pipelines/CRITIC_L-R2-otherworld-ruleset-impossible-space.md
Fresh-context adversarial refutation pass, 2026-07-28. Draft under review:
scratchpad/realm_wave/L-R2-otherworld-ruleset-impossible-space_DRAFT.md (20 findings).
Every citation below was opened in C:\dev\humanity-forgotten (read-only); every absence claim
was re-run independently rather than inherited from the draft.
Headline: GO WITH FIXES. 19 findings stand, 1 is downgraded, 3 items are killed. The audit's
evidence base is unusually strong — the nine-row threshold table, the per-chapter agility-course
counts, and the re-entry positive control all reproduce EXACTLY against live rows. Its defects are
concentrated in three places: one truncated-read false zero in its own control table, one
mis-cited ruling, and two recommendation clauses that fail the devil's-advocate-against-timidity
pass.
---
The draft asserts:
| The R-A aspect rows are not minted |docs/DESIGN_GAP_REGISTER.mdterm counts:realm0,
Realm0,physics0,ruleset0,impossible0,gravity0,otherworld0,aspect0 |
zero | same file, same reader:companion10,DIVERGENT9. The matcher demonstrably
matches; the zeros are real |
Re-run, same file, case-sensitive: realm 8 · Realm 1 · REALM 4 · ruleset
1 · Otherworld 1 · aspect 6. (physics, impossible, gravity are genuinely 0.)
Root cause is stated in the draft's own dedupe baseline: it read "docs/DESIGN_GAP_REGISTER.md
(all 375 lines…)". The file is 464 lines. The first realm token is at L384 — nine lines
past where the read stopped. The corpus was truncated, so the search could not match. This is the
a-search-that-cannot-match-reports-zero class the wave rule exists to prevent, committed inside
the control table built to prevent it.
What the truncation concealed, decisively:
per-aspect teacher state (3 taught / 4 partial / 13 teacherless) and definitions live in
docs/pipeline_review/REALM_ANALYSIS_LENS_GAPS_2026-07-27.md; the realm doctrine wave (L-R1 +
L-R2 + L-R3 + register-variety + corruption-as-condition) is the claiming wave for
R-A1/2/3/4/6/7a/9/11/12/13/14/15/18/20." The R-A rows ARE minted and L-R2 is named as their
claiming wave — the exact opposite of the claim row.
REALM_DESIGN_PROGRAM §4.4) | L-R1 + L-R2 (realm doctrine wave) + hades-death-run-realm-identity |
CLAIMED-RUNNING (realm Wave 1)". This audit's own unit is already a register row.
"L-R3's per-realm care tiering is the gating deliverable for every later art wave."
No finding dies from this. Nothing in the concealed region pre-covers a schema-field finding,
and the two DIVERGENT-BY-DESIGN entries the draft respected are real. But the control row must be
struck, the dedupe baseline corrected to 464 lines, and §1 must record that the register read was
truncated — otherwise the next reader inherits a false "unminted" premise.
Collateral, same cause: the divergence citations are off by two. "Markerless navigation (our
wayfinding legibility floor)" is at L207, not L205; "World readability & wayfinding at
pace-of-play" is at L213, not L211. Both entries are real and correctly respected.
Three independent failures.
(a) The ruling is mis-cited. The draft sources the pip meter to "docs/COMBAT_PROGRAM_ADDENDUM.md".
That file contains pip ×1 (inside "pipeline"), wipe ×0, "in a row" ×0, puzzle boss ×0 across
270 lines, and does not appear in a repo-wide grep for pip/wipe/chain-meter. The ruling's real home
is docs/FUN_REBUILD_PLAN.md §R2 (plus docs/PRE_5090_BUILD_PLAN.md).
(b) The finding's premise is already answered by the ruled design. The draft argues the
wipe-meter on a perception rule "is the guess-and-check trap with a progress bar attached."
FUN_REBUILD_PLAN.md R2.3 already rules the anti-degeneration mechanism:
the pair eases monotonically as pips fill (cadence slows, the safe lane widens, one serpent
withdraws) — the window never re-tightens; difficulty comes from varying WHICH counter-class
the next attack demands, not from shrinking the window
The audit did not read the section it was arguing against.
**(c) Devil's-advocate-against-timidity: this re-litigates a Josh ruling that already survived this
exact challenge. FUN_REBUILD_PLAN.md L81-82: "the chain WIPES on a mistake** (Josh's ruling
wins over the critic's banked-progress). See the reconciliation note." A fun critic proposed
softening the wipe once; Josh ruled against it. Re-proposing the softening as a *schema constraint*
is the same call in different clothing. Worse, the ban would forbid the ruled puzzle-boss meter from
running in Ch 71 specifically — the realm REALM_DESIGN_PROGRAM.md §1.1 calls "the flagship of
the minigame-as-boss category" and the spine types dominant_combat_mode puzzle_routed. That is a
schema-level ban on the flagship mechanic in its flagship venue.
Surviving half — KEEP. consequence_class ∈ `{position_loss, resource_cost, time_cost,
information_cost, damage}` is reuse-first and grounded in a real, verbatim-verified precedent:
docs/spine/CH_69.md L82 reads "at least two counters per clause, **a wrong declaration costs
POSITION never locks". Type the enum; drop the ban. Verdict downgrades WEAKER → NOTE**.
protected: false requirement on taught_at. Struck on two grounds.(i) It is self-defeating — it makes Ch 72 unsatisfiable. Read off the live §3b rows, Ch 72's
four sites are:
| site | protected | threat | affordance | functions |
|---|---|---|---|---|
patala_naga_depths | true | low | [sacred] | discovery, traversal |
narakas_weighing_place | true | low | [threshold, sacred] | traversal, investigation, learning |
world_axis_loka_rungs | false | high | [climactic_geometry, sacred] | discovery, combat, vril_management, traversal |
churned_sea_summit | true | low | [vista, sacred] | discovery, character, traversal |
The only site satisfying the draft's own threshold + {none, low} test is
narakas_weighing_place — which already carries the learning gameplay function and is therefore
the obvious teaching zone. Clause (b) disqualifies it, and every other Ch-72 candidate fails on
affordance or threat. Result: Ch 72 has no legal threshold_site, and the Wave-3 pass is forced
to author a NEW non-protected site inside the road's highest-care realm. A constraint that no row
can satisfy, whose remedy is more invented content in the most sensitive realm, is a defect.
(ii) It is care-INFLATED. protected is the §17.1 protected-community routing flag
(harness/check_build_readiness.py:48 — "field for the §17.1 protected-community routing"). What it
governs is ALTERATION and desecration, not presence or instruction: the live env-grammar precedent
EG_0002 "Sieidi Stone (protected)" sets alteration_permanence_default: hard_blocked and
env_boss_verb_disabled: TRUE — and nothing about being observed or learned from. The spine itself
assigns narakas_weighing_place the learning function, and CH72_B01's env line reads "the nagas
courteous and enormous and honored, never fought." Reading "this culture is living and honored" as
"the game may teach nothing here" is care-as-omission — precisely what the Care Doctrine bans
("care means weight and authenticity, never omission").
Clause (a) — the ORDERING clause — SURVIVES and is the strongest half of the finding. taught_at
must resolve to a site hosting a beat whose seq precedes the first beat where the law is
load-bearing. Both joins exist and are already validated: harness/check_spine_graph.py L508-521
("§3b sites: every beat's site (!= none) resolves to a declared site row"). Xibalba's only low
site being xibalba_climb_out_road, the EXIT, is verified and is a genuine tooth-defeating case.
---
The nine-row threshold table (L-R2-15) reproduces EXACTLY. All nine B01 beats and their §3b
rows were re-read: CH66_B01→realm_road_threshold [threshold] low false · CH67_B01→
vault_and_lulled_hall [sacred, climactic_geometry] moderate · CH68_B01→
primordial_deep_creation_floor high · CH69_B01→duat_gated_hours [threshold] high · CH70_B01→
acheron_shore [threshold] low · CH71_B01→xibalba_descent_road [threshold] moderate · CH72_B01→
patala_naga_depths [sacred] low protected true · CH73_B01→arawns_court_feast_hall [sacred] low ·
CH74_B01→trunk_and_boughs [vista, weather_dramatic] moderate. 4/9 threshold, 4/9 low, 2/9 both —
all correct. Note the author did NOT lazily take the first-listed row (CH_74 lists
three_wells_and_norns_hall first); it read the actual B01 join. Good work.
L-R2-14's arithmetic reproduces EXACTLY — agility_course per road chapter: 66×1, 67×1, 68×2,
69×2, 70×1, 71×1, 72×2, 73×3, 74×2, with HL_0079 cited 2-3× each. duat_twelve_gated_descent_course
confirmed at CH69_B02's systems line. All five earthly precedents confirmed present (Ch 3 mbaru ×15,
Ch 4 Uluwatu ×11, Ch 6 Sigiriya ×19, Ch 9 Lalibela ×28, Ch 11 Ituri ×22). COVERED is the right
verdict and this is the audit's best cost-reducing finding.
L-R2-17's positive control is exemplary. Across CH_66-74: reenter 0, reversible 0, revisit
0 — confirmed; re-enter/re-entry = 4, all four in CH_70 — confirmed; and the control term
threshold returns 181 across the same nine files, matching the draft's figure to the digit.
MISSING is correct.
L-R2-3's gate mechanics verified verbatim. `COUNTER_CLASSES = ("any", "dodge_only",
"parry_only", "neutralize") at check_counter_class_coverage.py:65; def a1_coverage` at L129 with
the g > n late-grant failure at L168-172; the UNREACHABLE idiom at L158-163 reading "a search that
cannot match must never report a clean pass". The monotone-unlock characterisation is correct — a
REVOKED capability passes A1 green by construction. Feeding ability_availability_delta into A1 as
an input, typed over the gate's own four-value vocabulary, is genuinely reuse-first and is the
highest-value item in the handoff.
L-R2-10 STRENGTHENED, not weakened. The draft's grep was under-inclusive (`SetGravity|GravityZ|
CustomGravity misses GravityScale/GravityDirection`). I re-ran a case-insensitive search for
gravity across all of Source/Humanity: zero hits, against a control (EnvironmentGrammar →
Public/Data/EnvironmentGrammarRegistryRow.h) that resolves. The zero is real and the finding is
better-founded than its own control.
L-R2-7's citations are verbatim-accurate throughout — T1_Vril [ACTIVE v2.2] §1.1: "A rift
dungeon is a MEMBRANE-POCKET: a pocket held open at the membrane"; "negative vril pools and hardens
in the vacated substrate-gap"; "The player-facing verb is ENTER THE RIFT"; HL_0005 ruin/tomb
placement; "Pop-up rift dungeons instantiate as membrane-pockets in the open world through the CVD
Pillar 1 Rift Instability Zones system"; Director ruling D-2, 2026-07-27. `T0_Hollow_Codex [DRAFT
v0.1] = 1 row, spawn_locus_type drain-deadzone`. Charter §1.3 "There is no inside-the-rift canon
at all" and §2.3's confirmation block both quoted correctly. Routing to the director is correct.
L-R2-8's schema citations verified verbatim — WGM §14.6's five closed values (vista · threshold ·
sacred · climactic_geometry · weather_dramatic); §14.2's eight zone fields with no terrain and no
physical-rule field; SPINE_SCHEMA_v3.md L56 (the site row) and L58 ("terrain feeds §1.2 PCG +
§6.5 required_zone_type"); CH66_B02's "attune the orichalcum resonance across the canal-stations in
Plato's ring-city geometry". The ART-DIRECTION HOOKS family does carry no structure field. MISSING is
correct.
L-R2-16's five arrival-beat quotes all verified verbatim (Ch 66, 70, 71, 73, 74 objectives).
Headline §2 verified — the four terrain values (Yggdrasil forest, Meru mountain, Xibalba
cave_system, Duat water_body at CH_69 L69) and the HumanityDensitySubsystem.h class law
("Difficulty IS the vril-density dial, and NOTHING else … telegraph slack, window width,
recharge-well latency, answer-pruning, Withdrawal aggression"; "Set by REALM + SITE").
Boundary HELD — R-A7a/R-A7b. FORKC_ANALYSIS_DESIGN_PRECEDENT_2026-07-27.md §0 FACT 3 verified
verbatim including the seam counts: Ch 66-69 dimensional transit ~8, "the same Vimana,
dimensional-space mode"; Ch 70-77 within-realm ~9, "nothing — same space, keep walking".
REALM_ANALYSIS_LENS_GAPS_2026-07-27.md L81 pre-registers R-A7b as owned live elsewhere, and L251
confirms L-R2's declared aspect scope. **No finding in the draft proposes a conveyance, load-mask,
travel card, or transition treatment.** The fork is not re-litigated. Clean.
§7's seven dropped candidates all check out, including the correctly-respected pre-registrations
(the markerless divergence; the vril_density vocabulary split at REALM_DESIGN_PROGRAM.md §5.1
Wave 0, confirmed present at L1232). The self-falsified "five of nine realms have no low-threat site"
candidate is honest work and the narrower claim that replaced it is the provable one.
---
1. L-R2-3 — ability_availability_delta occurs in four files (REALMARC_PATTERN_BRIEF,
REALMARC_PATTERN_CANDIDATES, REALM_ANALYSIS_HARMONY_DOCTRINE, REALM_DESIGN_PROGRAM), not
"exactly two files, both proposals." The load-bearing half — zero harness consumers — is verified.
Split the verdict honestly: the FIELD is DESIGNED-UNBUILT; the GATE-BLINDNESS is MISSING.
2. L-R2-6 — the live token is multi-construct:varies, a **well-formed key:value pair with an
out-of-vocabulary KEY, on two** rows (B_EW_003, B_EF_003) — not "the value
multi-construct appears where a key:value pair belongs," and not one row. This breaks
recommendation (c): a form-parser would PASS it. Re-cast the must-fire fixture as a
vocabulary-conformance failure, not a malformation. Everything else verified: 185 rows, 18
populated, six-term set at combat-and-spellcraft.md L36 verbatim, env-grammar 2 rows / 4 keys,
FString physical_properties_array in the UE struct, single harness hit at
check_build_space.py:105, and the 4-file vril_density control.
3. L-R2-12 — WGM §8's heading reads "(Ch 66-74; **geography entries Ch 66-69 below, Ch 70-74
pending)" and the sentence itself ends "these are the first four** of the nine-realm Realm
Road." So "Dreamlike, architecturally impossible" is applied to four cosmologies, not nine.
The draft's "applied to nine cosmologies" overclaims and contradicts its own next sentence. Amend
to four; the five-realm geographic blank is the separate, already-registered half.
4. L-R2-14 — HL_0079's verbatim text is "77 agility courses canonical, one minimum per chapter,
documented-site grounded. 3 variants per course canonical: Standard, Hardcore, Puzzle. 231 courses
total." It contains no "biome-rotated against its neighbours" clause; rotation/topography is
HL_0048/HL_0049 territory. Strike or re-attribute that clause.
5. L-R2-19 — three counts are wrong. ls registries/ returns 55 dirs (not 51);
ls harness/check_*.py returns 23 (not "30+"); §2.7 specifies thirteen deterministic
teeth (BASELINE, PARTITION, PROVENANCE, EXPRESSION, TRIGGER, FLOOR, UNIVERSAL, LEGIBILITY,
NODEADEND, COMPOUND, CURVE, CARE, REVEAL), not twelve. The three runtime tests and six rubrics are
correct, and both zeros re-verified true. DESIGNED-UNBUILT stands.
6. L-R2-2 — "no declared scope" is overstated. docs/spine/CH_69.md L224 declares it in prose:
"by canon weapons are unresponsive in the Architect encounter", and L11 "weapons are
unresponsive here" — which already answers the draft's own "Apep, three beats earlier?"
sub-question. Re-state the gap as: extent is prose-declared and UNTYPED; actor-scope (Law 4)
and restoration (Law 6) are genuinely absent. The three-location claim verifies (L11, L82, and
CH69_B08's objective "the weapons go quiet in her hands").
7. L-R2-9 — the strict-chain claim is slightly over-precise. CH_69 B01→B11 is single-threaded,
but CH69_B08 and CH69_B10 each read one additional ws.* key alongside the prior beat's out
key. "Each beat's in: consumes exactly the prior beat's single out: key" should read "consumes
the prior beat's single out: key, two beats adding a worldstate read." The finding is unaffected.
8. L-R2-4 — NATURAL_VOICE_DOCTRINE.md is 57 lines, and the control counts are clue 6
(not 4), puzzle 6 (not 5), person 6 (not 5). The load-bearing zeros (tutorial 0, teach 0,
instruct 0) verify, so the control still demonstrates matchability and the finding stands.
T0_Inscription_Spine [ACTIVE v0.1] confirmed live at 98 rows.
---
1. L-R2-7 (the membrane-pocket extent_class) — correctly routed, and **stronger than the draft
argues**. The draft treats §2.3's eleven-row roster as closed; but §2.3's own confirmation block
says "The eleven-row closed roster … stand[s]" while §2.5 simultaneously proposes "a twelfth row
for the Personal Dimension." The roster is already not literally closed, which weakens the
confirmation the finding is contesting. One collision to fix: the draft's proposed id `RLM_0011
pocket_template` clashes with §2.5's proposed twelfth row — renumber before the mint.
2. L-R2-13's residue — with the compose-ban struck and R2.3's monotonic de-escalation already
ruled, does consequence_class still earn a column? It is cheap and grounded in real canon
(CH_69 L82), but it is now a typing convenience rather than a collision fix. Director call.
3. L-R2-15's worklist arithmetic — "seven realms need a threshold site typed or authored" is
imprecise and will mis-size Wave 3. Under clause (a) alone, **7 of 9 already own a threshold+low
site. The real defects are: 2 realms with none (Ch 68, Ch 73), 3 whose qualifying site is
late** (Ch 69 hall_of_two_truths, Ch 71 xibalba_climb_out_road, Ch 74 high_bough_door_home),
and Ch 67 whose qualifying site is not the arrival (breaking_and_shore). Set the number
before costing the pass.
4. A THIRD schema disagreement the draft did not catch. It flags terrain (WGM §14.2 vs spine
§3b) and vril_density (§14.5 vs spine). There is also ambient_threat: WGM §14.4 rules five
values {safe, low, mid, high, lethal}; SPINE_SCHEMA_v3.md L56 rules four `{none, low,
moderate, high}. Since T-RLM-LEGIBILITY and the proposed threshold_site` both key on
ambient_threat, the tooth cannot be written until one home wins. Add to the Wave-0 vocabulary
reconciliation at REALM_DESIGN_PROGRAM.md §5.1 alongside the vril_density item already there.
5. REALM_DESIGN_PROGRAM.md §1.3 staleness — confirmed real. §1.3's "There is no inside-the-rift
canon at all" is superseded by T1_Vril [ACTIVE v2.2] §1.1's D-2 ruling of 2026-07-27. Zero-token
director edit; not a lens deliverable.
---
Run over every care-bearing recommendation in both directions.
protected: false). Bothshare a signature: a care/difficulty instinct converted into a *schema constraint*, which is the
most expensive form of timidity because it hard-codes the conservative option into the machine
where no playtest can reverse it.
is correct ordering, independently confirmed by DESIGN_GAP_REGISTER.md L459-460 ("L-R3's
per-realm care tiering is the gating deliverable for every later art wave").
it mirrors §2.6's already-ratified compounding budget, and the REPEAT RULE (each kind used ≥2× or
dropped) *increases* the amount of impossible space the player actually meets. Keep.
boss, or routes away from the R-17+ bar. L-R2-3 in particular pushes the opposite way — it makes
realm capability-SUPPRESSION expressible so realms can legitimately close a build's usual answer,
which is the bolder reading and the correct one.
---
GO WITH FIXES. Land after: striking the §1 control row and correcting the register baseline to
464 lines; downgrading L-R2-13 to a NOTE with the compose-ban removed and the citation re-pointed to
docs/FUN_REBUILD_PLAN.md R2; striking L-R2-15 clause (b) and keeping clause (a); and applying the
eight amendments in §3. §8's schema handoff is then consumable by the Wave-2 mint as written, minus
documented_at's two struck constraints and with the RLM_0011 id renumbered.
The audit's core argument — that the machine layer cannot distinguish a realm from a place, and that
the charter's §2.3 concept under-specifies against the corpus it must describe — survives refutation
intact and is well evidenced.