CRITIC_L-R1-mythic-cosmology-adaptation.md

supernatural/CRITIC_L-R1-mythic-cosmology-adaptation.md

CRITIC PASS — L-R1 · mythic-cosmology-adaptation

Fresh-context adversarial refutation. Read-only on both repos (nothing edited or staged). 2026-07-28.

Draft under review: scratchpad/realm_wave/L-R1-mythic-cosmology-adaptation_DRAFT.md — 19 findings.

Every citation below was opened and read. Every absence claim was re-run by me with my own

positive control. Counts were recomputed in Python (literal str.count, never shell grep) so

bracket/accent escaping could not fake a zero.

---

HEADLINE FOR THE DIRECTOR

Two findings would have caused real damage if landed as written. Both are the audit's own

top-priority items.

1. **F-10's "mis-cited 41 times" is false, and its repair would strip HL_0003 off ~30 places

where it is correct — two of them sitting directly on the §17 floor.** The count of 41 is

exact; the classification is not. Scope cut ~75%.

2. **F-01's premise is a category error and its named Wave-0 action edits a code mirror of

ratified CVD §17/§6.2 canon using DRAFT-tier data the audit itself calls polluted.**

harness/enrich.py is not lying — it is a faithful transcription of CVD §6.2. **Remedy

killed; a genuinely better finding is handed forward in its place.**

The other seventeen findings are unusually well-evidenced. I recomputed nine independent tallies

and every one matched the draft to the cell. This is a strong audit with two dangerous items in it.

---

1. F-10 — the unit-specific instruction, verified in full

(a) The 41 count — CONFIRMED EXACT

Recomputed over docs/spine/CH_66.md–CH_77.md:

Sum = 41. The draft's table is correct to the cell, all three rows. The inverse-correlation

observation (anchoring vanishes at Ch 71-74, the living-tradition realms) is confirmed by the

same numbers. No inflation in the arithmetic.

(b) The HL_0003 verbatim — CONFIRMED EXACT

registries/T0_Hard_Lines [ACTIVE v0.2]/Sheet1.csv, 123 rows, rule_id = HL_0003,

source_section = CVD §17.1, severity = canonical, applies_to_chapter_range = universal.

The draft's block quote matches the rule_text cell word for word through the closing sentence

*"The floor is anti-defamation and no-genocide-as-goal, not individual invincibility."*

No paraphrase, no truncation, no ellipsis.

The draft's reading is also right: **HL_0003 governs collective community protection and says

nothing about deity instantiation.**

(c) "Mis-cited forty-one times" — REFUTED

I pulled all 41 citations with ±400 chars of context and classified each:

Correct citations the blanket repair would have destroyed — quoted from the live spine:

• CH_67: *"never the 'hag' caricature, never racialized into an 'evil northern race,' never

Sámi-conflated / leaned on through 'Lapland witch' tropes."* HL_0003 **names Sámi noaidi in

its own rule text.** This is the most on-point HL_0003 citation on the entire road.

• CH_71 (its only HL_0003 citation): *"cruelty checked by the diminish-not-exterminate shape;

the lesser lords still spared on every path (HL_0001/HL_0003)."* This is HL_0003's

no-genocide-as-goal floor, verbatim domain. Stripping it touches the §17 floor.

• CH_67 ×3: *"the sleeping people of the north … never a threat nor a target, a protected

presence only"* / *"never a raid on a living community."* Core HL_0003 clause.

• CH_68 ×4: *"the Aboriginal sacred-tradition sound boundary respected per HL_0003"*,

*"no protected cultural sound reproduced"*, the Rainbow Serpent referenced-never-reproduced

with Ch-57 maximum-care carried. Living-tradition protection. Correct.

• CH_70 ×2: *"the honored dead of Asphodel are never a target"*, *"the egress-dead dangerous

and pitiable, never morally-evil ghouls."* Never-villain-framed. Correct.

• CH_69 — *quoted by the draft AS a mis-cite*: *"the Ma'at moral architecture honored as

Egyptian and African intellectual achievement — Karenga and Assmann — never curse-magic or a

tomb of traps per HL_0003."* This is anti-defamation of a tradition (refusing the

mummy-horror trope), i.e. precisely *"never villain-framed, defamed."* **The draft

mis-classified its own headline evidence.** Same for the parallel *"never curse-magic or

mummy-horror"* citation.

The genuine mis-cites — real, worth fixing, and all but one in CH_69:

• *"the assessor-deities (Ma'at, Anubis, Thoth, Ammit, Osiris, the forty-two) stay AUDIT-ONLY

throughout … never an NPC and never the boss (HL_0003)"*

• *"no deity instantiated per HL_0003"* · *"Ma'at and the assessor-deities audit-only per

HL_0003"* — and two more restatements in the beat/scene tables

• *"takes no realm-name or deity-name per HL_0003"* ×4 — the deity-name-on-a-mortal firewall,

which is CVD Component 7, not §17.1

• CH_68 ×1 partial: *"traditional figures are audit-only"* (the rest of that footer is

reveal-ceiling material co-cited to HL_0043)

(d) "No T0_Hard_Lines row for the deity rule" — CONFIRMED, my own positive control

Term scan over all 123 rows, all columns concatenated, case-folded:

deity       -> 2 rows (HL_0077, HL_0080 — neither governs instantiation)
deities     -> 0     pantheon -> 0     divin* -> 0
instantiat  -> 0     avatar   -> 0
god         -> 1 row (HL_0045)
revered     -> 2 rows (HL_0001, HL_0121)
CONTROL Velheim    -> 14 rows   (matcher works)
CONTROL malevolent ->  3 rows   (matcher works)

The zero is real; the absence claim survives. I also read HL_0121 and HL_0122 in full — both

match the draft's characterisation (malevolent-beings-are-legitimate-combat; honored-adversary

with the Ch 67 Louhi exemplar). Neither governs instantiation.

CVD Component 7 verbatim — CONFIRMED, quote is exact:

"Deities are equally real, portrayable, and usable as characters — a good deity may appear as

a disguised human quest-giver or a rescuer (rescue-and-re-encounter), an evil deity opposes

the player; the only firewall is that no deity-name sits on a mortal NPC and no revered or

real figure is cast as the villain."

(e) The proposed T-RLM-ANCHOR tooth would ship as a NOISY GATE

§6.2's tooth table names the must-fire fixture as *"any of the 41 HL_0003 realm citations."*

A topic-match assertion run against the Ch-67 Sámi-conflation cite or the Ch-71

no-extermination cite fires on correct canon. That is the repo's own *"a noisy gate

CONCEALS"* defect being designed in at mint time: ~30 false positives burying ~10 real ones.

Fixture must be a named CH_69 audit-only citation, not "any."

F-10 — AMENDED VERDICT

• Defect SURVIVES. No deity-presence hard-line row exists; CH_69 anchors ~10

deity-instantiation claims to a rule that does not say that; the constitutional rule lives in

CVD Component 7, is permissive, and has exactly two firewalls.

• Headline KILLED. Correct statement: *"~10 deity-instantiation mis-cites, concentrated in

CH_69, out of 41 HL_0003 realm-road citations of which the majority are correct."*

• Recommendation #2 REJECTED as written. *"Re-anchor the 41 realm-road citations"* would

strip the §17-floor anchors off Ch 67 and Ch 71. Replace with: *"Re-anchor the ~10 identified

deity-instantiation citations, enumerated by beat id; ADD the new row alongside HL_0003 where

a claim is genuinely dual-domain; leave every community-protection citation untouched."*

• Recommendation #1 SURVIVES — minting the row by transcription from CVD Component 7 into

the existing T0_Hard_Lines registry is the right reuse-first move.

• Priority HOLDS at HIGHEST. The diagnosis is real and Wave-2-blocking. It was the repair

scope that was wrong.

---

2. F-01 — KILLED as written. Category error + a canon-touching Wave-0 action

(a) The cited literals — CONFIRMED EXACT

harness/enrich.py:77-78 (line numbers exact):

HIGH_CARE = {29, 42, 57}                    # Sámi / Haiti / Australia
ELEVATED_CARE = {11, 12, 13, 47, 48, 50}    # Congo / West African / South Africa / Mesoamerica / Amazon

care_tier() at :233, called from goal_anchors() at :258. All three line refs correct.

No realm chapter appears in either set; all twelve return standard. Mechanically true.

(b) But enrich.py is a faithful transcription of RATIFIED CVD canon

I read the CVD. _source/01_Tier_1_Foundation/T1_CVD_Creative_Vision_Document [ACTIVE v1.4].md

§6.2 "Application by chapter tier" enumerates the tiers itself:

"Maximum depth tier … Ch 29 Sámi … Ch 42 Haiti / Vodou … Ch 57 Australia / Aboriginal

Elevated care tier … Ch 11 Congo, Ch 12 West African, Ch 13 South Africa, Ch 47-48

Mesoamerica, Ch 50 Amazon

Standard care tier — all other chapters."

The two Python literals are a cell-for-cell mirror of that list, and §6.2 explicitly assigns

standard to every other chapter. The pack is not lying. It is quoting the constitution.

(c) The Ch 69 row of the draft's contradiction table is a conflation of two different fields

docs/spine/CH_69.md resolves this itself, in text I read:

"The chapter's cultural_depiction_tier is standard, but its dominant care register is the

ELEVATED §17.1 duty the source header names … FORK_CH69_A and the boss/beat care register are

typed care_band elevated, not open."

CH_70.md states the same distinction and cites CH_69 as its precedent. So the corpus

consistently distinguishes three things:

The draft treats these as one field with a contradiction. They are three fields with three

authorities. Ch 69 returning standard is exactly what the live spine says it should return.

(d) The Wave-0 action is canon-touching, self-contradictory, and care-inflating

The draft's step 1: *"extend the two existing literal sets in harness/enrich.py … seeded from

the live T0_Boss_Encounter_Registry.care_tier values … (zero-token, Wave 0)."* Three

independent objections:

1. It edits a mirror of CVD-enumerated chapter canon. Chapter-level care-tier assignment is

CVD §6.2 text. CLAUDE.md's cardinal rule reserves chapter/numerical/§17 canon to Josh and

says the CVD text is never auto-corrected. Changing the mirror without ruling on §6.2 mints

exactly the doc-vs-code drift F-14 (correctly) complains about — one document further down.

2. It seeds ratified-tier code from DRAFT-tier data the same audit calls polluted. F-02

proves T0_Boss_Encounter_Registry.care_tier carries 15 values including full sentences.

F-01 proposes using it as the source of truth. The two findings contradict each other.

3. It is care-tier INFLATION applied mechanically — the move CLAUDE.md explicitly names a

defect ("care-tier INFLATION is a defect; the timid option is usually the wrong one").

The draft ran a devil's-advocate-against-timidity pass on F-12 and **did not run one here,

where it was needed just as much.** Raising nine chapters to elevated/maximum adds Stage-2

review + author-in-loop load to the whole Realm Road on no ruling.

Labelling all of this "zero-token" makes it read as a safe mechanical edit. It is not.

(e) What genuinely survives — a better finding than the one written

**Three care vocabularies with no declared relationship, and a Wave-2 pass that will read one

and believe it read another.** That IS decision-grade, it IS Wave-2-blocking, and it needs no

canon edit to state. Plus one real substantive question worth surfacing:

CVD §6.2 places Ch 47-48 Mesoamerica in the elevated tier but not Ch 71 Xibalba, the

game's flagship living-Maya content (Popol Vuh, K'iche'). §6.2 was written for the earthly arc

and predates the Realm Road cascade.

Correct handling: a decision brief to L-R3 / the director under the standing protocol

(alternatives + deep-dive + adversarial + devil's-advocate-against-over-conservatism +

recommendation), not a literal edit in Wave 0. Reuse-first fix that needs no canon call:

have the Wave-2 T0_Realm_Ruleset_Registry carry care_band and make enrich.py read it

for chapters 66-77 — which is the draft's own step 2, and is sound. **Step 2 survives; step 1

is withdrawn.**

• Verdict reclassified: WEAKER → spec-doc coordination gap + one surfaced canon question.
• Priority reclassified: HIGHEST → MEDIUM (nothing silently breaks; the schema hook is

Wave 2 either way).

• SCHEMA HOOK survives unchanged and is the right landing.

---

3. Findings verified and SURVIVING (independently recomputed)

Pre-registered-divergence check: I read docs/DESIGN_GAP_REGISTER.md (8 DIVERGENT-BY-DESIGN

entries, all FF7R / WoW / GTA / MHW / Elden — combat, UI, economy, navigation). Term scan:

deity 0, care_tier 0, literacy 0, Yggdrasil 0, Duat 0, Xibalba 0, theme 0.

No finding in this audit re-litigates a pre-registered divergence. The markerless-navigation

and no-scoreboard guards named in the runbook are untouched.

Reuse-first check: every recommendation names an existing mechanism — the hard-lines

registry, the layers: tag, check_registry_extensions.py, the HL_0049 topography

comparator, node_kind = mythic_realm, T0_Theme_Registry, the zone catalog's single

sanctioned writer, check_no_dead_end.py. One brand-new tooth family, justified. Passes.

---

4. Amendments (finding survives, detail corrected)

• F-04 — the prose says *"Four beats on the road carry no L1 at all"* but its own enumeration

(CH_66 ×2 L3, CH_67 ×1 L3 + ×1 L2+L3, CH_74 ×1 L2+L3) sums to five. I recount

five. The enumeration is right; the word is wrong. Matters because it is the T-RLM-LITERACY

fixture population.

• F-16 — *"cultural_substrate is empty on all twelve rows"* is false: FAMILIAR_BOND

carries *"varies per species region per Pillar 10 regional combat identity."* One of twelve,

not zero. The substance (the column is essentially unexercised; no region/realm

theme_type; hard_line_relevance populated on all 12) is confirmed.

• F-18 — *"zero occurrences of any of the nine realm names"* is false: Hades appears

once in PRESENTATION_DOCTRINE.md. All other eight are genuinely 0, and realm ×3,

art direction ×0, ceiling-setting ×0, palette ×1, silhouette ×2 are all exact. The

DR-2 COVERED half is confirmed verbatim at COMPARATOR_LENS_PROGRAM.md:144. Finding holds;

drop the absolute.

• F-11 — flag the framing on treatment 4. Calling instantiated-ally Arawn *"the sharpest

gap"* resting on *"the thinnest anchoring"* invites a Wave-2 author to restrict him. The

draft does say it is "legal and good," but the fix must be stated affirmatively:

anchor Arawn to the new Component-7 row; never reduce his instantiation. CVD Component 7

explicitly blesses a good deity as host/friend.

• F-02 — the recommendation to move sentence-values into a new care_notes column touches

a live registry and shifts docs/fidelity_baseline.json. Not "free." Must carry the

--emit-baseline refresh in the same commit or the gates go red.

---

5. Devil's-advocate-against-timidity pass

F-12 — the audit's one deliberate care call. I weighed it both directions and ENDORSE it.

• *Direction A — loosen Ch 69.* CVD Component 7 permits instantiation with exactly two

firewalls; "care means weight and authenticity, never omission" pushes toward letting Osiris

appear. Rejected. Ch 69's reason is craft, not flinch: the Architect's Final Negotiation

is a fifty-two-chapter terminus and a second judge in the room demotes him at his own payoff.

And the chapter demonstrably does not defang — Apep is genuinely fought and unmade by the

tradition's own rite (HL_0121, correctly cited, 11 times). Loosening buys nothing and costs

the terminus. The R-17+ bar is met here, not dodged.

• *Direction B — declare it and block its propagation.* **Correct, and it is the

anti-timidity move**, not a conservative one: it stops the road's strictest treatment from

being copied by default onto living Hindu, Buddhist and Jain deities at Ch 72, where the CVD's

permission is identical, the protagonist canonically echoes the Indra-role, and Indra is

already 25 mentions in the prose.

• **rationale_kind enum{design|care|both} plus "a realm whose deity treatments are all care

is the timidity signature" is the best anti-timidity instrument in this audit** — it makes

over-conservatism machine-detectable instead of a vibe. Endorse and elevate; it belongs in the

Wave-2 mint, not as an optional extra.

• T-RLM-DEITY-FLOOR encoding exactly two firewalls is right, and the draft says so

explicitly ("encoding exactly two of them is what keeps the tooth from becoming a timidity

engine"). Hold that line in the mint — do not let a third firewall accrete.

Timidity flags raised against the draft:

1. F-01 is the care-inflated finding (§2d above) — a mechanical tier raise across nine

chapters, presented as zero-token, with no devil's-advocate pass run on it. The audit applied

the discipline to F-12 and skipped it exactly where it also applied.

2. F-11's Arawn framing (§4 above) — reads as an invitation to tighten a treatment the CVD

blesses.

3. Otherwise clean. F-06 (declare, do not rename), F-08 (decompose, do not soften), F-12

(do not loosen Ch 69) are all correctly calibrated to the R-17+ bar.

---

6. Contested — for the director

1. The enrich.py / CVD §6.2 care question (F-01 residual). Should Ch 71 Xibalba (living

K'iche' Maya, Popol Vuh) sit in §6.2's elevated tier when Ch 47-48 Mesoamerica does? §6.2 is

CVD-enumerated chapter text. **My position: surface as a decision brief to L-R3, do not edit

the literals in Wave 0.** I did not decide it and I edited nothing.

2. Authority for F-10's repair. ~10 citation edits across CH_68/CH_69 is a canon-bearing

citation change: director-level apply, fresh critic, full gate suite. The draft says this

correctly — but it says it about 41 edits. **The enumerated ~10 must be listed by beat id in

the work order so no one re-derives "all HL_0003 hits" and repeats the error.**

3. F-14 edits REALM_DESIGN_PROGRAM.md, a director-owned document, and its Ch 74 half moves

the premise of live Josh brief C-1. Both divergences are verified real. Surfacing, not

resolving — correct posture by the draft, and I confirm the underlying facts.

4. F-05's T-RLM-REGISTER ("no two adjacent road positions share a register_class") is a

hard constraint minted by L-R1 on a field L-R5 owns. If L-R5 rules that two adjacent death

realms legitimately share a register, this tooth blocks canon. **Recommend it land as a

WARN-tier check until L-R5's values exist**, then promote.

---

7. Closing verdict

Killed: 1 · Survived: 18 · Amended: 6 · Contested: 4

• *Killed:* F-01 as written — premise is a three-field conflation, enrich.py faithfully

mirrors ratified CVD §6.2, and the named Wave-0 action edits canon-mirroring code from

draft-tier data while care-inflating nine chapters. A reframed residual (three care

vocabularies undeclared + the Ch-71/§6.2 question) is handed forward at MEDIUM.

• *Amended:* F-10 (headline and repair scope cut ~75%, tooth fixture replaced), F-04, F-11,

F-16, F-18, F-02.

• *Survived unchanged:* F-03, F-05, F-06, F-07, F-08, F-09, F-12, F-13, F-14, F-15, F-17, F-19.

Landing recommendation: GO WITH FIXES

This audit is well above bar — I recomputed nine independent tallies and every one matched to

the cell, the positive controls are real and one self-reported failure is honest. Land it after

three mandatory edits, none of which need new research:

1. Rewrite F-10's headline and repair scope to the enumerated ~10 CH_69/CH_68

deity-instantiation citations. **Do not land the "re-anchor 41" instruction in any work

order** — it would strip HL_0003 from the Ch-67 Sámi-conflation guard and the Ch-71

no-extermination guard, both §17-floor material.

2. Replace T-RLM-ANCHOR's must-fire fixture with a named CH_69 audit-only citation, not

"any of the 41."

3. Strike F-01's Wave-0 step 1 from §6.3's zero-token list and reissue it as a decision

brief. Keep step 2 (the registry becomes authority, enrich.py becomes a reader) — it is

correct and needs no canon call.

F-17 should be promoted to the top of the Wave-0 actions: it is the only finding describing

something that fails silently while reporting success, and it is fully verified.

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