HOOK_AND_TENSION_ARCHITECTURE.md

music/HOOK_AND_TENSION_ARCHITECTURE.md

HOOK CONSTRUCTION AND TENSION/RELEASE ARCHITECTURE

CANON SUBORDINATION — this document is PROPOSAL-TIER: it serves canon and never outranks it.

Canon served: the composition floor at docs/spine/DECISIONS_PENDING_JOSH.md (commit da627060), and the

per-chapter music_mood anchors of docs/spine/CH_03.md, docs/spine/CH_04.md and

docs/spine/CH_10.md.

If this document disagrees with canon, CANON WINS and this document is the defect.

Tier: RESEARCH SYNTHESIS, proposal-tier. This document is subordinate to canon: the CVD, the T1

foundation docs, the spine (docs/spine/CH_*.md), the Tier-2 region pages

(_source/02_Tier_2_Region_Pages/) and the T0 registries. Where it disagrees with canon, canon

wins and this document is the defect. Nothing here is applied until it is ratified.

Every substantive claim carries a source. Claims that are craft consensus rather than evidence are

labelled CONSENSUS. Claims measured from our own formula cards for this document are labelled

MEASURED-HERE and carry no calibrated multiplicity bar — they are descriptive reads at N = 28,

in the same honest register as PATTERN_FINDINGS_V2.md section 10.

Commissioned by: the composition floor at docs/spine/DECISIONS_PENDING_JOSH.md (MUSIC ROUND 1

GRADED + THE COMPOSITION FLOOR, commit da627060) — "its got to have the hooks and complexities.

There cant just be stacked noise", "no track will be boring or filler", and the requirement to set

"the next bar for video game sound tracks".

0. DERIVATION — what canon was read before this was written

• Read-set resolved live with python harness/route.py music. It names, in order:

docs/spine/CH_NN.md (Asset anchors, the music_mood bullet), the region page Section 10

(the Music Sound and Composition Brief), registries/T0_Theme_Registry,

docs/translation/T99_Translation_Audio.md Music Steps 1-7, and

docs/proposals/MUSIC_COMPOSITION_DOCTRINE.md.

• Canon nodes actually opened: docs/spine/CH_03.md L245, docs/spine/CH_04.md L242 and

docs/spine/CH_10.md L224 (the music_mood asset anchors); and Section 10 of

_source/02_Tier_2_Region_Pages/ for flores_island (L844), bali (L1894),

central_africa_congo (L1017), ethiopia (L713), south_africa (L599), south_india (L787),

sri_lanka (L908), sumatra_java (L2229), swahili_coast (L838), rift_valley (L624) and

west_africa (L2398).

• Measurement substrate read: build/audio/exemplars/formula_cards/EX_124.json (schema

formula-card/v1), build/audio/exemplars/PATTERN_FINDINGS_V2.md,

harness/music_gen/exemplar_hook_signatures.json (13 signatures, 14 declared gaps),

harness/music_gen/nostalgia_score.py and docs/proposals/music/NOSTALGIA_RUBRIC.md.

• What this document realizes: the music_mood anchor's demand that a chapter's music be an

identity a player carries, and the region pages' Section 10 briefs, which already specify the

cultural substrate. This document supplies the craft layer between those two and never re-decides

either. It changes no canon; every canon-facing consequence below is stated as a finding.

1. What a hook is

1.1 Three traditions, one object

• The pop-craft definition is Burns's. In "A typology of 'hooks' in popular records" (Popular Music

6/1, 1987, 1-20, https://www.tagg.org/xpdfs/burns87.pdf) a hook is the part of a record that

stands out and recurs, and Burns's contribution is that the hook need not be melodic at all. He

sorts hooks into musical (rhythm, melody, harmony, instrumentation), textual (lyric), performance

(tempo, dynamics, improvisation) and production/technological (editing, mix, channel balance,

signal distortion, sound effects) families. The practical consequence for us is direct: a

gong-waning cycle, a whip-crack placed on a beat, a single detuned bronze pair, or a sudden

register drop are each admissible hooks by Burns's taxonomy, and a cue with no singable tune is

not automatically hookless.

• The classical definition is the theme or head motive: the identity-bearing opening idea that the

rest of the movement is heard against. Caplin's account (Classical Form, Oxford, 1998) makes this

operational rather than impressionistic — see section 2.

• The game-score definition is the one Josh is using: the tune you hum leaving the room. It is a

memory claim, not a structural one, and no purely structural measurement can certify it. That gap

is the honest centre of this document.

1.2 Hook against riff and ostinato

• Middleton's distinction is the useful one (Studying Popular Music, Open University Press, 1990,

summarized at https://mtosmt.org/issues/mto.02.8.4/leydon_text.html). Musematic repetition is

near-unvaried repetition of a museme — the smallest meaningful unit — and its paradigm case is the

riff or ostinato; it is circular, synchronic and open. Discursive repetition is repetition of

longer, syntactically complex units — phrases, strophes, sections — and it is linear and

self-sufficient.

• A riff is therefore not a weak hook; it is a different mechanism. A riff earns memorability by

saturation and groove, a hook by shape and return. A cue that has only musematic repetition is the

thing Josh named as stacked noise: material that recurs without ever being answered.

• CONSENSUS craft rule that follows: every cue needs at least one discursive unit — something that

begins, goes somewhere and comes back — even when its surface is an ostinato culture.

• PIPELINE HOOK: motif_economy.interval_3gram_repetition and interval_4gram_repetition measure

musematic density. motif_economy.self_similarity_lift (peak minus baseline) is the closest thing

the card has to a discursive-return measure. Corpus medians, MEASURED-HERE at N = 28: 3-gram

repetition 0.432 (p10 0.157, p90 0.791), self-similarity lift 0.181 (p10 0.058, p90 0.328). A

candidate with high 3-gram repetition and near-zero lift is a riff with no theme.

2. Phrase structure — the engineering under a memorable tune

2.1 The sentence

• Caplin's eight-bar sentence has three formal functions: presentation (beginning), continuation

(being in the middle) and cadential (ending). The presentation is a two-bar basic idea plus its

near-literal repetition, usually over a tonic prolongation. The continuation fragments the basic

idea, accelerates harmonic rhythm and increases surface activity. The cadential function

liquidates the idea's characteristic features and closes

(https://mtosmt.org/issues/mto.14.20.1/mto.14.20.1.aziz.html;

http://shanahdt.github.io/MUSI4331/lessons/phrases1.html).

• Why this is memorable is a prediction argument, not an aesthetic one. The presentation states a

unit and immediately proves it is a unit by repeating it; from bar 3 the listener can predict bar

4. The continuation then rewards that acquired competence by breaking the unit into pieces the

listener already owns. Fragmentation is only intelligible because the whole was stated twice

first.

• PIPELINE HOOK: nostalgia_score.py axis 4 already encodes this (P1 basic-idea repeat

similarity, hard-gated at 0.80; P2 continuation fragmentation; P3/P4 the 8-bar grid). The

census reports P1 at 3 of 14 — our landed cohort answers its head with a mirror instead of

repeating it (docs/proposals/music/NOSTALGIA_RUBRIC.md section 6). That is the single most

diagnosable phrase-level defect we have measured.

2.2 The period, elision and hypermeter

• The period pairs an antecedent ending in a weak cadence with a consequent that restates the basic

idea and closes strongly. Caplin treats antecedent/consequent as functionally comparable to

presentation/continuation, which is why a tune can feel classically shaped without being a

textbook example of either.

• Elision — the last bar of one unit being the first bar of the next — is the standard device for

removing the seam between phrases without shortening either. CONSENSUS.

• Hypermeter is the grouping of bars into higher-order strong and weak positions, canonically four

(Rothstein, Phrase Rhythm in Tonal Music, 1989, cited in

https://mtosmt.org/issues/mto.22.28.4/mto.22.28.4.temperley.pdf). Hypermetric regularity is what

makes a listener feel where a section will end before it ends, and hypermetric irregularity

(a five-bar or six-bar group) is a tension device precisely because that expectation exists.

• PIPELINE HOOK: not measurable from the current card. complexity_law.bar_length_s and bars give

bar arithmetic at an assumed 4/4 (tempo_and_time_grammar.metre.measured is null on EX_124), so

hypermeter has no representation. Closing measurement named in section 12.

3. Melodic anatomy

3.1 Contour archetypes

• Huron's Essen-corpus study of Western folksong found a predominance of convex (arch) phrase

contours over ascending and descending shapes ("The Melodic Arch in Western Folksongs", Computing

in Musicology 10, 1996; https://www.academia.edu/50292501/The_Melodic_Arch_in_Western_Folksongs).

The arch is the default phrase shape of a large, mostly European folk corpus — which is a fact

about that corpus, and the Essen database's over-use as a proxy for music in general is a known

methodological complaint (https://dl.acm.org/doi/10.1145/3469013.3469016).

• The axial (oscillation around a central pitch) and gap-fill (a leap answered by stepwise motion

filling the gap) archetypes come from Meyer and Narmour. Gap-fill originates with Meyer (1973):

large intervals imply smaller intervals in the opposite direction

(https://mutor-2.github.io/ScienceOfMusic/units/08/).

• CAUTION, and it matters: gap-fill as a listener-side classification principle is weakly supported.

Reanalyses find that gap-fill does poorly as a way listeners classify melodies

(https://www.researchgate.net/publication/271681259_Questioning_a_Melodic_Archetype_Do_Listeners_Use_Gap-Fill_to_Classify_Melodies).

• PIPELINE HOOK: the_hook.pitch_contour is a U/D/- string; EX_124 reads ----UDUD-UDUD--, which

is axial with repeated pitches, not an arch. Our own card carries a gap-fill-after-leap axis whose

hit median and sibling median are both 0.000 with AUC 0.533 (PATTERN_FINDINGS_V2.md section 10)

— gap-fill does not separate our corpus rows from their album siblings, and should be used as a

craft option, never as a gate.

3.2 Post-skip reversal, and the deflationary explanation

• Narmour's implication-realization model (1990, 1992) predicts melodic continuation from the last

two notes: a small interval implies continuation in the same direction, a large interval implies

reversal. Huron reduced it to five governing principles — registral direction, intervallic

difference, registral return, proximity, closure (Huron 1997; summarized at

https://mutor-2.github.io/ScienceOfMusic/units/08/).

• Von Hippel and Huron's deflationary alternative is that post-skip reversal is largely regression

to the mean: a leap tends to land far from the tessitura's centre, and the next note statistically

returns toward it, which looks like a reversal principle without being one ("Why Do Skips Precede

Reversals? The Effect of Tessitura on Melodic Structure", Music Perception;

https://www.researchgate.net/publication/224982434).

• The craft consequence is not that reversal is fake. It is that a composer who writes a leap and

then a step back has done the ordinary thing, and the ordinary thing is not a fingerprint. If a

leap is to be an identity, it has to be the same leap every time, in the same metric position.

• PIPELINE HOOK: the_hook.distinctive_deviation (largest interval, note index, is_licensed_leap)

and interval_vocabulary_semitones. EX_124 carries a single -14 semitone leap at note index 6 out

of an interval vocabulary of exactly four values {0, 10, 12, 14}. That is the shape to aim for:

a tiny vocabulary with one deliberate outlier at a fixed index.

3.3 Tessitura, span and rhythmic distinctiveness

• The singable-range constraint is a physiological fact plus a craft convention: a melody a listener

can reproduce without notation must fit an untrained voice. Our rubric encodes it as C5

(range <= 17 semitones, alphabet <= 8 pitches, tessitura MIDI 55-84), hard-gated

(NOSTALGIA_RUBRIC.md axis 2).

• Rhythmic distinctiveness and the anacrusis: a hook is identified as much by its rhythm as by its

pitches — Burns's rhythm-hook category exists for exactly this reason, and Kondo's own

three-note head cells are described by their half-quarter-half template as much as by their

intervals (NOSTALGIA_RUBRIC.md H7, citing ECON-07 rhythmic economy). An upbeat entry changes

which note lands on the downbeat and therefore which note is heard as the identity.

• Repeated pitches are not a defect. The Super Mario Bros. Ground Theme's own head is E E E and

carries two unisons in six intervals — a fact our own rubric had to write into feature I0 as a

declared derived floor after a texture control exposed the naive reading

(NOSTALGIA_RUBRIC.md axis 3).

• PIPELINE HOOK: register.melody_span_semitones and melody_tessitura_midi;

the_hook.range_semitones. Corpus hook range median 22.0 semitones, MEASURED-HERE, range

0.67-52.0 — far wider than the rubric's 7-17 singability band, because the card's hook extractor

reads the audio's melodic band and the rubric reads an authored head cell. These two numbers are

not comparable and must never be quoted against each other.

4. The empirical memorability literature, and its limits

4.1 Earworms

• Jakubowski, Finkel, Stewart and Müllensiefen matched 100 frequently-named involuntary-musical-

imagery tunes against 100 never-named tunes controlled for popularity and style, and compared them

on 83 melodic features (Psychology of Aesthetics, Creativity, and the Arts 11/2, 2017, 122-135;

https://www.apa.org/pubs/journals/releases/aca-aca0000090.pdf). The reported discriminating

features are faster tempo, more common global melodic contour patterns (the plain rise-then-fall

arch being a positive predictor), and unusual interval structure — larger or less expected leaps

than the contour alone would imply.

• What the study actually supports is narrower than the folk reading. Song popularity and chart

recency were themselves strong predictors; the musical effects were modest; the design is

correlational and the sample is UK self-report. The authors' own framing is that these properties

interact rather than that any single feature determines stickiness.

• The one usable design rule is a paradox and should be stated as one: ordinary shape, unusual

detail. Put the novelty in the interval and the harmony; keep the contour common. Our own rubric

already says this at axis 2 and cites the same paper.

4.2 Repetition and mere exposure

• Margulis (On Repeat, Oxford, 2013) argues repetition is what converts a listener from observer to

participant — repeated material is heard as something you are inside rather than something you are

watching (https://mtosmt.org/issues/mto.14.20.4/mto.14.20.4.albrecht.html).

• The mere-exposure effect (Zajonc 1968) makes repeated stimuli more liked; Berlyne's two-factor

model adds a tedium factor, producing the inverted-U: liking rises with exposure to a peak and

then falls. The peak's location is stimulus-dependent and is not a constant anyone has pinned

down (https://arxiv.org/pdf/2210.16226).

• This is the theoretical content of Josh's own 3-to-5-cycle law. He is asserting that for a game

cue heard in a loop, the tedium factor overtakes the habituation factor at roughly the fourth

cycle unless something new enters. That number is a ruling, not a measurement, and it is

consistent with the shape of the literature without being derivable from it.

• PIPELINE HOOK: duration_and_structure.repeat_map plus section_count. Corpus section-count

median is 17 against a sibling median of 9 (PATTERN_FINDINGS_V2.md section 10), the strongest

measured axis in our whole bank. A 17-section track that runs 196 s changes something roughly

every 11 s. That is the measured form of the 3-to-5-cycle law.

4.3 Expectancy models

• IDyOM (Pearce 2005) is a variable-order Markov model trained by statistical learning that emits,

per event, a probability distribution over the next event, and from it an information content

(surprise) and entropy (uncertainty). It outperforms static rule-based models such as I-R for

pitch expectation in many contexts (https://onlinelibrary.wiley.com/doi/full/10.1111/j.1756-8765.2012.01214.x;

https://nyaspubs.onlinelibrary.wiley.com/doi/10.1111/nyas.13654).

• Information content also functions as an empirical definition of perceived musical complexity

(Music Perception 37/2, 165; https://online.ucpress.edu/mp/article-abstract/37/2/165/109530), and

the predictability-liking relation recovered from it is the same inverted-U shape as section 4.2.

• HONEST LIMIT: IDyOM needs symbolic melody. We do not have transcriptions of our corpus or our

candidates, so no expectancy model can be run in our pipeline today. Named as a gap in section 12.

5. Hook placement and economy

• Arrival time. Léveillé Gauvin's attention-economy study measured 303 US top-10 singles from 1986

to 2015 and found time-before-voice-entry and time-before-first-title-mention both fell

significantly across the period (Musicae Scientiae 22/3, 2018, 291-304;

https://journals.sagepub.com/doi/abs/10.1177/1029864917698010), with the popular summary of the

same work putting average intros at over 20 s in the mid-1980s against about 5 s today

(https://news.osu.edu/has-music-streaming-killed-the-instrumental-intro/). This is a fact about a

distribution channel, not about memorability, and it should be read as one.

• Our corpus behaves the same way anyway. MEASURED-HERE, opening_and_flow_law.time_to_hook_s

across the 28 hits has median 0.575 s (p10 0.104, p90 5.12, max 61.6). Half of the tracks Josh

named put identifiable melodic material inside the first six-tenths of a second. But

time_to_hook_frac sits at AUC 0.344 against siblings (PATTERN_FINDINGS_V2.md section 4) —

the siblings do it too, so early arrival is a floor and not a differentiator.

• Texture arrival is separate from hook arrival and is much faster than intuition suggests.

MEASURED-HERE, time_to_full_texture_s / total_length_s has median 0.032 across the 28: the

arrangement is essentially open within the first three percent of the runtime. The corpus does not

do long ramps into the material. It states the thing, then varies it for three minutes.

• State it plainly once. CONSENSUS across the classical, pop and film literatures: the identity

should appear once in its plainest form before it is transformed, because every later variation is

parsed against that statement. Caplin's presentation function is the classical form of this rule;

the film-scoring version is the restraint argument — the theme lands at the climax because the

film withheld it (https://www.cinemagicscoring.com/post/behind-the-music-how-film-scores-work).

• The withheld full statement is the complementary device and is not in conflict with the previous

bullet: state a fragment plainly early, withhold the complete harmonized statement for a later

structural payoff. This is the design the leitmotif architecture already assumes.

• PIPELINE HOOK: opening_and_flow_law (onset_type, first_sound_at_s, time_to_hook_s,

time_to_hook_bars, time_to_full_texture_s) plus motif_economy.recurrence_lag_s. Note that

recurrence_lag_norm was V1's best hypothesis and is now the second-worst axis in the bank at

AUC 0.505 (PATTERN_FINDINGS_V2.md section 6) — do not build a delayed-return gate on it.

6. Hook in a non-vocal, non-Western, period-authentic frame

The slice's traditions are not tune cultures with exotic timbres. In several of them the

identity-bearing unit is not a melody at all, and the region pages already say so. The craft answer

per tradition:

6.1 The cycle is the hook (Bali, Java, Sumatra)

• Javanese and Balinese gamelan is stratified and cyclic: a core melody (balungan/pokok) is

punctuated by a colotomic structure of gongs marking a cycle of fixed length, with elaborating

parts above (https://ecampusontario.pressbooks.pub/beyondtheclassroom/chapter/balinese-music-gamelan/;

https://grokipedia.com/page/Colotomy). Kotekan is the interlocking of paired parts into a single

line faster than either player produces (https://en.wikipedia.org/wiki/Kotekan).

• The hook here is the cycle length plus the placement of the gong. A listener learns where the

cycle closes and hears everything against that. The identity a player carries out of the room is

a period and a punctuation pattern, not a tune.

• Our own canon states the same thing more strongly than any secondary source: bali.md Section 10

(L1894) says Bali is "the arc's first region where music is not scoring but structure" and that

the music "is built to be READ"; it also documents the ombak beat-note engineered by deliberately

mistuned pairs [SRC_00157, p. 316-317, p. 319-320]. sumatra_java.md Section 10 (L2229) states

plainly that its processed sources carry almost no instrumental ethnography and that the region's

sound is built around the voice; and the Minangkabau talempong pacik tradition is itself

interlocking, three players on six gong-chimes playing anak, tengah and peningkah.

• PIPELINE HOOK: cycle length is tempo_and_time_grammar.dominant_onset_period_s (EX_124: 2.879 s,

i.e. 20.84 cycles per minute) read together with section_boundaries_s. A colotomic cue should

show a strong, stable dominant onset period and section boundaries that are integer multiples of

it. Nothing in the card checks that multiple relation today; it is a cheap addition.

6.2 The phrase is speech (Ile-Ife/Yoruba, and the talking drum)

• Yoruba dundun drumming is a speech surrogate: the tension strings let the drummer track the three

lexical tone levels of Yoruba, and the repertoire consists of real texts — oriki praise poetry and

owe proverbs (https://www.frontiersin.org/journals/communication/articles/10.3389/fcomm.2021.652542/full).

Villepastour's study of the bata shows the encoding is no less precise for being less iconic

(https://ethnomusicologyreview.ucla.edu/journal/volume/15/piece/481).

• Our canon carries the same fact from its own shelf: west_africa.md Section 10 (L2398) quotes

the tradition's own scholar that the dundun "can use codes to produce a type of language which

nearly all Yoruba people can decode" [SRC_00260, p. 135], and records the Ifa corpus imitating the

drums by name [SRC_00229, p. 112].

• The craft answer: the hook is a fixed rhythmic-tonal phrase that means something. It recurs

verbatim because it is a sentence, and it is memorable for the same reason a catchphrase is. Write

a drum phrase of fixed length and fixed contour, use it as a call, and let the ensemble answer it.

Do not develop it — a proverb that is developed stops being a proverb.

• PIPELINE HOOK: none. The card has no percussion-phrase extractor. the_hook reads a pitched

melodic band and returns available: false or noise on a drum-led cue. Named in section 12.

6.3 The ostinato is a body (Ethiopian highlands, Sri Lanka, Tamil South India)

• Ethiopian secular music is organized around four pentatonic qenet — tizita, bati, ambassel,

anchihoye — carried on krar, begena, masenqo and washint

(https://en.wikipedia.org/wiki/Qenet; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10118148/).

Our own ethiopia.md Section 10 (L713) supplies the vernacular roster with descriptions attached

— karar the six-string lyre, bagana the eight-or-ten-string floor harp, masanqo the one-string

fiddle, kabaro, nagarit, sanasal the sistrum, maqwamiya the prayer-staff [SRC_00254, p. 104-105].

• The hook in a modal ostinato culture is the mode's characteristic degree sounded at a fixed point

over a fixed cyclic accompaniment. The listener's memory hook is the interval that identifies the

qenet, not a phrase.

• Carnatic practice gives the strongest formal model for the slice's South Indian chapter. A kriti

runs pallavi, anupallavi, charanam; the pallavi establishes the raga's identity and returns as a

refrain after each later section; the anupallavi rises in register and introduces contrasting

material; each line is then varied through composed sangati

(https://artiumacademy.com/blogs/what-is-kriti-in-carnatic-music/;

https://www.mtosmt.org/issues/mto.15.21.4/mto.15.21.4.schachter.html). That is a refrain-and-

graded-variation architecture that solves the loop-fatigue problem natively, and it is a far

better template for a 3-5 minute looping cue than a Western AABA.

• Our Sri Lanka canon points at voice, not ensemble: sri_lanka.md Section 10 (L908) documents the

all-night preaching hall and the protective chant with its procedure [SRC_00256, p. 138, p. 152;

SRC_00237, pp. 62-66]. The hook there is a recitation cadence and a room.

• PIPELINE HOOK: tonality.key / mode / windowed_key_agreement see major/minor only. A qenet or

a raga is invisible to the card. nostalgia_score.py Y12 (characteristic-degree rate) is the

right shape of feature and exists only for authored symbolic material.

6.4 The texture is the hook (Baka/Mbuti forest, /Xam San)

• Central African forest polyphony is dense contrapuntal communal improvisation built from short

hocketed motives and yodel register-shifts over interlocking ostinatos (Arom, African Polyphony

and Polyrhythm, Cambridge, 1991; https://en.wikipedia.org/wiki/Simha_Arom;

https://www.melodigging.com/genre/mbenga-mbuti-music). Among the Mbuti the singing uses short

hocket-like motives connected to the voco-instrumental hocket of the small flute.

• Our own central_africa_congo.md Section 10 (L1017) is more specific than any secondary source

and is already a complete composition brief: overlapping entries, each singer holding her note as

the next comes in, a chorus cascading downward like a waterfall, swelling until breath runs out,

then all stopping together sharply and cocking their heads to listen for the forest's echo

[SRC_00253, p. 171-172]. It names that listened-for reply as the region's musical signature.

• That is a hook made of an entry pattern and a cut-off — Burns's rhythm-and-production families,

not his melody family. The memorable object is the collective attack, the swell and the hard

unison stop.

• The /Xam case is different again: south_africa.md Section 10 (L599) records that the archive's

organizing principle is that a song belongs to a specific creature, person or state, and names the

catalogue — the Cat's Song, the Song of the Caama Fox, the Broken String and the rest [SRC_00265].

Our canon's own design line is a catalogue of short named songs discovered rather than triggered,

sparse, vocal-first, with a deliberate absence of harmony. Khoisan practice supplies the technique:

cyclical forms, interlocking parts, handclap polyrhythm, yodel-like register shifts, hocketing,

and a gradual upward drift over the course of a song

(https://www.melodigging.com/genre/khoisan-folk-music).

• PIPELINE HOOK: lane_analysis.entry_exit_map and simultaneity_curve are the right instruments

for entry-pattern hooks, but max_lanes is pinned at 9 for hits and siblings alike — a

measurement ceiling in the spectral_band_proxy method, confirmed across two full re-runs

(PATTERN_FINDINGS_V2.md section 10). No arrangement-width claim is admissible until a real

instrument roster exists.

6.5 Olduvai, and the honest case of no record

• rift_valley.md Section 10 (L624) opens by stating that no instruments are attested in its corpus

and that no rhythmic structure, scale, call pattern or instrument is described anywhere on the

page. The chapter's musical content is the Ju|'hoansi healing-dance clapping carried from ratified

canon at reference register, plus "a deep-time silence at Olduvai".

• The deep-time stratum has no ethnomusicology by construction. What exists is archaeology: the

Hohle Fels vulture-radius flute with five finger holes, roughly 35-40 thousand years old, plus

ivory flute fragments (https://www.sciencedaily.com/releases/2009/06/090624213346.htm). The

defensible register is therefore body percussion, voice, struck stone and end-blown pipe, and the

honest composition is near-absence with a floor of wind and stone, which is exactly what the

region page already specifies.

• FINDING, not an edit: the deep-time cue is the one place in the slice where the composition floor

("no track will be boring or filler") and the canon brief ("the arc's quietest bed") pull against

each other. The resolution available inside both is that the Olduvai bed be short, be event-bearing

rather than static — a single struck-stone figure that recurs at a long, irregular period — and

never be the chapter's only cue. This is a craft recommendation to the music lane, not a canon

change.

7. The tension parameters, one by one

Farbood's model is the best single empirical map. Both her experiments manipulated harmony, pitch

height, melodic expectation, dynamics, onset frequency, tempo, meter, rhythmic regularity and

syncopation, and her model is built on a moving perceptual window and trend salience — tension at a

moment is a function of the recent trajectory, not the instantaneous state (Music Perception 29/4,

2012, 387-428; http://mp.ucpress.edu/content/29/4/387). Lerdahl and Krumhansl's hierarchical model

combines prolongational structure, pitch-space distance, surface dissonance and attraction, and

implementations of it correlate strongly with listener tension ratings (Music Perception 24/4, 2007,

329-366; https://www.fredlerdahl.com/s/Modeling-Tonal-Tension.pdf).

7.1 Harmonic

• Dissonance, tonal distance from the prevailing key, an unresolved dominant, a pedal that holds

under changing harmony, and modal mixture all raise tension. Lerdahl and Krumhansl's pitch-space

distance is the formal version of tonal distance; surface dissonance is a separate additive term

in the same model.

• PIPELINE HOOK: not measurable today. tonality.windowed_key_agreement is a modulation proxy only,

and the card itself declares that a pivot-chord map needs symbolic transcription. Closing

measurement in section 12.

7.2 Melodic

• Register ascent raises tension; Farbood treats pitch height as a direct contributor. Tessitura

strain — a line sustained near the top of its range — is the extreme case. An unresolved leading

tone is the melodic form of an unresolved dominant.

• PIPELINE HOOK: register.melody_span_semitones, melody_tessitura_midi. There is no per-time

melodic-register curve on the card, only a global span; a time-series version is the cheap fix.

7.3 Rhythmic

• Acceleration, syncopation, and metric dissonance raise tension. Krebs's taxonomy splits metric

dissonance into grouping dissonance (pulse layers at different speeds, of which hemiola is the

commonest case) and displacement dissonance (same speed, misaligned)

(https://viva.pressbooks.pub/openmusictheory/chapter/metrical-dissonance/;

https://mtosmt.org/issues/mto.14.20.2/mto.14.20.2.biamonte.html). Hemiola clusters at cadences in

triple metre precisely because it heightens the tension the cadence then resolves.

• Subdivision doubling — the same tempo with twice the note rate — is the game-score workhorse

because it raises perceived intensity without breaking a loop's grid. CONSENSUS.

• PIPELINE HOOK: tempo_function.adjustment_events (count, direction), groove.swing_ratio,

groove.onset_rate_per_s, windowed_tempo_stability. Corpus tempo_adjust_n median 13.5 against

a sibling median of 6 (PATTERN_FINDINGS_V2.md section 10) — the tracks Josh named bend tempo

more than twice as often as their album neighbours. Metric dissonance itself is unmeasured.

7.4 Dynamic

• Crescendo raises tension continuously. A sudden change raises it discontinuously and more

powerfully: sudden loudness is one of the acoustic events that trips the brain-stem reflex, which

has no opt-out, and cross-cultural work correlates chill incidence with sudden peaks in loudness,

brightness and roughness

(https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2017.02044/full;

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4756174/).

• The subito is therefore worth more than the crescendo per unit of dynamic range spent, and a track

that has spent its range on a long ramp has none left for the surprise. This is the craft argument

against front-loading loudness.

• PIPELINE HOOK: dynamics.dynamic_range_db_p95_minus_p10, arc_curve_db, and

event_grammar.raises[].slope_db_per_s. Corpus dynamic-range median 11.7 dB against a sibling

median of 13.7 (AUC 0.369) — the corpus rows are slightly narrower, not wider, which is a warning

against reading dynamic range as a quality axis.

7.5 Textural

• Density accumulation and register expansion raise tension; the drop-out lowers it instantly and

is the strongest release gesture available without harmony. Sloboda's structural inventory found

shivers most reliably evoked by new or unexpected harmony and tears by appoggiatura and sequence

(Psychology of Music 19, 1991, 110-120;

https://journals.sagepub.com/doi/10.1177/0305735691192002), which locates the strongest affective

events in harmony and voice-leading rather than in texture — texture is the amplifier.

• PIPELINE HOOK: lane_analysis.simultaneity_curve, register_spacing_octaves,

event_grammar.drops[].lanes_removed, complexity_law.index_per_bar.

7.6 Timbral

• Brightness (spectral centroid rise), noise and roughness all raise tension. Roughness has the

best-grounded psychoacoustic account: Plomp and Levelt showed maximum sensory dissonance at

roughly a quarter of the critical bandwidth, with consonance recovering as the separation exceeds

it, and the complex-tone case computed as the summed roughness of all partial pairs (JASA 38,

1965, 548-560; https://www.semanticscholar.org/paper/1d3ccc073b1b3f13f95e2392ff81a9eff0e7a4d2).

Vassilakis's model is the standard modern implementation

(https://www.mat.ucsb.edu/Masters/Brian_Hansen_Masters.pdf).

• PIPELINE HOOK: timbre.spectral_centroid_curve_hz, spectral_rolloff85_hz_mean,

spectral_flatness_mean, signature_colour. Brightness is measured; roughness is not. Adding a

Plomp-Levelt/Vassilakis roughness curve is the single cheapest high-value feature we could add,

because it is computed from the spectrum we already extract.

8. Arch form, climax placement and release

8.1 Where the climax actually sits — our own answer

• The golden-section climax claim is the standard one and its evidential base is weak. The Bartok

case that made it famous is contested on its own arithmetic: Lendvai's reading of the first

movement of Music for Strings, Percussion and Celesta requires an added bar of rest to reach 89,

the dynamic climax is at bar 55 but the tonal climax is at 44, and Somfai's work on the sketches

found no proportional calculation of any kind

(https://blogs.ams.org/jmm2021/2021/01/08/did-bartok-use-fibonacci-numbers-in-his-music/).

• MEASURED-HERE, on the 28 corpus rows, using dynamics.peak_position_normalised and

opening_and_flow_law.flow_curve.peak_position_normalised:

• The honest reading: the central tendency of both peak measures sits close to the golden section

(0.618), and the distribution is nearly flat. Thirteen of twenty-eight tracks peak before the

midpoint on both measures. Golden

section describes the median of this pool and is not a rule any of these composers appear to have

followed. A generator that pins every peak at 0.618 would be more uniform than the corpus it is

imitating, which is itself a defect. Use 0.5-0.75 as a soft target and treat anything outside

0.16-0.87 as worth a second look, not a rejection.

• Per-purpose-class peak medians, MEASURED-HERE and purely descriptive at n = 2 to 8 per class:

BATTLE 0.672 dynamic / 0.282 flow; TENSION 0.704 / 0.809; CREDITS_TRIUMPH 0.543 / 0.493;

EXPLORATION 0.462 / 0.763; MELANCHOLIC 0.459 / 0.640; TITLE_CHARACTER 0.460 / 0.346;

BOSS 0.291 / 0.805. No per-class claim is a test; five classes sit below the five-row floor.

8.2 Single climax against multi-peak, and the release

• A single-climax design concentrates all the arrangement's reserve into one moment and is the

strongest shape for a 90-second cue. A multi-peak design is the shape a 3-5 minute cue needs,

because a single peak leaves too much runtime as approach and aftermath.

• The release is the part that is usually skipped and is the part that makes the peak mean anything.

CONSENSUS, and it follows from Farbood's moving-window result: tension is read from trajectory, so

a peak with no descent registers as a plateau. After a peak the arrangement must lose lanes, lose

register width, or lose harmonic motion — and it must do so for long enough to be heard as a

descent, not as a gap.

• The false climax is the standard device for buying a second peak: build fully, arrive, and then

either withhold the expected full statement or resolve deceptively, so the listener's spent

expectation is available to spend again.

• PIPELINE HOOK: dynamics.arc_curve_db and opening_and_flow_law.flow_curve.values carry the full

trajectory, so peak count, peak prominence and post-peak descent depth are all derivable today and

none of them is currently extracted. Adding peak-count and post-peak-descent-dB to the card is

low-cost and directly serves this section.

9. The drop

• The EDM break routine is breakdown, build-up, drop. Solberg's analysis identifies the concrete

mechanics: extensive uplifters and risers, the drum-roll effect, large frequency changes and

filter sweeps, the removal and reintroduction of bass and bass drum, and a contrasting breakdown —

all of which create tension and anticipation

(https://dj.dancecult.net/index.php/dancecult/article/view/451).

• The empirical test is Solberg and Dibben. Twenty-four participants heard four EDM excerpts

containing break routines and one without, with skin conductance, self-reported affect and

embodied experience measured. The emotional-intensity peak starts at or immediately after the drop

and continues into the core section, accompanied by greater motion and increased skin-conductance

response; dancers' overall activity level shows a sudden decrease then increase matching the

music's structure (Music Perception 36/4, 2019, 371-389;

https://online.ucpress.edu/mp/article/36/4/371/62966).

• Why removal creates more energy than addition: the drop is not a maximum, it is a contrast. The

breakdown removes the low end and the pulse, so the listener's reference level falls; when the

same bass returns it is heard against that lowered reference, not against the track's own opening.

This is the textural form of the subito argument in section 7.4, and it is why a drop preceded by

a full-texture build lands harder than a drop preceded by a louder build.

• The craft distinction between a drop that lands and one that merely stops is entirely about what

is guaranteed across the gap. A drop lands when the grid survives the removal — the listener can

still count, so the return is arrival rather than restart. A drop merely stops when tempo,

harmony and grid all vanish together, because then there is nothing to be early or late against.

CONSENSUS, and it is the operational content of Medina-Gray's meter parameter: strong smoothness

is all pulse streams continuing across the seam

(https://www.mtosmt.org/issues/mto.19.25.3/mto.19.25.3.medina.gray.html).

• Josh listed "beat drop" as a valid novelty event alongside a new instrument, a new catch or a new

melody. The rule that follows is that a drop is a structural event and gets structural preparation:

it needs a build it terminates and a re-entry it justifies. A drop with neither is the over-drop

defect measured in section 10.

• PIPELINE HOOK: event_grammar.drops[] already carries db_delta, lanes_removed, bars_held_s

and re_enters_at_s, and raises[] carries slope_db_per_s and lanes_added. A real drop is

identifiable today as a raise whose end is within a bar of a drop's start, with the drop's

bars_held_s at least a bar and a re-entry that restores at least the removed lanes. Nothing

computes that pairing. It is the highest-value derived feature in this document.

10. Expectation mechanics, and the 2.68-17.43 band

10.1 ITPRA

• Huron's ITPRA splits the response to an expected event into five systems across two epochs.

Pre-outcome: imagination (contemplating the future state) and tension (arousal and attention

raised immediately before the outcome, scaled by uncertainty, stakes and time-to-outcome).

Post-outcome: prediction (reward or punishment for the accuracy of the expectation itself),

reaction (fast, worst-case) and appraisal (slow, considered)

(Sweet Anticipation, MIT Press, 2006; https://mtosmt.org/issues/mto.09.15.3/mto.09.15.3.aversa.html).

• The design consequence that matters most is the prediction response: it rewards being right,

independently of whether the outcome was good. This is why a listener enjoys a resolution they saw

coming, and why a composer must establish a pattern before violating it — the violation is only

worth anything if the prediction response was engaged.

• Delayed resolution and the deceptive cadence both work by extending the tension epoch: the tension

response scales with estimated time to outcome, so holding the dominant longer literally raises

arousal before the resolution arrives.

10.2 Silence

• Margulis showed experimentally that musical silences are syntactically differentiated even though

they are acoustically identical apart from duration: silences following tonal closure are

identified faster and rated less tense than silences following unclosed music, and the preceding

material and the expectation of what follows both "seep into the gap" (Music Perception 24/5,

2007, 485-506; https://online.ucpress.edu/mp/article-abstract/24/5/485/95245).

• A rest is therefore a composed object whose value comes entirely from its harmonic and metric

context. A general pause after an unresolved dominant is tension; the same pause after a cadence

is punctuation. Our own card already declares this posture — the silence_budget field carries

the note that section 11.2 "treats rest as a FIRST-CLASS parameter, not an absence."

10.3 What our narrow band is really measuring

• X_rest_grammar is defined in harness/music_gen/derive_patterns.py L368 as

silence_frac * 100 + drops_per_min. It is a composite of how much of the track is below the

silence threshold and how often the texture is cut.

• Its status is precise and easy to get wrong. It FAILS as a separator: AUC 0.424 pooled, and it was

one of three carried hypotheses that died at N = 28 (PATTERN_FINDINGS_V2.md section 6). It

SUCCEEDS as a containment axis: the hit envelope 2.68-17.43 holds 26 of 28 held-out hits and is

one of the three axes in the joint box (section 8 of the same document). Corpus rows occupy a

narrow band of composed rest without sitting high or low on it.

• MEASURED-HERE, the two pools side by side:

• The candidate median drops the texture 2.53 times as often per minute as the corpus median, and

the worst candidate sits at 42.55 against a corpus ceiling of 17.43. Every one of the fourteen

rejections is an over-drop, not an under-drop.

• The craft reading, and this is the load-bearing sentence of section 10: a drop is a debt against

an expectation, and the expectation is built by the texture staying up. Cut every eight bars and

the listener stops predicting continuation, so removal stops being an event and becomes the

texture's normal behaviour — at which point it costs a lane and buys nothing. Farbood's

moving-window model says the same thing formally: tension is read from trend salience, and a

signal that reverses every few seconds has no trend to be salient. Over-dropping does not merely

fail to build tension; it destroys the mechanism by which tension could be built.

• The corpus figure to design toward is roughly one drop every ten to eleven seconds at the median

(5.56 per minute) with a hard ceiling near fourteen per minute, and a silence budget near two

percent. That is a descriptive band from a containment axis, and it is necessary-not-sufficient:

it rejects, it never certifies.

11. Long-form architecture for a 3-5 minute cue that loops

11.1 The two requirements are in tension, and the resolution is architectural

• A satisfying single listen wants an arc with a peak and a release. A loop wants a seam the

listener cannot hear. A cue that peaks at 0.6 and releases to nothing by 1.0 has a large

discontinuity at the wrap; a cue that is flat enough to wrap invisibly has no arc.

• The resolution is the refrain architecture, and the Carnatic kriti is its cleanest statement:

return to the pallavi after every section, so the loop point is a return the listener has already

heard four times rather than a splice. The Western equivalent is a rondo. Either way the wrap

falls on material that has already been established as a place the music comes back to.

• The second mechanism is the anti-fatigue cadence rule our own rubric already hard-gates: never

close the loop with a perfect authentic cadence (NOSTALGIA_RUBRIC.md Y1, from Gervais on

Zelda's Theme's avoidance of the tonic and of PACs). A loop that closes fully asks to stop; a loop

that closes on an open half cadence asks to continue.

• Medina-Gray's five smoothness parameters are the analytical form of the seam question: meter

(do the pulse streams continue), timbre (is the instrumentation shared), pitch (are the new pitch

classes inside the previous five seconds' macroharmony), volume (is loudness consistent across the

seam) and abruptness (is the entry a cut or a decay). She explicitly refuses to collapse them into

one number, because the correct weighting is context-dependent, and she notes that disjunction is

a legitimate design choice when the game wants the player to notice a state change

(https://www.mtosmt.org/issues/mto.19.25.3/mto.19.25.3.medina.gray.html).

• HONEST WARNING about our own corpus: it cannot teach us seam craft. MEASURED-HERE, the 28 hits'

cadence_and_ending.loop_seam.seam_ratio has median 24.17 (p10 5.34, p90 63.98), and EX_124 reads

seam_clean: false with baked_fade_out: true. These are album masters with fades, not game

loops. Any loop-seam standard must be derived from authored material or from extracted game loops,

never from this pool.

11.2 Repetition density has a measured, era-invariant home

• Hunt analysed 21,391 pieces of video game music across console generations and found the

repetition structure essentially consistent across platforms and generations, disproving his own

hypothesis that early hardware limits produced above-average repetition

(https://boblsturm.github.io/aimusic2020/papers/CSMC__MuMe_2020_paper_10.pdf;

https://uwe-repository.worktribe.com/output/6818560/). Our rubric's L4 band (0.55-0.70) is taken

from that survey's mean of 62.79% with SD 3.49.

• Our own corpus agrees from a different direction: repeat_sim_mean is above sibling median and

novelty_mean is below it, both clearing the calibrated bar, and the six clearing axes hold the

same sign and magnitude in the 16-bit half and the modern half of the pool

(PATTERN_FINDINGS_V2.md sections 4 and 9). Structural self-return is era-invariant across the

two eras we hold.

11.3 Per-purpose-class tension arcs

Recommendations below are craft syntheses grounded in the parameter evidence of section 7, the

class-descriptive medians of section 8.1, and the region-page briefs. They are proposal-tier.

• EXPLORATION — multi-peak and shallow. Three to five gentle rises with full releases between,

dynamic peak around 0.45-0.50, no single dominating climax, wrap on an open cadence over a

returning refrain. Novelty enters as instrument swaps rather than as dynamic events.

• BATTLE — front-weighted with a late dynamic peak. Full texture inside the first bars (corpus

time_to_full_texture median is 3% of runtime), flow peak early at about 0.28 and the dynamic

peak late near 0.67, so the arrangement is wide throughout while loudness still has somewhere to

go. One drop, held one to two bars, in the final third.

• BOSS — phased, with the flow peak very late (class median 0.81) and the dynamic peak deliberately

earlier. Each phase gets its own build and its own release; the last phase withholds the full

harmonized statement of the boss motif until the final section.

• MELANCHOLIC — single peak just before the midpoint (class median 0.46), long release, silence

budget at or slightly above the corpus median. Tension is carried by harmony and appoggiatura

rather than by density; Sloboda's tears-inducing structures are the target.

• TENSION — the only class that should sit high on both measures (class medians 0.70 dynamic, 0.81

flow). Sustained unresolved harmony, one long delayed resolution, drops used sparingly and always

with a preceding build.

• TITLE_CHARACTER — statement first, development after. Plain hook inside the first second (the

corpus median time_to_hook_s is 0.575 s), dynamic peak near 0.46, and the definitive full

statement of the theme reserved for the last third.

• CREDITS_TRIUMPH — the most symmetrical class in our pool (0.54 dynamic, 0.49 flow). Build to a

central peak, release, then a second smaller peak that quotes the arc's earlier themes. This is

the one class where a perfect authentic cadence is correct, because it does not loop.

12. The nostalgia rubric — what it gets right, and what it is missing

12.1 What it gets right

• It refuses a single scalar (LAW 2) and reports per feature. That is the correct response to the

fact that our own composite is 0.744 with all axes and 0.497 with the length-independent ones —

a single number would hide exactly that.

• It declares what it cannot measure rather than counting it as a pass (LAW 3, nine

NOT_COMPUTABLE features excluded from the denominator). Section 11 of this document reaches the

same conclusion about seam measurement from the corpus side.

• It hard-gates the anti-fatigue cadence (Y1) and the multi-motif loop (L1, L2), which are the

two mechanisms with the most direct bearing on a cue surviving a hundred hearings, and it grounds

L2 in Kondo's own stated method.

• Its anti-plagiarism tooth is mutation-proved in both directions and honest about its seed set: 13

signatures with 14 declared gaps, and CLEAR explicitly meaning "no seed signature matched" rather

than "clean".

• Its census correctly localised the landed cohort's failure — not the head cell, but everything

after it.

12.2 What this research says it is missing

• No tension axis at all. The rubric has seven axes and none of them measures tension, release,

climax placement or arc shape. Every principle in sections 7 through 10 above is outside its

scope. This is the largest structural gap.

• No drop or event grammar. Josh's floor names the beat drop as a valid novelty event; the rubric

cannot see one, because it scores a symbolic head cell rather than an arrangement.

• No rest placement, only rest density. C6 gates rest_density_melody >= 0.15, which is a budget.

Margulis's result is that rest VALUE comes from context — the same silence after a cadence and

after an unresolved dominant are different objects. A density gate cannot distinguish them.

• The 3-to-5-cycle variation law is unrepresented. L2 requires three or four motifs per loop but

says nothing about their scheduling, and Josh's floor is explicitly about scheduling: no more than

three to five cycles without a new instrument, catch, melody or drop.

• Two incommensurable hook definitions coexist in the stack. The rubric's C3 bands melodic range

at 7-17 semitones from a symbolic head cell; the formula card's the_hook.range_semitones reads

the audio's melodic band and has a corpus median of 22. Neither is wrong; they measure different

objects, and nothing in the stack says so. FINDING: the two should be named distinctly

(head_cell_range against measured_melodic_band_range) before either is quoted in a verdict.

• The seed set's 14 declared gaps are still open, and the rubric says correctly that a fingerprint

invented from memory is a false negative wearing a green tick. With the Nintendo discs arriving,

direct transcription of the named exemplars is now unblocked as a polish pass.

• No per-tradition dialect layer. The rubric's dialect is Kondo/SNES-JRPG throughout. Sections 6.1

through 6.5 above describe five identity mechanisms — cycle, speech-phrase, modal ostinato,

entry-pattern, named-song catalogue — that the rubric would score as failures for not being tunes.

FINDING: the rubric needs a declared dialect parameter, or its verdicts will push every region

toward the same Western dialect the region pages explicitly refuse.

13. The measurement gaps, named

Principles that can be enforced as COMPOSITIONAL CONSTRAINTS in the authored path today:

• Phrase architecture (sentence, basic-idea repeat, fragmentation, 8-bar grid) — nostalgia_score.py

axis 4, pre-render.

• Hook economy (small interval vocabulary, one licensed leap at a fixed index, small pitch alphabet,

tappable rhythm) — axes 1 through 3, pre-render.

• Loop cadence discipline (no PAC at the wrap, open half cadence, refrain return) — Y1, L6.
• Motif count and contrast per loop — L2, L3.
• Section count and length targets (the six bar-clearing axes) — enforceable as a spec, not as a

prompt, since the generator responds on only 4 of 8 probed dimensions and section_count is not

among them.

• Drop scheduling (a drop must follow a build and precede a re-entry; drops per minute inside

1.5-14) — enforceable as an arrangement rule in the authored path.

Principles measurable only after the fact, from the rendered audio:

• Climax position and arc shape — dynamics.arc_curve_db, flow_curve.
• Rest grammar — silence_budget plus event_grammar.drops.
• Brightness trajectory — timbre.spectral_centroid_curve_hz.
• Loop seam distance — cadence_and_ending.loop_seam.

Principles NOT measurable from the current card, with the measurement that would close each:

• Harmonic tension. Close with a chroma-based tonal-tension estimator. Herremans and Chew's spiral

array gives three computable quantities — cloud diameter, cloud momentum and tensile strain (the

distance between local and global tonal context) — with a Python implementation in midi-miner

(http://dorienherremans.com/sites/default/files/paper_tenor_dh_preprint_small.pdf). Running it on

audio requires a chroma-to-pitch-set step we do not have; running it on the authored symbolic path

requires nothing new.

• Sensory dissonance and roughness. Close with a Plomp-Levelt/Vassilakis roughness curve computed

from the spectrum already extracted for timbre. This is the cheapest closure in this list.

• Melodic expectancy and surprise. Close with IDyOM information content over a transcribed melody

(https://www.sciencedirect.com/science/article/abs/pii/S0165027024002929 for the current Python

implementation). Blocked on melodic transcription, which is also blocking cadence census, mode

identification and the hook-signature seed set — one blocker, four consumers.

• Hypermeter and metric dissonance. Blocked on a measured time signature; metre.measured is null

and bar arithmetic assumes 4/4.

• Instrument roster and arrangement width. Blocked on the spectral_band_proxy ceiling

(max_lanes = 9 for hits and siblings alike across two re-runs). Source separation or a real

roster is the only route.

• Percussion-phrase hooks. No extractor exists, so the entire west-African and Sri Lankan drum

repertoire is invisible to the_hook.

THE SINGLE MOST IMPORTANT GAP: melodic transcription. Without a symbolic melody line, we cannot run

an expectancy model, cannot compute a cadence census, cannot identify a mode or a raga or a qenet,

cannot expand the anti-plagiarism seed set beyond its 13 signatures, and cannot connect the

rubric's authored-side features to the card's rendered-side features at all. Every other gap in this

list is a feature; this one is the bridge between our two measurement worlds.

14. Sources

• Burns, Gary. "A typology of 'hooks' in popular records." Popular Music 6/1 (1987), 1-20.

https://www.tagg.org/xpdfs/burns87.pdf

• Caplin, William. Classical Form (Oxford, 1998) / Analyzing Classical Form (Oxford, 2013).

https://mtosmt.org/issues/mto.14.20.1/mto.14.20.1.aziz.html ·

http://shanahdt.github.io/MUSI4331/lessons/phrases1.html

• Middleton, Richard. Studying Popular Music (Open University Press, 1990), musematic and discursive

repetition. https://mtosmt.org/issues/mto.02.8.4/leydon_text.html

• Rothstein, William. Phrase Rhythm in Tonal Music (1989); Krebs, Harald. Fantasy Pieces (1999).

https://viva.pressbooks.pub/openmusictheory/chapter/metrical-dissonance/ ·

https://mtosmt.org/issues/mto.14.20.2/mto.14.20.2.biamonte.html

• Huron, David. "The Melodic Arch in Western Folksongs." Computing in Musicology 10 (1996).

https://www.academia.edu/50292501/The_Melodic_Arch_in_Western_Folksongs

• Huron, David. Sweet Anticipation (MIT Press, 2006), the ITPRA theory.

https://mtosmt.org/issues/mto.09.15.3/mto.09.15.3.aversa.html

• Narmour, Eugene. The Analysis and Cognition of Basic Melodic Structures (1990) and of Melodic

Complexity (1992); Meyer 1973 on gap-fill. https://mutor-2.github.io/ScienceOfMusic/units/08/

• von Hippel, Paul, and David Huron. "Why Do Skips Precede Reversals? The Effect of Tessitura on

Melodic Structure." Music Perception. https://www.researchgate.net/publication/224982434

• Gap-fill critique: "Questioning a Melodic Archetype: Do Listeners Use Gap-Fill to Classify

Melodies?" https://www.researchgate.net/publication/271681259

• Jakubowski, Finkel, Stewart and Müllensiefen. "Dissecting an earworm." Psychology of Aesthetics,

Creativity, and the Arts 11/2 (2017), 122-135.

https://www.apa.org/pubs/journals/releases/aca-aca0000090.pdf

• Margulis, Elizabeth Hellmuth. On Repeat (Oxford, 2013).

https://mtosmt.org/issues/mto.14.20.4/mto.14.20.4.albrecht.html

• Margulis, Elizabeth Hellmuth. "Silences in Music are Musical Not Silent." Music Perception 24/5

(2007), 485-506. https://online.ucpress.edu/mp/article-abstract/24/5/485/95245

• Pearce, Marcus. IDyOM; "Auditory Expectation," Topics in Cognitive Science (2012); "Statistical

learning and probabilistic prediction in music cognition," Annals NYAS (2018).

https://onlinelibrary.wiley.com/doi/full/10.1111/j.1756-8765.2012.01214.x ·

https://nyaspubs.onlinelibrary.wiley.com/doi/10.1111/nyas.13654 ·

https://www.sciencedirect.com/science/article/abs/pii/S0165027024002929

• Farbood, Morwaread. "A Parametric, Temporal Model of Musical Tension." Music Perception 29/4

(2012), 387-428. http://mp.ucpress.edu/content/29/4/387

• Lerdahl, Fred, and Carol Krumhansl. "Modeling Tonal Tension." Music Perception 24/4 (2007),

329-366. https://www.fredlerdahl.com/s/Modeling-Tonal-Tension.pdf

• Herremans, Dorien, and Elaine Chew. "Tension Ribbons: Quantifying and Visualising Tonal Tension."

TENOR 2016. http://dorienherremans.com/sites/default/files/paper_tenor_dh_preprint_small.pdf

• Plomp, R., and W. J. M. Levelt. "Tonal Consonance and Critical Bandwidth." JASA 38 (1965),

548-560. https://www.semanticscholar.org/paper/1d3ccc073b1b3f13f95e2392ff81a9eff0e7a4d2 ·

Vassilakis roughness: https://www.mat.ucsb.edu/Masters/Brian_Hansen_Masters.pdf

• Sloboda, John. "Music Structure and Emotional Response: Some Empirical Findings." Psychology of

Music 19 (1991), 110-120. https://journals.sagepub.com/doi/10.1177/0305735691192002

• Solberg, Ragnhild. "'Waiting for the Bass to Drop.'" Dancecult (2014).

https://dj.dancecult.net/index.php/dancecult/article/view/451

• Solberg, Ragnhild, and Nicola Dibben. "Peak Experiences with Electronic Dance Music." Music

Perception 36/4 (2019), 371-389. https://online.ucpress.edu/mp/article/36/4/371/62966

• Medina-Gray, Elizabeth. "Analyzing Modular Smoothness in Video Game Music." MTO 25.3 (2019).

https://www.mtosmt.org/issues/mto.19.25.3/mto.19.25.3.medina.gray.html

• Hunt, Samuel J. "An Analysis of Repetition in Video Game Music." CSMC/MuMe 2020.

https://boblsturm.github.io/aimusic2020/papers/CSMC__MuMe_2020_paper_10.pdf

• Léveillé Gauvin, Hubert. "Drawing listener attention in popular music." Musicae Scientiae 22/3

(2018), 291-304. https://journals.sagepub.com/doi/abs/10.1177/1029864917698010 ·

https://news.osu.edu/has-music-streaming-killed-the-instrumental-intro/

• Arousal and sudden acoustic change:

https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2017.02044/full ·

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4756174/

• Golden-section critique (Lendvai, Howat, Somfai on Bartók):

https://blogs.ams.org/jmm2021/2021/01/08/did-bartok-use-fibonacci-numbers-in-his-music/

• Arom, Simha. African Polyphony and Polyrhythm (Cambridge, 1991).

https://en.wikipedia.org/wiki/Simha_Arom · https://www.melodigging.com/genre/mbenga-mbuti-music

• Yoruba speech surrogacy:

https://www.frontiersin.org/journals/communication/articles/10.3389/fcomm.2021.652542/full ·

Villepastour, Ancient Text Messages of the Yorùbá Bàtá Drum:

https://ethnomusicologyreview.ucla.edu/journal/volume/15/piece/481

• Gamelan colotomic structure and kotekan: https://grokipedia.com/page/Colotomy ·

https://en.wikipedia.org/wiki/Kotekan ·

https://ecampusontario.pressbooks.pub/beyondtheclassroom/chapter/balinese-music-gamelan/

• Carnatic kriti and sangati: https://artiumacademy.com/blogs/what-is-kriti-in-carnatic-music/ ·

https://www.mtosmt.org/issues/mto.15.21.4/mto.15.21.4.schachter.html

• Ethiopian qenet: https://en.wikipedia.org/wiki/Qenet ·

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10118148/

• Khoisan musical practice: https://www.melodigging.com/genre/khoisan-folk-music
• Minangkabau talempong pacik:

https://www.auralarchipelago.com/auralarchipelago/talempongbotuang

• Hohle Fels Paleolithic flutes: https://www.sciencedaily.com/releases/2009/06/090624213346.htm
• Repo canon and instruments: docs/spine/CH_03.md L245, CH_04.md L242, CH_10.md L224;

_source/02_Tier_2_Region_Pages/*.md Section 10; build/audio/exemplars/PATTERN_FINDINGS_V2.md;

build/audio/exemplars/formula_cards/EX_124.json; harness/music_gen/derive_patterns.py L368;

harness/music_gen/nostalgia_score.py; docs/proposals/music/NOSTALGIA_RUBRIC.md;

harness/music_gen/exemplar_hook_signatures.json.

15. Reproduction of the MEASURED-HERE numbers

Every MEASURED-HERE figure in this document is read from the formula cards on disk through the same

fields derive_patterns.flatten uses. The 28 corpus rows are the hit list in

PATTERN_FINDINGS_V2.md section 2; the 32 candidates are

build/audio/generated/slice_v1/formula_cards/*.json. Fields used: dynamics.peak_position_normalised,

opening_and_flow_law.flow_curve.peak_position_normalised, opening_and_flow_law.time_to_hook_s,

opening_and_flow_law.time_to_full_texture_s, event_grammar.counts.drops,

event_grammar.silence_budget.fraction, duration_and_structure.total_length_s,

duration_and_structure.section_count, complexity_law.boundary_step_share,

cadence_and_ending.loop_seam.seam_ratio, motif_economy.self_similarity_lift,

motif_economy.interval_3gram_repetition, the_hook.range_semitones. No audio was decoded, no

exemplar bytes moved, and no calibrated multiplicity bar was computed for any of them — they are

descriptive reads, and the section 8.1 and section 10.3 tables are the two that matter.

Generated by harness/site/structure_site.py — the URL path is the repo path. review root