CRITIC_animal-crossing-cadence-collection.md

pipelines/CRITIC_animal-crossing-cadence-collection.md

CRITIC PASS — refutation of LENS_animal-crossing-cadence-collection_DRAFT.md

Fresh-context adversarial critic, 2026-07-28. Read-only on C:\dev\humanity-forgotten and

C:\dev\Humanity\Humanity. Every citation in the draft was opened; every zero was re-run with an

independent positive control; every MISSING was re-attacked as a possible DESIGNED-UNBUILT.

Verdict key: SURVIVES (finding + class + citations hold) · AMENDED (substance holds, a

citation/class/scope is corrected) · KILLED (finding fails) · CONTESTED (director call).

---

Headline assessment

This is a strong draft. The re-verification rate on load-bearing numbers is unusually high: the

scene_type 11-value enum, the festival participation split (11/4/1), the window_kind split

(6/5/3/1/1), the 4-region festival distribution (7/5/2/2), creature roster 150 rows with

dex_completion_class and collect_verb at exactly 1/150, trade_8_tameable 95/150 split

36/31/16/12, flora 25×12, mineral 19×11, reputation ladder 2 rows, voice registry 10 rows,

minigame registry 2×19, gather nodes 24, region index 76 — **all counted independently and all

exact.** The headline correction (the museum EXISTS as ACTIVE T3 canon) is correct and is the most

valuable thing in the document.

Three real defects were found, all citation-integrity rather than substance, and one of them is

propagating from the wave spec itself and should be corrected upstream.

---

Per-finding verdicts

AC-1 — scene_type cannot express participation — SURVIVES (class confirmed MISSING)

Verified. harness/check_spine_graph.py SCENE_TYPES holds exactly the 11 values quoted, in the

order quoted, with the source comment "§5.5's scene_type set (from T0_Scene_Spec_Registry + §4

roster)". docs/SPINE_SCHEMA_v3.md L83 independently recites them and says "Adopt as-is; do not

re-invent" — quoted accurately.

The registry side verified: player_participation = opt-in-participate ×11 / observe ×4 / assist

×1 across 16 rows. FST_0008's banten-tegeh verb is real and quoted accurately ("carry it balanced

on the head and stoop through the inner temple gate without spilling"). FST_0004 does seat WPN_033

and ground the CH_03 PARRY verb, verbatim from gameplay_hooks.

The "never spectacle" attribution to FST_0001 is exact: care_note reads "sacrificial weight,

never spectacle; §17.1 collective protection binds". The self-refutation (a) therefore lands.

Class is right: this is a schema *absence*, not a designed-unbuilt — no doc anywhere proposes a

participation scene class. Recommendation is reuse-first (one enum value, three existing homes).

The contested note about riding one commit with the gate change correctly invokes the

never-write-a-gate-enrolled-file-mid-check lesson.

AC-2 — no festival window-resolution policy — SURVIVES (class confirmed MISSING)

Verified end to end. VOL2 L1034 (LW-3, "the deterministic 210-day / 10-concurrent-week

combinatorial calendar", co-req "the cited 55 day-name + tika good-day data list"), L1038 (LW-12

keys time_of_day, weather_state, pawukon_day, active_festival), L1001 (LW-5 "map Pawukon

good-days → festival windows"), L1087 (W-LW3 exit criteria) all read as cited. The density-curve

ruling is at L205-207 not L202-205 — the draft cites "§A.1 L202-205", and L202-204 is the framing

sentence with the verbatim Josh block at L205-207. **De minimis; the block is inside the cited

span's immediate context and the quote "≈100-200h main" is exact.** Not scored as a defect.

The window-kind arithmetic in self-refutation (b) checks: 6 calendar_annual + 1 lunar + 3

pawukon_cycle = exactly the 10 non-event_triggered rows. FST_0001 and FST_0011's "SHAPE only /

lands with LW-3" deferrals are verbatim from window_spec.

I re-ran the absence independently: no chapter-duration or in-game-day-budget doctrine exists in

the plan corpus, and no festival row carries a resolution rule. The zero is real and the positive

control (the same corpus returns the density curve and the LW rows) is sound.

AC-3 — museum is ACTIVE canon with no read surface, no build item, ~1% data — SURVIVES

The T3 citations are the most heavily checked in the draft and every one is exact:

T3_Creatures_Tameable [ACTIVE v1.0] L98-100 (TAME/OBSERVE/DANGEROUS), L113 ("seven non-combat

observations per species is the canonical completionist benchmark, following the Ch 02 Komodo

observation chain template"), L170 (rare variants across 420-plus), L210-212 (the three reward

tiers, quoted verbatim including "the chapter Ecological Witness title and permanent Intuition

plus-2"), and L217 — **"Completion does not grant combat advantages — the tier rewards knowledge

and attention"** — verbatim. The L9-vs-L78 count divergence (222 vs 420-plus, same ACTIVE doc) is

real and is a genuine living-source currency defect.

Data claim verified by direct count: 1/150 on both dex_completion_class (OBSERVATION_CHAIN) and

collect_verb (DOCUMENT); 95/150 on trade_8_tameable.

Surface claim re-attacked independently. I ran a repo-wide bestiary|codex sweep over docs/**

(212 hits across 51 files — the positive control is emphatic, the search matches). The only

build-plan rows that touch it are VOL2 U11.25 (photo / scan / codex-capture, the capture half)

and U11.20's screens shell, which enumerates "book/journal/Grimoire/map/ability" and not a

bestiary. No read surface is queued anywhere. The claim holds against a much wider search than the

draft ran.

Class WEAKER is defensible and the draft's self-refutation (b) states the reason correctly: the

design is designed-unbuilt, but the *planning* gap (ACTIVE canon with no line item) is the finding.

I note one nuance the draft did not: docs/proposals/systems/creature-system.md L281 declares

"Completionist/dex (DOCUMENT-all) is the universal spine that covers EVERY class at once …

Everything earns a Beastiary/codex entry even when nothing is captured." That strengthens AC-3

(the dex spine is even more load-bearing than cited) and does not refute it — it is proposal-tier

and still has no read surface. Recommend the draft add it as supporting citation.

AC-4 — documentation ladder is fauna-only — SURVIVES (with one strengthening amendment)

Flora 25 rows × 12 columns and mineral 19 rows × 11 columns confirmed exactly, headers as quoted,

and neither carries any discovery/documentation/completion field. I re-ran the header search with

an independent pattern — no observ|document|dex|complet|discover in either header; rarity_class

and care_note both present, so the read is sound.

The language wing: CVD L1217 does contain "Language fragments accumulate to roughly fifteen

documented languages by arc completion" — verbatim and at the cited line. `T0_Language_Script_

Registry [ACTIVE v0.1]` is 84 rows × 19 columns.

Amendment (strengthens the finding, does not refute it): the draft says the language registry

is "an accumulation with no completion tier and no payout". It does carry decipherment_status

(FULLY_DECIPHERED ×56, ORAL_NOT_SCRIPT ×13, PHONETIC_ONLY_NO_SEMANTICS ×9, FULLY_UNDECIPHERED ×3,

+3 singletons) and mastery_tier_threshold (82/84 populated). Neither is a *player* completion

state — decipherment_status is real-world scholarly status and mastery_tier_threshold is a

gating threshold. So the finding is right, but the precise claim should be "carries scholarly

status and a gate threshold, but no player-side documentation state and no completion payout",

which is sharper and pre-empts a later reader's false refutation.

The devil's-advocate handling of (c) is correct and correctly calibrated: withholding the ladder

from flora on care grounds would be care-inflation. No timidity defect.

AC-5 — non-combat progression viability unmeasured — SURVIVES (see contested note)

Substance verified below; see the verification block for the boss-registry counts. Class MISSING is

correctly scoped: the draft explicitly hands has_nonkill_path population back to FR-038 and the

BG3 lens and claims only the arc-level union, which is genuinely unclaimed. That is the correct

discipline and it is the reason this finding is not a duplicate.

The devil's-advocate pass is present and lands in the right direction — no violence reduced, PACIFY

integrity cost untouched, gladiator arena untouched, and the Ch-2 CAVE_BEAR forced-run cited as

proof that an unkillable encounter is still a combat beat. No care-inflation defect. If

anything this finding is the correct anti-timid framing: it asks for *measurement*, not mercy.

AC-6 — return-is-paid-in-change has no pre-Vimana surface — AMENDED (citation defect)

The substance survives; one quotation is misattributed and must be fixed before landing.

The draft quotes, as verbatim from T1_World_Geography_Master [ACTIVE v2.3] L766: "the five-

world-state revisit mechanic, which continues to govern the previously visited ground each node

still reflects on".

WGM L766 does not contain that text. L766 is the *Revisit-band rescope* paragraph ("several nodes

are re-tasked from re-walking seen ground to reaching ground the outbound arc skipped, with the

world-state reflection function preserved as the framing layer"). The five-world-state mechanic is

real but lives at WGM L999: "the five-world-state revisit mechanic governs previously visited

ground". The draft's rendering is a paraphrase presented inside quotation marks at the wrong line.

Amendment: re-cite as WGM L999 (mechanic) + L766 (revisit-band rescope framing), and drop the

quotation marks or quote L999 exactly. The finding itself — that no worldstate key represents

elapsed-time-since-visit, and that the Ch 59-65 band is the only return surface — is unaffected and

the LW-12 key list at VOL2 L1038 is verbatim correct.

AC-12 — minigame layer is DESIGNED-UNBUILT — AMENDED (citation defect, propagating upstream)

The verdict (DESIGNED-UNBUILT, no finding minted) is correct and the restraint is right. But the

CVD citation is wrong in three ways at once and it is the one error in this draft most likely to

poison a later agent, because it is quoted as ratified canon text.

The draft says CVD §17.12 ratifies, at L1148, "max upgrades on every craft (the Vehicle/Mount

and base Domains buildables) and top score on every minigame".

"New-systems reconciliation (AAA re-architecture, ratified 2026-07-12). The L1 monster-breeding

husbandry loop operates on the 400-plus ordinary tameable roster only…" — nothing about

minigames.

returns 4 CVD hits (L51, L248, L981, L1148), so the search matches and the corpus is reachable.

minigame**, fully upgrade every vehicle and base, complete every side quest."

Amendment: re-cite as CVD L51 (Layer 2 completionist), not §17.12/L1148. The substance —

minigame top-score completion is ratified CVD canon — is TRUE, so nothing downstream changes.

Escalation for the director: the wave runbook's own divergence guard carries the same

mis-citation ("CVD §17.12 ratifies minigame top-score completion"). The draft almost certainly

inherited it. Every lens in this wave that touched the no-scoreboard guard may be carrying it.

Worth a one-line correction at the wave-spec level so the whole roster re-anchors on L51.

The rest of AC-12 verifies: minigame-galaxy.md §2.1 Tier-A quote is verbatim at L28 (tavern

knucklebones, cup-and-ball, coin-pitch, drinking-song call-and-response … "pure L1.5 immersion"),

the Games Master completion meta is at L52-55, and VOL2 L867 MG-1 UHumanityScoredInteractionSubsystem

reads as cited. T0_Minigame_Registry is 2 rows × 19 columns. DESIGNED-UNBUILT is the right class.

AC-7 — faucet-sink discipline has no spec — AMENDED (positive control is wrong) + CONTESTED (class)

The substance holds: DESIGN_GAP_REGISTER L208 does read "/ auction house (faucet-sink discipline)"

as a DIVERGENT-BY-DESIGN rationale, and no faucet-sink spec exists anywhere in the repo. VOL2

L1059's Opt-3 con is verbatim ("leaves gathered materials + drops with no sink; the slice reads

incomplete"). The recommendation is properly reuse-first — base-building modules, the §9 2nd/3rd

loadout sets (verified at COMBAT_PROGRAM_ADDENDUM L156), T0_Imprint_Affix and T0_Rune_Registry

(both exist on disk), vehicle upgrades — and is explicitly bounded "no timers, no upkeep, no

repossession", which respects the anti-grind guard.

Amendment — the positive control is factually wrong and must be corrected. The draft states

"faucet returns exactly 3 hits across docs/". I re-ran it: 17 total, of which 8 are

pre-existing repo hits** (the other 9 are the lens draft itself and WAVE_SPECS.md). The draft

found 3 of 8 and missed 5, all in a third sense the draft did not name — the anti-grind sense:

CANON_CORRECTION_CHANGESET.md L388 and canon_correction/cc4_tier_combat.md L77 both read "hold

the anti-grind guarantee ('innocent RNG,' no farm faucet) as the binding constraint";

combat-and-spellcraft.md L89/L250; SECTION5_NUMERICALS_BRIEF.md L44.

This does not refute the finding — none of the five is a faucet-sink *spec*, so the zero is

still real. But under the wave's own positive-controlled-zeros rule an undercounted control is

precisely the failure the rule exists to catch, and it must be corrected before landing. The

correction also strengthens the recommendation: "no farm faucet" is already a *binding

constraint* on record, so the sink ladder must be written to compose with it — cite it.

CONTESTED (class, for the director). The draft's own header assigns the owner as "spec-doc

(rides the D-ECONOMY brief)" while self-refutation (a) argues the sink question is *separate* from

the blocked brief. Those two statements are in tension. If the sink structure rides D-ECONOMY, this

is arguably DESIGNED-UNBUILT/BLOCKED-ON-BRIEF rather than WEAKER. I lean to the draft's reading

(the medium ruling genuinely does not answer "what absorbs it"), but the director should settle it

so the register does not carry a class the economy brief later contradicts.

AC-8 — integrity skin vs player decoration collide — AMENDED (citation defect ×1, overstated ×1)

Substance survives. The owning doc really is an empty stub: register entry #22 DOC-STUB reads

verbatim "Four T1 docs are byte-level EMPTY stubs, not drafts: Audio_Spec, UI_UX_Spec,

Personal_Dimension_Spec, Tier_B_DLC_Catalog" (L323-324), and all three files measure 3 lines on

disk (the draft says 2 — 2 content lines plus a trailing newline; de minimis, the stub claim is

right). U11.25 photo mode confirmed at VOL2 L752, so the "photo mode is not a zero" note is right.

Amendment 1 (citation). The draft attributes to base-building.md §5.1 the text "L1

Truly-Good → verdant-light substrate; L3 Drifting → neutral-grey balance; L4-L5 Evil →

shadow-substrate variants". §5.1 is titled "Canon (do not contradict)" and is entirely about the

Dimension Pocket (Ether-Crown, Tier-6, creature housing, growth-rate-free garden) — it does not

contain that text. The real homes are the comparison table at L24 ("YES — canon (verdant-light

L1 / neutral-grey L3 / shadow L4-L5)") and §5.3 at L101 ("verdant paradise, ashen

shadow-realm, or the razor-balance grey"). The §5.3 quote the draft uses for the AAA hook — "a

screenshot tells your whole playthrough's story" — is verbatim correct at L101. §4.3's free-form

module placement quote is verbatim correct at L77.

Amendment 2 (overstatement). "Nobody has split them" is too strong. The L24 table already

draws a partial split — for the culture's own home it rules integrity-visible **"No (it is the

*culture's* home — care firewall); only your added modules read your path."** That is a split by

*location*, not the substrate-vs-objects split within one surface that AC-8 proposes, so the

finding survives — but it should acknowledge L24 and position the recommendation as *generalising

an existing partial rule* rather than inventing the first one. That is a stronger reuse-first

posture, not a weaker one.

Amendment 3 (dedupe transparency). The register's claim board carries 14b — "Personal

Dimension (split from 14) | the five existing sandbox lenses | DESIGNED-UNBUILT routing (never

MISSING; T1 spec stub = register #22/23)" (L410). AC-8 is not a duplicate of 14b (14b is routing

status; AC-8 is a design-collision ruling), but the draft should cite it so a later reader does not

mistake the overlap for a missed dedupe.

AC-9 — no warmth axis in the voice doctrine — SURVIVES (positive control independently exact)

docs/NATURAL_VOICE_DOCTRINE.md is 57 lines (draft says 56 — de minimis). §2's cultural register

is at L23-27 including "an elder's cadence, a healer's directness, a broker's oil — drawn from the

persona system (the 13-field personas) + T0_Voice_Registry". §5's two critic questions at L52 are

verbatim: "does a person say this?" and "does this carry the culture?". §4 at L42-45 is verbatim

including "the strictly-no-kill path's special rewards" and "They are PRIZES for the players who

want them — never walls."

I re-ran the draft's own positive control (warm|tone|cozy|gentle|register) independently: **2

hits, L23 and L50, both register lines.** Exactly as stated. This is the most honest control in the

document. T0_Voice_Registry confirmed at 10 rows.

Devil's-advocate check: self-refutation (c) rejects the "warmth defangs a rated-R arc" objection

with "the arc's emotional payload *depends* on the ordinary mornings being warm enough to lose."

That is correctly calibrated against timidity — no care-inflation defect.

AC-10 — festival coverage 4/76 with a NULL gate — AMENDED (one line number)

Verified precisely. Region distribution is exactly flores_island ×7, bali ×5, sumatra_java ×2,

angkor_khmer ×2 = 16 rows; chapter_range spans CH_02–CH_05 only; T0_Region_Index is 76 rows.

FR-039's gate column reads verbatim `NONE(no per-chapter festival or ambient coverage assertion

exists; rank 8 authors the cue table). daily` really is 1 of 16 (FST_0010). Class DESIGNED-UNBUILT

is correct and correctly reasoned. Recommendation lands on the existing gate suite — reuse-first.

Amendment. The drift lesson is cited as OPEN_DECISIONS_BRIEF_2026-07-22 L489. The file is

at docs/OPEN_DECISIONS_BRIEF_2026-07-22.md (1162 lines) and L489 is about the 90 seeds and quest

types, not enforcement. The lesson the draft wants is at L505: Option F, "C with teeth" — "my

recommendation is one unenforced sentence away from being Option E's nightmare". Same document,

same argument thread, wrong line. Re-cite L505.

AC-11 — no low-stakes acquaintance tier — AMENDED (host doc is an orphan) + thinnest finding

Verified: T0_Reputation_Ladder_Registry is 2 rows; RUNTIME_GENERATIVE_LAYER.md L64 is verbatim

("NPC ambient dialogue (barks, crowd chatter, vendor lines) | Ambient NPC | Line content within

topic + mood | Persona card; region; worldstate facts flagged safe-to-reference; integrity band").

I re-ran the absence: recognition_state returns zero repo-wide while 13-field persona

returns 8+ real hits (FR-035, factory_contract.json L584, NATURAL_VOICE_DOCTRINE L27,

PRE_5090_BUILD_PLAN L667, VOL2 L1389, ROADMAP L144), so the control is sound and the tier is

genuinely absent. Dedupe against register #2 (GTA daily schedules, L34) is correctly distinguished.

Amendment. The recommendation should cite FR-035 (FACTORY_CONTRACT.md L149 — "Flesh a

complete persona on every entity this chapter surfaces — the 13-field persona is what the

natural-voice and living-world lanes read from") as the landing obligation, not only LW-8/LW-9.

Caveat worth recording: the persona schema's home,

docs/proposals/ENTITY_PERSONA_SYSTEM.md, is flagged [ORPHAN] in docs/WIRING_AUDIT.md (L11,

L120), so "rides an existing card" is true of the schema but the schema itself has no live consumer.

Thinness note (not a kill). This is the weakest finding in the set. The generative layer already

grants ambient NPCs "worldstate facts flagged safe-to-reference" plus a persona card, which is

arguably enough to produce a villager who references what you did. The draft's distinction — a

person who *recognises you* with no ledger consequence — is real but narrow. MEDIUM-LOW priority is

the honest call and the draft assigns exactly that. It survives on its own honesty.

AC-13 — no-fail play is COVERED — AMENDED (path; the tier claim holds)

The Wellspring quote is verbatim — "while it holds you close the world moves with you; no trial can

strand you here. This is the arc for the traveller who came for the journey and the wonder" — and

"The dial is ONLY vril density; NEVER HP / damage / resistance inflation" is verbatim at L12-13.

Amendment. The draft cites it as bare DIFFICULTY_SYSTEM.md; the file is at

docs/proposals/DIFFICULTY_SYSTEM.md (112 lines). I checked the tier question this raises,

because a proposal-tier source would be weak evidence for a COVERED verdict: it is not proposal

tier in substance — the title reads "Difficulty System — RULED" and the Status block at L93-95 reads

"RULED and CLOSED (Josh, 2026-07-04)". COVERED stands. harness/check_no_dead_end.py exists.

The DIVERGENT-BY-DESIGN note (posture vs tier-plus-routing-rule) is a sound pre-registration, and

handing the death/defeat axis to the hades lens rather than claiming it is correct discipline.

AC-14 — real-clock gating / dailies / FOMO as DIVERGENT-BY-DESIGN — SURVIVES

Correct and valuable. Pre-registering the three rejected mechanisms is exactly what the register is

for, the rationales are sound (single-player-first, offline, anti-grind, no-dead-end, and the

AI-QA-loop verifiability point is the sharpest of them), and it forecloses a future lens

re-litigating AC's retention engine as a gap. No citation to check; nothing to amend.

---

The 8 refuted-and-dropped candidates — spot-checked

I re-verified the three that carry the most weight.

quoted (see AC-3). This is the draft's best work.

verbatim "registry creation deferred until consumer demand demonstrates need." The

markdown-escape trap the draft reports is real: the CVD writes

T0\_Journal\_Entry\_Template\_Registry, so a naive grep returns zero. **The draft caught a live

instance of the documented a-search-that-cannot-match-reports-zero failure and recorded it

honestly. That is the single best methodological moment in the document** and should survive into

the landed file verbatim.

respawn_class is fully populated with the seasonal vocabulary described. Correctly killed.

The remaining five (#3 journal, #4 has_nonkill_path, #5 photo mode, #6 voice rows, #8 economy)

all check out against citations verified elsewhere in this pass.

---

Devil's-advocate-against-timidity pass

Run over every recommendation. **No care-inflation defects found — this draft is calibrated

correctly, and in two places it explicitly pushes back against the timid option:**

right: the fauna ladder already pays cultural profiles *with attribution*, and care_note is

already a live column on both flora and mineral registries.

flattening failure.

content softened, no violence reduced, PACIFY's integrity cost untouched, the gladiator arena

untouched, and it cites the Ch-2 CAVE_BEAR forced-run as proof that an unwinnable encounter is

still a combat beat. Verified: PACIFY's ruled definition at T99_Translation_Combat L502-503 is

verbatim as quoted, and win_condition/primary_verb distributions are exact to the row.

The opposite failure — defanging by *omission* — does not appear either. Nothing in the draft

proposes removing violence, softening a fight, or protecting a monster.

---

Numbers independently re-counted (all exact unless noted)

ClaimDraftVerified
scene_type enum11 closed values11, verbatim, check_spine_graph.py
Festival player_participation11 / 4 / 1exact
Festival window_kind6/5/3/1/1exact
Festival regions7/5/2/2 across 4exact
Creature roster150 rowsexact
dex_completion_class · collect_verb1/150 eachexact
trade_8_tameable95/150 (36/31/16/12)exact
Flora · Mineral25×12 · 19×11exact
Boss rows · has_nonkill_path281 · 16exact
win_condition top values46/33/16/16/10/6exact
primary_verbstrike 64 / perceive 52 / …exact
Region index76 rowsexact
Voice · Reputation · Minigame · Gather10 · 2 · 2×19 · 24exact
append grammar count154exact (L97, L502)
Language registryACTIVE84 rows × 19 cols
Worldstate_Variables42 (WS_001-042)46 (WS_001-046) live — see below
faucet hits in docs/**38 pre-existing — see AC-7

Worldstate amendment (AC-6). The draft quotes VOL2 L1038's "42 rows (WS_001-042)" correctly,

but did not check the live registry, which now holds 46 rows (WS_001-WS_046). I inspected the

four new rows: bonus_selection_state, bonus_respec_ledger, loadout_set_state,

item_instance_ledgernone is a time / weather / calendar / festival key. So the finding is

*confirmed against live data* and is stronger than the draft argued; only the count is stale. Cite

the live registry (46 rows, none temporal), not the plan doc's 42.

---

One check the draft did not run — and it strengthens AC-3

The draft proved "no bestiary read surface" from the *plan* only. I checked the built side in

C:\dev\Humanity\Humanity\Source: the only matches for bestiary|codex|festival|minigame are

Data/HollowCodexRow.h (the Hollow containment record — a different system), `Data/

MinigameRegistryRow.h` (a data row, no subsystem), and two Hollow combat headers. **There is no

bestiary/codex UI, no festival system, and no scored-interaction subsystem in the engine.**

Data/BossEncounterRegistryRow.h carries no non-kill field either. AC-3 and AC-10 therefore hold

on both the design side and the built side; recommend the landed file add this engine-side control.

---

Landing recommendation

GO-WITH-FIXES. All 14 findings survive on substance; none is refuted; none is a re-litigation of

a pre-registered divergence; none is care-inflated. The verdict classes are right, including the two

hardest calls (AC-3 WEAKER-not-DESIGNED-UNBUILT and AC-10 DESIGNED-UNBUILT-not-MISSING), and the

dedupe discipline against BG3 #42/#4, register #2/#9/#22, and the P1-A creature lens is genuine

rather than decorative.

Land after these seven citation fixes, none of which changes a verdict:

1. AC-12 — CVD "top-score every minigame" is at L51 (Layer 2), not §17.12/L1148. *(Also

correct the wave spec, which is the likely origin — escalate.)*

2. AC-6 — the five-world-state quote is WGM L999, not L766; drop the quote marks or quote

L999 exactly.

3. AC-6Worldstate_Variables is 46 rows live, not 42; the finding gets stronger.

4. AC-7 — the faucet control is 8 pre-existing hits, not 3; add the "no farm faucet"

anti-grind sense and cite it as a binding constraint on the sink ladder.

5. AC-8 — the integrity-substrate quote is at L24 + §5.3/L101, not §5.1; soften "nobody has

split them" to acknowledge L24's partial split; cite register 14b for dedupe.

6. AC-10 — the enforcement lesson is at L505, not L489.

7. AC-13 — path is docs/proposals/DIFFICULTY_SYSTEM.md (RULED and CLOSED, so COVERED holds).

Recommended additions: the creature-system.md L281 "universal dex spine" citation to AC-3; the

engine-side absence control to AC-3/AC-10; FR-035 to AC-11; the sharper language-registry wording to

AC-4.

Director-owned items (unchanged from the draft's own list, plus one): AC-7's class; AC-3's T3

222-vs-420-plus numeric fix (ACTIVE-doc numeric string → director applies, per the

subagent-classifier lesson); AC-1 riding a single commit with its gate change; and the wave-spec

CVD §17.12 mis-citation, which may have propagated to other lenses in this wave.

Killed: 0 · Survived: 14 · Amended: 7 · Contested: 3

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