": 0.7841,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.3523,
"term": "voice_dialogue",
"slate_percentile": 0.5568,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0037v2_Waterfall_B": {
"cue_class": "EXPLORATION",
"character_reading": 0.6069,
"leave_one_term_out": {
"without_turn_pacing": 0.5966,
"without_motivic_dev": 0.7675,
"without_dynamic_breath": 0.561,
"without_voice_dialogue": 0.5023
},
"axes": {
"structural_yield": {
"value": 1.2857,
"term": "turn_pacing",
"slate_percentile": 0.6375,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 4.5,
"term": null,
"slate_percentile": 0.6625,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 18.65,
"term": "dynamic_breath",
"slate_percentile": 0.7444,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 15.53,
"term": null,
"slate_percentile": 0.9222,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.6918,
"term": "voice_dialogue",
"slate_percentile": 0.9205,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.6094,
"term": "voice_dialogue",
"slate_percentile": 0.9205,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 104.0,
"to_s": 119.6,
"duration_s": 15.6,
"mean_energy": 0.1549
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 15.6,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0038_BossP1_A": {
"cue_class": "BOSS",
"character_reading": 0.6368,
"leave_one_term_out": {
"without_turn_pacing": 0.52,
"without_motivic_dev": 0.5385,
"without_dynamic_breath": 0.7269,
"without_voice_dialogue": 0.762
},
"axes": {
"structural_yield": {
"value": 14.0,
"term": "turn_pacing",
"slate_percentile": 0.9875,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 7.0,
"term": null,
"slate_percentile": 0.9375,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 9.7,
"term": "dynamic_breath",
"slate_percentile": 0.3667,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 3.59,
"term": null,
"slate_percentile": 0.0778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 10.5,
"term": "motivic_dev",
"slate_percentile": 0.9091,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 5.0,
"term": "motivic_dev",
"slate_percentile": 0.9545,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.1716,
"term": "voice_dialogue",
"slate_percentile": 0.2386,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.2283,
"term": "voice_dialogue",
"slate_percentile": 0.2841,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.1333,
"term": null,
"slate_percentile": 0.7778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.3501,
"floor_class": "BOSS",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0038_BossP1_B": {
"cue_class": "BOSS",
"character_reading": 0.4117,
"leave_one_term_out": {
"without_turn_pacing": 0.4365,
"without_motivic_dev": 0.3766,
"without_dynamic_breath": 0.3379,
"without_voice_dialogue": 0.4959
},
"axes": {
"structural_yield": {
"value": 0.5,
"term": "turn_pacing",
"slate_percentile": 0.3375,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.0,
"term": null,
"slate_percentile": 0.4,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 15.75,
"term": "dynamic_breath",
"slate_percentile": 0.6333,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 8.58,
"term": null,
"slate_percentile": 0.6333,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 2.0,
"term": "motivic_dev",
"slate_percentile": 0.6364,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.1023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.2133,
"term": "voice_dialogue",
"slate_percentile": 0.2159,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.2667,
"term": null,
"slate_percentile": 0.9333,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.3501,
"floor_class": "BOSS",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0039_BossP2_A": {
"cue_class": "BOSS",
"character_reading": 0.0556,
"leave_one_term_out": {
"without_turn_pacing": 0.0556,
"without_motivic_dev": 0.0556,
"without_dynamic_breath": null,
"without_voice_dialogue": 0.0556
},
"axes": {
"structural_yield": {
"value": null,
"term": "turn_pacing",
"slate_percentile": null,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": null,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": null,
"term": null,
"slate_percentile": null,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": null,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 5.83,
"term": "dynamic_breath",
"slate_percentile": 0.0556,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 4.18,
"term": null,
"slate_percentile": 0.1667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": null,
"term": "motivic_dev",
"slate_percentile": null,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": null,
"term": "motivic_dev",
"slate_percentile": null,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": null,
"term": "voice_dialogue",
"slate_percentile": null,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": null,
"term": "voice_dialogue",
"slate_percentile": null,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0667,
"term": null,
"slate_percentile": 0.6667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.3501,
"floor_class": "BOSS",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0039_BossP2_B": {
"cue_class": "BOSS",
"character_reading": 0.4997,
"leave_one_term_out": {
"without_turn_pacing": 0.3787,
"without_motivic_dev": 0.5412,
"without_dynamic_breath": 0.6625,
"without_voice_dialogue": 0.4162
},
"axes": {
"structural_yield": {
"value": 3.3333,
"term": "turn_pacing",
"slate_percentile": 0.8625,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 5.0,
"term": null,
"slate_percentile": 0.7875,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 4.69,
"term": "dynamic_breath",
"slate_percentile": 0.0111,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 3.75,
"term": null,
"slate_percentile": 0.1,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.5,
"term": "motivic_dev",
"slate_percentile": 0.3523,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.4964,
"term": "voice_dialogue",
"slate_percentile": 0.5341,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.7113,
"term": "voice_dialogue",
"slate_percentile": 0.9659,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.3501,
"floor_class": "BOSS",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 119.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0050_Motif_A": {
"cue_class": "SERENE",
"character_reading": 0.5748,
"leave_one_term_out": {
"without_turn_pacing": 0.7373,
"without_motivic_dev": 0.4483,
"without_dynamic_breath": 0.4591,
"without_voice_dialogue": 0.6547
},
"axes": {
"structural_yield": {
"value": 0.36,
"term": "turn_pacing",
"slate_percentile": 0.0875,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 7.2,
"term": null,
"slate_percentile": 0.9625,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 32.3,
"term": "dynamic_breath",
"slate_percentile": 0.9222,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 13.55,
"term": null,
"slate_percentile": 0.8556,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 12.0,
"term": "motivic_dev",
"slate_percentile": 0.9545,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 5.0,
"term": "motivic_dev",
"slate_percentile": 0.9545,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.5583,
"term": "voice_dialogue",
"slate_percentile": 0.6477,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.0227,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.1333,
"term": null,
"slate_percentile": 0.7778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2222,
"floor_class": "MELANCHOLIC",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": true,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 74.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_CUE_MC0050_Motif_B": {
"cue_class": "SERENE",
"character_reading": 0.5589,
"leave_one_term_out": {
"without_turn_pacing": 0.6327,
"without_motivic_dev": 0.5823,
"without_dynamic_breath": 0.4155,
"without_voice_dialogue": 0.605
},
"axes": {
"structural_yield": {
"value": 0.5,
"term": "turn_pacing",
"slate_percentile": 0.3375,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 6.4,
"term": null,
"slate_percentile": 0.9125,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 39.99,
"term": "dynamic_breath",
"slate_percentile": 0.9889,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 7.8,
"term": null,
"slate_percentile": 0.5667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 1.6,
"term": "motivic_dev",
"slate_percentile": 0.5795,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.387,
"term": "voice_dialogue",
"slate_percentile": 0.375,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.3406,
"term": "voice_dialogue",
"slate_percentile": 0.4659,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.1333,
"term": null,
"slate_percentile": 0.7778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2222,
"floor_class": "MELANCHOLIC",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": true,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 74.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_EDIT_BATTLE_V2": {
"cue_class": "BATTLE",
"character_reading": 0.5519,
"leave_one_term_out": {
"without_turn_pacing": 0.5025,
"without_motivic_dev": 0.4518,
"without_dynamic_breath": 0.6803,
"without_voice_dialogue": 0.573
},
"axes": {
"structural_yield": {
"value": 1.4286,
"term": "turn_pacing",
"slate_percentile": 0.7,
"corpus_band": {
"n": 5,
"p10": 0.71,
"p50": 1.875,
"p90": 2.6133,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.0,
"term": null,
"slate_percentile": 0.4,
"corpus_band": {
"n": 5,
"p10": 1.7474,
"p50": 3.329,
"p90": 4.5384,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 7.2,
"term": "dynamic_breath",
"slate_percentile": 0.1667,
"corpus_band": {
"n": 5,
"p10": 5.452,
"p50": 8.19,
"p90": 11.256,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 5.65,
"term": null,
"slate_percentile": 0.4111,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 4.5,
"term": "motivic_dev",
"slate_percentile": 0.75,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 5.0,
"term": "motivic_dev",
"slate_percentile": 0.9545,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.2011,
"term": "voice_dialogue",
"slate_percentile": 0.2841,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.412,
"term": "voice_dialogue",
"slate_percentile": 0.6932,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.5018,
"floor_class": "BATTLE",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 199.9,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_EDIT_FLORES_P3": {
"cue_class": "BOSS",
"character_reading": 0.2656,
"leave_one_term_out": {
"without_turn_pacing": 0.3166,
"without_motivic_dev": 0.1798,
"without_dynamic_breath": 0.2837,
"without_voice_dialogue": 0.2821
},
"axes": {
"structural_yield": {
"value": 0.4,
"term": "turn_pacing",
"slate_percentile": 0.1125,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 1.143,
"term": null,
"slate_percentile": 0.0375,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 7.48,
"term": "dynamic_breath",
"slate_percentile": 0.2111,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 5.34,
"term": null,
"slate_percentile": 0.3667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.5714,
"term": "motivic_dev",
"slate_percentile": 0.4205,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 2.0,
"term": "motivic_dev",
"slate_percentile": 0.625,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.0,
"term": "voice_dialogue",
"slate_percentile": 0.1023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.2344,
"term": "voice_dialogue",
"slate_percentile": 0.3295,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.3501,
"floor_class": "BOSS",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 209.5,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_EDIT_KOMODO_FULLPOWER": {
"cue_class": "BOSS",
"character_reading": 0.4947,
"leave_one_term_out": {
"without_turn_pacing": 0.3804,
"without_motivic_dev": 0.4153,
"without_dynamic_breath": 0.6485,
"without_voice_dialogue": 0.5346
},
"axes": {
"structural_yield": {
"value": 2.375,
"term": "turn_pacing",
"slate_percentile": 0.8375,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 4.471,
"term": null,
"slate_percentile": 0.6375,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 4.82,
"term": "dynamic_breath",
"slate_percentile": 0.0333,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 3.34,
"term": null,
"slate_percentile": 0.0556,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 3.0588,
"term": "motivic_dev",
"slate_percentile": 0.6932,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 3.0,
"term": "motivic_dev",
"slate_percentile": 0.7727,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.4037,
"term": "voice_dialogue",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.2518,
"term": "voice_dialogue",
"slate_percentile": 0.3523,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.3333,
"term": null,
"slate_percentile": 0.9889,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.3501,
"floor_class": "BOSS",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 254.9,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_EDIT_SERENE_V2": {
"cue_class": "SERENE",
"character_reading": 0.7289,
"leave_one_term_out": {
"without_turn_pacing": 0.676,
"without_motivic_dev": 0.7597,
"without_dynamic_breath": 0.7163,
"without_voice_dialogue": 0.7635
},
"axes": {
"structural_yield": {
"value": 3.4444,
"term": "turn_pacing",
"slate_percentile": 0.8875,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 7.75,
"term": null,
"slate_percentile": 0.9875,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 18.69,
"term": "dynamic_breath",
"slate_percentile": 0.7667,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 5.03,
"term": null,
"slate_percentile": 0.3222,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 8.5,
"term": "motivic_dev",
"slate_percentile": 0.875,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.4631,
"term": "voice_dialogue",
"slate_percentile": 0.4432,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.4803,
"term": "voice_dialogue",
"slate_percentile": 0.8068,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 0.0,
"to_s": 13.4,
"duration_s": 13.4,
"mean_energy": 0.1393
}
],
"energy_floor": 0.2222,
"floor_class": "MELANCHOLIC",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": true,
"interior_dead_s": 0,
"edge_dead_s": 13.4,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_EDIT_T2_MELODY_1": {
"cue_class": "EXPLORATION",
"character_reading": 0.4455,
"leave_one_term_out": {
"without_turn_pacing": 0.4231,
"without_motivic_dev": 0.4614,
"without_dynamic_breath": 0.5458,
"without_voice_dialogue": 0.3515
},
"axes": {
"structural_yield": {
"value": 0.7333,
"term": "turn_pacing",
"slate_percentile": 0.5125,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 2.87,
"term": null,
"slate_percentile": 0.3,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 6.65,
"term": "dynamic_breath",
"slate_percentile": 0.1444,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 2.99,
"term": null,
"slate_percentile": 0.0333,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.5217,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.4907,
"term": "voice_dialogue",
"slate_percentile": 0.5114,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.6349,
"term": "voice_dialogue",
"slate_percentile": 0.9432,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 229.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_EDIT_V2_1": {
"cue_class": "EXPLORATION",
"character_reading": 0.6291,
"leave_one_term_out": {
"without_turn_pacing": 0.5763,
"without_motivic_dev": 0.6228,
"without_dynamic_breath": 0.6943,
"without_voice_dialogue": 0.6229
},
"axes": {
"structural_yield": {
"value": 1.8,
"term": "turn_pacing",
"slate_percentile": 0.7875,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 4.909,
"term": null,
"slate_percentile": 0.7125,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 11.18,
"term": "dynamic_breath",
"slate_percentile": 0.4333,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 7.16,
"term": null,
"slate_percentile": 0.5444,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 3.0,
"term": "motivic_dev",
"slate_percentile": 0.6705,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 2.0,
"term": "motivic_dev",
"slate_percentile": 0.625,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.5716,
"term": "voice_dialogue",
"slate_percentile": 0.7159,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.3828,
"term": "voice_dialogue",
"slate_percentile": 0.5795,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0667,
"term": null,
"slate_percentile": 0.6667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 0.0,
"to_s": 20.1,
"duration_s": 20.1,
"mean_energy": 0.1456
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 20.1,
"analysed_s": 219.9,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_EDIT_V2_2": {
"cue_class": "EXPLORATION",
"character_reading": 0.7699,
"leave_one_term_out": {
"without_turn_pacing": 0.8306,
"without_motivic_dev": 0.7159,
"without_dynamic_breath": 0.7413,
"without_voice_dialogue": 0.7916
},
"axes": {
"structural_yield": {
"value": 0.8333,
"term": "turn_pacing",
"slate_percentile": 0.5875,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 1.304,
"term": null,
"slate_percentile": 0.0875,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 23.24,
"term": "dynamic_breath",
"slate_percentile": 0.8556,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 4.37,
"term": null,
"slate_percentile": 0.2111,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 12.2609,
"term": "motivic_dev",
"slate_percentile": 0.9886,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 4.0,
"term": "motivic_dev",
"slate_percentile": 0.875,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.5509,
"term": "voice_dialogue",
"slate_percentile": 0.625,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.4554,
"term": "voice_dialogue",
"slate_percentile": 0.7841,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 210.3,
"to_s": 229.6,
"duration_s": 19.3,
"mean_energy": 0.1128
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 19.3,
"analysed_s": 229.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_EDIT_V2_3": {
"cue_class": "EXPLORATION",
"character_reading": 0.5176,
"leave_one_term_out": {
"without_turn_pacing": 0.3693,
"without_motivic_dev": 0.6484,
"without_dynamic_breath": 0.6049,
"without_voice_dialogue": 0.4477
},
"axes": {
"structural_yield": {
"value": 4.0,
"term": "turn_pacing",
"slate_percentile": 0.9625,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 5.0,
"term": null,
"slate_percentile": 0.7875,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 7.98,
"term": "dynamic_breath",
"slate_percentile": 0.2556,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 4.55,
"term": null,
"slate_percentile": 0.2556,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.7318,
"term": "voice_dialogue",
"slate_percentile": 0.9659,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.3411,
"term": "voice_dialogue",
"slate_percentile": 0.4886,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_EDIT_V2_4": {
"cue_class": "EXPLORATION",
"character_reading": 0.3793,
"leave_one_term_out": {
"without_turn_pacing": 0.3683,
"without_motivic_dev": 0.4641,
"without_dynamic_breath": 0.3761,
"without_voice_dialogue": 0.3088
},
"axes": {
"structural_yield": {
"value": 0.5417,
"term": "turn_pacing",
"slate_percentile": 0.4125,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.25,
"term": null,
"slate_percentile": 0.4875,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 9.99,
"term": "dynamic_breath",
"slate_percentile": 0.3889,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 6.86,
"term": null,
"slate_percentile": 0.5222,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.673,
"term": "voice_dialogue",
"slate_percentile": 0.875,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.231,
"term": "voice_dialogue",
"slate_percentile": 0.3068,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.2667,
"term": null,
"slate_percentile": 0.9333,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_GEN_KOMODO_V3_A": {
"cue_class": "BOSS",
"character_reading": 0.3154,
"leave_one_term_out": {
"without_turn_pacing": 0.3622,
"without_motivic_dev": 0.2425,
"without_dynamic_breath": 0.3576,
"without_voice_dialogue": 0.2993
},
"axes": {
"structural_yield": {
"value": 0.4444,
"term": "turn_pacing",
"slate_percentile": 0.175,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 1.0,
"term": null,
"slate_percentile": 0.0125,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 7.23,
"term": "dynamic_breath",
"slate_percentile": 0.1889,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 4.7,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.75,
"term": "motivic_dev",
"slate_percentile": 0.4432,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 2.0,
"term": "motivic_dev",
"slate_percentile": 0.625,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.485,
"term": "voice_dialogue",
"slate_percentile": 0.4886,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.2157,
"term": "voice_dialogue",
"slate_percentile": 0.2386,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.3501,
"floor_class": "BOSS",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_GEN_KOMODO_V3_B": {
"cue_class": "BOSS",
"character_reading": 0.4689,
"leave_one_term_out": {
"without_turn_pacing": 0.5544,
"without_motivic_dev": 0.3904,
"without_dynamic_breath": 0.5178,
"without_voice_dialogue": 0.4131
},
"axes": {
"structural_yield": {
"value": 0.4545,
"term": "turn_pacing",
"slate_percentile": 0.2125,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 1.25,
"term": null,
"slate_percentile": 0.0625,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 9.38,
"term": "dynamic_breath",
"slate_percentile": 0.3222,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 7.85,
"term": null,
"slate_percentile": 0.5889,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 2.0,
"term": "motivic_dev",
"slate_percentile": 0.6364,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 3.0,
"term": "motivic_dev",
"slate_percentile": 0.7727,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.4573,
"term": "voice_dialogue",
"slate_percentile": 0.4205,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.5221,
"term": "voice_dialogue",
"slate_percentile": 0.8523,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.3501,
"floor_class": "BOSS",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_B1_v1_1": {
"cue_class": "BATTLE",
"character_reading": 0.2876,
"leave_one_term_out": {
"without_turn_pacing": 0.271,
"without_motivic_dev": 0.3418,
"without_dynamic_breath": 0.3057,
"without_voice_dialogue": 0.2319
},
"axes": {
"structural_yield": {
"value": 0.5,
"term": "turn_pacing",
"slate_percentile": 0.3375,
"corpus_band": {
"n": 5,
"p10": 0.71,
"p50": 1.875,
"p90": 2.6133,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 1.429,
"term": null,
"slate_percentile": 0.1125,
"corpus_band": {
"n": 5,
"p10": 1.7474,
"p50": 3.329,
"p90": 4.5384,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 7.68,
"term": "dynamic_breath",
"slate_percentile": 0.2333,
"corpus_band": {
"n": 5,
"p10": 5.452,
"p50": 8.19,
"p90": 11.256,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 5.53,
"term": null,
"slate_percentile": 0.3889,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.1864,
"term": "voice_dialogue",
"slate_percentile": 0.2614,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.3997,
"term": "voice_dialogue",
"slate_percentile": 0.6477,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.2,
"term": null,
"slate_percentile": 0.8667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.5018,
"floor_class": "BATTLE",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 209.5,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_B1_v1_2": {
"cue_class": "BATTLE",
"character_reading": 0.3252,
"leave_one_term_out": {
"without_turn_pacing": 0.3252,
"without_motivic_dev": 0.3116,
"without_dynamic_breath": 0.4488,
"without_voice_dialogue": 0.215
},
"axes": {
"structural_yield": {
"value": null,
"term": "turn_pacing",
"slate_percentile": null,
"corpus_band": {
"n": 5,
"p10": 0.71,
"p50": 1.875,
"p90": 2.6133,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": null,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": null,
"term": null,
"slate_percentile": null,
"corpus_band": {
"n": 5,
"p10": 1.7474,
"p50": 3.329,
"p90": 4.5384,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": null,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 6.23,
"term": "dynamic_breath",
"slate_percentile": 0.0778,
"corpus_band": {
"n": 5,
"p10": 5.452,
"p50": 8.19,
"p90": 11.256,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 4.67,
"term": null,
"slate_percentile": 0.2778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.3,
"term": "motivic_dev",
"slate_percentile": 0.3068,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.3812,
"term": "voice_dialogue",
"slate_percentile": 0.3523,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.4407,
"term": "voice_dialogue",
"slate_percentile": 0.7386,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.5018,
"floor_class": "BATTLE",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 199.9,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_B2_HYBRID_1": {
"cue_class": "BATTLE",
"character_reading": 0.5761,
"leave_one_term_out": {
"without_turn_pacing": 0.4973,
"without_motivic_dev": 0.5201,
"without_dynamic_breath": 0.7348,
"without_voice_dialogue": 0.5523
},
"axes": {
"structural_yield": {
"value": 1.8889,
"term": "turn_pacing",
"slate_percentile": 0.8125,
"corpus_band": {
"n": 5,
"p10": 0.71,
"p50": 1.875,
"p90": 2.6133,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 4.857,
"term": null,
"slate_percentile": 0.6875,
"corpus_band": {
"n": 5,
"p10": 1.7474,
"p50": 3.329,
"p90": 4.5384,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 6.34,
"term": "dynamic_breath",
"slate_percentile": 0.1,
"corpus_band": {
"n": 5,
"p10": 5.452,
"p50": 8.19,
"p90": 11.256,
"provisional": true,
"resolved_from": "BATTLE",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 4.28,
"term": null,
"slate_percentile": 0.1889,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 4.2857,
"term": "motivic_dev",
"slate_percentile": 0.7159,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 3.0,
"term": "motivic_dev",
"slate_percentile": 0.7727,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.2365,
"term": "voice_dialogue",
"slate_percentile": 0.3068,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.7346,
"term": "voice_dialogue",
"slate_percentile": 0.9886,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.5018,
"floor_class": "BATTLE",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 209.5,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_E1_v4_1": {
"cue_class": "EXPLORATION",
"character_reading": 0.6911,
"leave_one_term_out": {
"without_turn_pacing": 0.6757,
"without_motivic_dev": 0.717,
"without_dynamic_breath": 0.6512,
"without_voice_dialogue": 0.7208
},
"axes": {
"structural_yield": {
"value": 1.5,
"term": "turn_pacing",
"slate_percentile": 0.7375,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 6.0,
"term": null,
"slate_percentile": 0.8875,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 20.94,
"term": "dynamic_breath",
"slate_percentile": 0.8111,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 11.16,
"term": null,
"slate_percentile": 0.7667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 1.75,
"term": "motivic_dev",
"slate_percentile": 0.6023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 2.0,
"term": "motivic_dev",
"slate_percentile": 0.625,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.5786,
"term": "voice_dialogue",
"slate_percentile": 0.7614,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.333,
"term": "voice_dialogue",
"slate_percentile": 0.4432,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0667,
"term": null,
"slate_percentile": 0.6667,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 0.0,
"to_s": 20.1,
"duration_s": 20.1,
"mean_energy": 0.1377
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 20.1,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_E1_v4_2": {
"cue_class": "EXPLORATION",
"character_reading": 0.3063,
"leave_one_term_out": {
"without_turn_pacing": 0.35,
"without_motivic_dev": 0.3667,
"without_dynamic_breath": 0.3083,
"without_voice_dialogue": 0.2
},
"axes": {
"structural_yield": {
"value": 0.4444,
"term": "turn_pacing",
"slate_percentile": 0.175,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 3.0,
"term": null,
"slate_percentile": 0.4,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 9.15,
"term": "dynamic_breath",
"slate_percentile": 0.3,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 6.69,
"term": null,
"slate_percentile": 0.4778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.6523,
"term": "voice_dialogue",
"slate_percentile": 0.8295,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.3146,
"term": "voice_dialogue",
"slate_percentile": 0.4205,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.2667,
"term": null,
"slate_percentile": 0.9333,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_E1_v5_1": {
"cue_class": "EXPLORATION",
"character_reading": 0.438,
"leave_one_term_out": {
"without_turn_pacing": 0.4923,
"without_motivic_dev": 0.459,
"without_dynamic_breath": 0.4136,
"without_voice_dialogue": 0.387
},
"axes": {
"structural_yield": {
"value": 0.4667,
"term": "turn_pacing",
"slate_percentile": 0.275,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 1.75,
"term": null,
"slate_percentile": 0.15,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 14.37,
"term": "dynamic_breath",
"slate_percentile": 0.5111,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 9.98,
"term": null,
"slate_percentile": 0.7111,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.5,
"term": "motivic_dev",
"slate_percentile": 0.3523,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.5131,
"term": "voice_dialogue",
"slate_percentile": 0.5682,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.3835,
"term": "voice_dialogue",
"slate_percentile": 0.6136,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 219.2,
"to_s": 239.3,
"duration_s": 20.1,
"mean_energy": 0.1403
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 20.1,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_E1_v5_2": {
"cue_class": "EXPLORATION",
"character_reading": 0.438,
"leave_one_term_out": {
"without_turn_pacing": 0.4923,
"without_motivic_dev": 0.459,
"without_dynamic_breath": 0.4136,
"without_voice_dialogue": 0.387
},
"axes": {
"structural_yield": {
"value": 0.4667,
"term": "turn_pacing",
"slate_percentile": 0.275,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 1.75,
"term": null,
"slate_percentile": 0.15,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 14.37,
"term": "dynamic_breath",
"slate_percentile": 0.5111,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 9.98,
"term": null,
"slate_percentile": 0.7111,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.5,
"term": "motivic_dev",
"slate_percentile": 0.3523,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.5131,
"term": "voice_dialogue",
"slate_percentile": 0.5682,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.3835,
"term": "voice_dialogue",
"slate_percentile": 0.6136,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [
{
"from_s": 219.2,
"to_s": 239.3,
"duration_s": 20.1,
"mean_energy": 0.1403
}
],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 20.1,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_E1_v6_1": {
"cue_class": "EXPLORATION",
"character_reading": 0.6447,
"leave_one_term_out": {
"without_turn_pacing": 0.5471,
"without_motivic_dev": 0.5718,
"without_dynamic_breath": 0.767,
"without_voice_dialogue": 0.693
},
"axes": {
"structural_yield": {
"value": 3.6667,
"term": "turn_pacing",
"slate_percentile": 0.9375,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 2.87,
"term": null,
"slate_percentile": 0.3,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 8.59,
"term": "dynamic_breath",
"slate_percentile": 0.2778,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 5.04,
"term": null,
"slate_percentile": 0.3444,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 12.0,
"term": "motivic_dev",
"slate_percentile": 0.9545,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 3.0,
"term": "motivic_dev",
"slate_percentile": 0.7727,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.5489,
"term": "voice_dialogue",
"slate_percentile": 0.6023,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.2862,
"term": "voice_dialogue",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 229.6,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_E1_v6_2": {
"cue_class": "EXPLORATION",
"character_reading": 0.6143,
"leave_one_term_out": {
"without_turn_pacing": 0.6149,
"without_motivic_dev": 0.7774,
"without_dynamic_breath": 0.5413,
"without_voice_dialogue": 0.5236
},
"axes": {
"structural_yield": {
"value": 1.0,
"term": "turn_pacing",
"slate_percentile": 0.6125,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 2.5,
"term": null,
"slate_percentile": 0.2625,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 21.48,
"term": "dynamic_breath",
"slate_percentile": 0.8333,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 15.66,
"term": null,
"slate_percentile": 0.9444,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 0.0,
"term": "motivic_dev",
"slate_percentile": 0.125,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.684,
"term": "voice_dialogue",
"slate_percentile": 0.8977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.5523,
"term": "voice_dialogue",
"slate_percentile": 0.875,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [
{
"from_s": 124.1,
"to_s": 144.9,
"duration_s": 20.8,
"mean_energy": 0.1258
}
],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 20.8,
"edge_dead_s": 0,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_S1_v1_1": {
"cue_class": "SERENE",
"character_reading": 0.3946,
"leave_one_term_out": {
"without_turn_pacing": 0.5136,
"without_motivic_dev": 0.4144,
"without_dynamic_breath": 0.3742,
"without_voice_dialogue": 0.2761
},
"axes": {
"structural_yield": {
"value": 0.2424,
"term": "turn_pacing",
"slate_percentile": 0.0375,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 2.0,
"term": null,
"slate_percentile": 0.225,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 12.82,
"term": "dynamic_breath",
"slate_percentile": 0.4556,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 6.15,
"term": null,
"slate_percentile": 0.4556,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.25,
"term": "motivic_dev",
"slate_percentile": 0.2727,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.5606,
"term": "voice_dialogue",
"slate_percentile": 0.6705,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.5135,
"term": "voice_dialogue",
"slate_percentile": 0.8295,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.1333,
"term": null,
"slate_percentile": 0.7778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2222,
"floor_class": "MELANCHOLIC",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": true,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_ELEVENLABS_TP_S1_v1_2": {
"cue_class": "SERENE",
"character_reading": 0.5406,
"leave_one_term_out": {
"without_turn_pacing": 0.5333,
"without_motivic_dev": 0.6091,
"without_dynamic_breath": 0.606,
"without_voice_dialogue": 0.414
},
"axes": {
"structural_yield": {
"value": 0.8,
"term": "turn_pacing",
"slate_percentile": 0.5625,
"corpus_band": {
"n": 28,
"p10": 0.363,
"p50": 0.7649,
"p90": 2.0222,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 4.0,
"term": null,
"slate_percentile": 0.6125,
"corpus_band": {
"n": 28,
"p10": 2.3576,
"p50": 5.0875,
"p90": 6.9362,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 9.43,
"term": "dynamic_breath",
"slate_percentile": 0.3444,
"corpus_band": {
"n": 28,
"p10": 5.737,
"p50": 11.715,
"p90": 21.015,
"provisional": false,
"resolved_from": "POOLED_ALL_CLASSES",
"borrowed": false,
"borrow_declared_but_unavailable": "MELANCHOLIC",
"pooled": true,
"honest_tier": "HOUSE-DECLARED-PROVISIONAL"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 4.1,
"term": null,
"slate_percentile": 0.1444,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.25,
"term": "motivic_dev",
"slate_percentile": 0.2727,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.7162,
"term": "voice_dialogue",
"slate_percentile": 0.9432,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.5916,
"term": "voice_dialogue",
"slate_percentile": 0.8977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.0,
"term": null,
"slate_percentile": 0.3,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2222,
"floor_class": "MELANCHOLIC",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": true,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 239.3,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
},
"BENCH_R1_INHOUSE_BASELINE_R4": {
"cue_class": "EXPLORATION",
"character_reading": 0.3692,
"leave_one_term_out": {
"without_turn_pacing": 0.4881,
"without_motivic_dev": 0.3484,
"without_dynamic_breath": 0.3034,
"without_voice_dialogue": 0.337
},
"axes": {
"structural_yield": {
"value": 0.1296,
"term": "turn_pacing",
"slate_percentile": 0.0125,
"corpus_band": {
"n": 8,
"p10": 0.4157,
"p50": 0.8559,
"p90": 2.284,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "higher is more change per cut, both tails reported",
"honest_tier": "AUDIO-COMPLETE -- a GENERATION-DEFECT DETECTOR, not a taste oracle -- floor_instruments.py:41-45 measures it at AUC 0.423 against album siblings, the regime a slate of edits of one track sits in.",
"auc_over_the_graded_set": 0.9688,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.8333 kept / 0.7333 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3 (monotony is overcome only by variation); harness/music_gen/floor_instruments.py:28-56"
},
"boundaries_per_min": {
"value": 1.88,
"term": null,
"slate_percentile": 0.1875,
"corpus_band": {
"n": 8,
"p10": 4.6279,
"p50": 5.8915,
"p90": 7.2478,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": false,
"direction": "REPORTED; the order assumed FEWER is better and the corpus says the opposite",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.875,
"t2_pair_behaviour": "INVERTS IT (1.304 kept / 2.87 rejected) -- why it is not a term",
"practice_cite": "harness/music_gen/floor_instruments.py:36-38 (hits 5.09, turned-down 2.78)"
},
"dynamic_range_db": {
"value": 14.67,
"term": "dynamic_breath",
"slate_percentile": 0.5667,
"corpus_band": {
"n": 8,
"p10": 6.99,
"p50": 10.11,
"p90": 28.979,
"provisional": false,
"resolved_from": "EXPLORATION",
"borrowed": false,
"borrow_declared_but_unavailable": null,
"pooled": false,
"honest_tier": "CORPUS-DERIVED"
},
"inside_corpus_band": true,
"direction": "higher is more breathing, both tails reported",
"honest_tier": "AUDIO-COMPLETE",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (23.24 kept / 6.65 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S9 (dynamics are composed, not a humanise knob); PIPE_AUDIO_MUSIC_2026-07-29.md §0"
},
"loudness_range_lu": {
"value": 8.61,
"term": null,
"slate_percentile": 0.6556,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REFUSED as deciding -- measured INVERTED on the graded set",
"honest_tier": "AUDIO-COMPLETE, INFORMATIONAL",
"auc_over_the_graded_set": 0.3,
"t2_pair_behaviour": "orders it correctly (4.37 / 2.99) and inverts five other rulings",
"practice_cite": "harness/music_gen/mix_policy.py:159-168; the exclusion is asserted by benchmark_judge control C8"
},
"dev_ops_per_min": {
"value": 0.8056,
"term": "motivic_dev",
"slate_percentile": 0.4659,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more development",
"honest_tier": "AUDIO-ESTIMATE -- runs on the extracted predominant line",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (47 ops kept / 2 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"dev_kinds": {
"value": 1.0,
"term": "motivic_dev",
"slate_percentile": 0.3977,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more varied development",
"honest_tier": "AUDIO-ESTIMATE",
"auc_over_the_graded_set": 0.625,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (4 kinds kept / 1 rejected)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S3"
},
"counter_melody_presence": {
"value": 0.6614,
"term": "voice_dialogue",
"slate_percentile": 0.8523,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more second-voice presence",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy, NOT source separation",
"auc_over_the_graded_set": 0.6,
"t2_pair_behaviour": "ORDERS IT CORRECTLY (0.5509 / 0.4907)",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6"
},
"orchestrational_dialogue": {
"value": 0.0681,
"term": "voice_dialogue",
"slate_percentile": 0.0795,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "higher is more trading",
"honest_tier": "AUDIO-ESTIMATE -- register-band proxy",
"auc_over_the_graded_set": 0.575,
"t2_pair_behaviour": "INVERTS IT (0.4554 / 0.6349) -- carried anyway, because the term is theory-led",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S6; benchmark_judge.UNMEASURED_BY_ANY_INSTRUMENT names carrier migration unmeasured at INSTRUMENT grain, which this proxy does not claim to close"
},
"syncopation_audio": {
"value": 0.1333,
"term": null,
"slate_percentile": 0.7778,
"corpus_band": null,
"inside_corpus_band": null,
"direction": "REPORTED -- reads exactly 0.0 on 13 of 19 engine rows and its EXPLORATION p10 is 0.0, so below-band is unreachable; a zero with no positive control on real material is the a-search-that-cannot-match class",
"honest_tier": "AUDIO-COMPLETE by construction, UNCONTROLLED in practice",
"auc_over_the_graded_set": 0.525,
"t2_pair_behaviour": "TIE (0.0 / 0.0) -- it cannot see this pair at all",
"practice_cite": "docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S7 (composite syncopation is the number the empirical literature ties to groove)"
}
},
"terms_that_stood": [
"dynamic_breath",
"motivic_dev",
"turn_pacing",
"voice_dialogue"
],
"dead_space": {
"available": true,
"reason": "",
"interior_spans": [],
"edge_spans": [],
"energy_floor": 0.2737,
"floor_class": "EXPLORATION",
"known_limitation": "KNOWN-WRONG AND INHERITED, NOT A CLEAN READING. `instr_dynamic_arc.sedation_spans` receives only the rendered curve, so it is PLAN-BLIND: a DECLARED silence window (PASS7_FLORES_FALLS carries one) and composed exploration silence are indistinguishable from a flatline once they run past the 12 s threshold, and the instrument will penalise exactly the design the exemplar lineage validates. Graded WRONG at docs/proposals/music/MUSIC_PRACTICE_GAP_AUDIT.md S13 (\"Composed silence is not sedation\"); the repair is specified and UNBUILT at docs/proposals/music/craft_research/GAME_SCORING_AND_ADAPTIVE_PRACTICE.md R8 (pass the plan, exclude declared silence, report the excluded spans separately so the exclusion cannot hide a real flatline) with the mechanism at its §12.4. Until R8 lands, a span here is a QUESTION for a composer -- 'was this silence composed?' -- and never a defect finding on its own.",
"borrowed_floor": false,
"interior_dead_s": 0,
"edge_dead_s": 0,
"analysed_s": 222.9,
"honest_tier": "AUDIO-COMPLETE (card flow curve); a FINDING, never a term",
"fires_on": "1 of 19 engine rows carries an INTERIOR span -- nothing here clears a multiplicity bar, and the low firing rate is the finding's own limit"
}
}
}
}