CRITIC_SLICE-corruption-as-condition.md

pipelines/CRITIC_SLICE-corruption-as-condition.md

CRITIC PASS — SLICE-corruption-as-condition (fresh-context refutation)

Draft under review: scratchpad/realm_wave/SLICE-corruption-as-condition_DRAFT.md (17 findings, 3

contested). Repo read-only: C:\dev\humanity-forgotten. Nothing written, edited or staged in either

repo by this agent. Date 2026-07-28.

VERDICT: GO-WITH-FIXES. 2 findings killed, 15 survive, 8 amended. The draft's citation hygiene is

unusually good — I re-derived twenty-plus quotes and line anchors and they land. The failures are

concentrated in exactly the three places the draft named as unread (HOLLOW_EXPANSION §5-12, CH_75

tail, REALM_DESIGN_PROGRAM §6-8) plus ONE under-search it did not name: it never opened

registries/T0_Boss_Encounter_Registry, which is the live home of its own primary deliverable.

---

1. KILL LIST

KILLED — F-11 "the Hollow's REALM ledger axis reaches the fairy fold by construction"

The finding's load-bearing claim is: "Nothing in either document says whether Ch 75-77 is inside or

outside Hollow placement — so on the face of the specs, it is inside."

Decisive evidence, docs/pipeline_review/HOLLOW_EXPANSION_INTEGRATION_2026-07-27.md L1043-1045:

"The MYTHOLOGICAL road Ch 66-74 (Ch 66 Atlantis dimensional-space via the Vimana; Ch 67-69
threshold traversal; the fairy realm via the World Tree crossing at Ch 74). Realm cleansing as an
ARC is therefore a Ch 66-74 structure, which is exactly where the ruled Final Ascent lives — the
two rulings compose rather than compete."

Realm cleansing is SCOPED Ch 66-74. Ch 75-77 is outside it by the spec's own sentence. L100-101

confirms the direction of travel — the deferred yield "BANK[s] its power yield, releasing it at the

ruled climb points (Ch 72-74) and the ceiling (Ch 75)" — Ch 75 is where banked yield RELEASES, not

where Hollows are placed.

Second, independent: the live T0_Hollow_Codex [DRAFT v0.1] singular row HW_0001 carries

chapter_window = CH_39-CH_65. The only Hollow row in canon is already bounded ten chapters short of

the fold.

Third: HOLLOW_EXPANSION L1047-1051 shows the doc DOES do explicit fairy-fold-adjacent placement

exclusions when it means them ("THE ONE PLACE A RIFT DUNGEON MUST NOT GO: the PROLOGUE"), so silence

elsewhere is not the open door the finding reads it as.

The draft pre-registered this exact downgrade at §7.3 ("if a later section already excludes the fairy

fold, F-11 downgrades from MISSING to COVERED"). Honoring it. MISSING → COVERED. Contested item

C-2 dissolves with it — the fork is already answered in the spec; what remains is a one-line

record in §2.4, not a ruling. The T-RLM-TOPOLOGY mechanism proposed under F-4 survives on its own

merits and does not need F-11 to justify it.

KILLED — F-10 "the fairy fold's grief register — no register row anywhere claims it"

docs/proposals/REALM_DESIGN_PROGRAM.md §4.2 L1139-1144, the L-R5 charter line, verbatim:

"L-R5 underworld-grief-and-register-variety, craft remainder. Spiritfarer · Grim Fandango · Edith
Finch · Hades's register layer · Kentucky Route Zero as an optional minor. **Spiritfarer is grief as
an entire game's register and its ferrying structure is the closest analog to the Ch 75
grief-geography encounter (grief crystallised into terrain, addressed by restraint and consent
rather than damage).** … Aspects R-A6, R-A12, R-A18."

The charter already claims the fairy fold's grief register, already names the Ch-75 encounter by

node, already carries R-A6, and already uses the finding's own two verbs — RESTRAINT and CONSENT.

The finding's assertion that "no register row anywhere claims it" is not a search that could not

match; the draft states at §7.3 that it read §4 in full (1060-1260), so it read L1139-1144 and

reported the zero anyway. That is the more serious form of the failure.

**MISSING → COVERED (lens claimed, not yet run: DESIGN_GAP_REGISTER L411/L444-448 has the realm set

CLAIMED-QUEUED at P2).** The finding's residual value is real but small: the extracted

register-variety SLICE is death-realm-scoped (L1119-1124, correctly cited), so the fairy register

rides the FULL L-R5 lens at P2, not the slice. Re-issue as a one-line routing note, not a finding.

---

2. AMENDED (survive, with the correction stated)

F-2 — REUSE-FIRST FAILURE. The registry already types the restoration register.

The finding never opens registries/T0_Boss_Encounter_Registry [DRAFT v0.1]; it reads only the

spine's thin join index at CH_75 L77-82 and concludes "not one typed field" exists. The live

registry carries 281 rows, 48 columns and six BE_0075* rows, and the columns include

win_condition, win_shape, has_nonkill_path, ontology_class, collect_verb, restore_target,

target_type, care_tier, ledger_axis, counts_toward_cleansing_meter.

The restoration register IS typed, and it discriminates exactly the way the finding wants:

rowwin_conditioncare_tier
BE_0075_S1 Unlidded EyePART_BREAKstandard-with-the-war-register
BE_0075_S2 Westering HostFREE;PACIFYelevated-the-realms-own-dead-given-dignity-win-is-rest
BE_0075_S3 Pitch-CourtREDUCE;PURGE…-and-the-HL_0109-guardrail-realms-own-dark-not-a-cabal
BE_0075_S4 Cinder-AdvanceREDUCE;PURGEstandard-with-the-war-register
BE_0075_S5 Last CoilREDUCEstandard-with-the-war-register

FREE;PACIFY against REDUCE;PURGE is the restoration/subjugation split in an enum, and S2's

care_tier string literally ends "win-is-rest."

AMENDMENT: DROP the proposed resolution_register enum — minting it creates a second vocabulary for

a concept win_condition already carries, which is precisely the two-vocabularies-one-concept defect

F-3 exists to prevent. KEEP affliction_state (native_dark | turned | uncorrupted), which is the

one axis genuinely absent — and check first whether it is derivable from the existing

awareness_category / target_type strings before minting a column. T-RLM-CONDITION then reads

win_condition + affliction_state, and it belongs on a boss-registry gate, not on

check_realm_ruleset.py (which is per-realm-row and, note, does not yet exist — §2.7 says "proposed

as harness/check_realm_ruleset.py"; the draft's "zero new gate files" is true only relative to the

charter's own proposal).

F-2 — DEVIL'S-ADVOCATE-AGAINST-TIMIDITY: the proposed tooth is care-inflated as written.

T-RLM-CONDITION as drafted: "every encounter whose affliction_state = turned must carry a

resolution_register from the restoration set (rested | unseated | restored), never ended."

Applied literally that forbids KILLING a turned noble. The live BE_0075_S3 win_condition is

REDUCE;PURGE; CH_75 L80's ratified win is "the Pitch-Court UN-THRONED and driven to its deep

earth-holds, a sovereignty unseated, never a race destroyed and never a slaughter." The canon

line is about FRAMING (no extermination of a people), not about whether individual turned combatants

die — and the standing bar is explicit that §17.1 protection is COLLECTIVE, never individual

invincibility (HL_0003: "it does not render individual members mechanically un-attackable, and it

never sanitizes toward rated-G"). A tooth that converts a framing guardrail into a mechanical

no-kill rule is exactly the care-tier inflation the calibration forbids. AMEND the tooth to assert

the framing field (target_type / care_tier must carry the never-a-people clause), never the combat

outcome.

Everything else in the draft passes the timidity check. F-5's `native_dark_risen → purge_eligible =

FALSE` is correctly bold and correctly reasoned (the tradition's own monsters are FOUGHT); C-1's

closing paragraph runs the anti-timidity pass itself and lands on the right side. No defanging

found.

F-3 — collides with a RULED, UNBLOCKED mint the finding did not see.

Worklist X-9 mints BLIGHT and CONTAIN_BLIGHT as the Hollow's encounter-row vocabulary

(HOLLOW_EXPANSION L1581, L1725). So the Hollow's canonical noun family already has a ruled token,

and it is BLIGHT — not only "contamination." The finding's CONTAMINATION/CORRUPTION lock must land

THROUGH X-9's mint, not beside it, or the repo gets three nouns instead of two.

Two smaller corrections. (a) T-RLM-CARE's declared scope is "the fringe-uptake firewall as a

FIXED-STRING scan for retired coinages and fringe-tradition names on any player-facing surface"

(§2.7 L672-674). T0_Hollow_Codex rows and RLM_0010 fields are authorial/registry, and

HOLLOW_EXPANSION §6.2's ruling makes "the Hollow" authorial-and-registry-only with the

player-facing noun attaching to the RIFT. The extension therefore WIDENS the tooth's scope and must

say so, or it will be written against a surface class the tooth does not cover. (b) The

CORRUPTION_SEAM citation creature-system.md L244 FU.2 is a stale line number inherited from

HOLLOW_EXPANSION §4.1 L650 (which describes the pre-amendment file). In the current file L244 is

the CONTAIN relocation; the trigger and dual spawn loci are at L256/L258. The token is real

(verified at creature-system.md L258 and HOLLOW_EXPANSION L652) — only the anchor is wrong.

F-4 — internal contradiction with F-3: the proposed field name imports the collision it forbids.

blight_topology is offered as ONE enum consumed by both registries. But BLIGHT is the Hollow's own

ratified X-9 token (CONTAIN_BLIGHT), and creature-system.md L242 defines a Hollow as "a blight."

Naming the shared field blight_topology therefore stamps the Hollow's noun onto the RLM_0010

fairy row — the exact blur F-3 was written to prevent, committed by F-3's own sibling finding.

AMEND: name the shared field for the SUBSTANCE both share, not for one side's product —

negative_vril_topology (the substance is HL_0114's, and the draft's own §3.1 rule 1 is phrased on

negative vril, not on blight). Everything else in F-4 stands: both halves of the topology claim are

verbatim-verified (REALM_DESIGN_PROGRAM §1.3 L308-310; HOLLOW_EXPANSION L528 "IT IS NOT DEATH AND

THERE IS NO CORPSE — the body was never alive", L579 "the eroded silhouette of what that place lost";

CH_75 L29 "woke and turned the realm's OWN dark mythology", CH75_B07 L126).

F-5 — mis-attributed hard line (substance intact, ID wrong).

The finding quotes as HL_0121: "a myth's own dragon, chaos-serpent, sea-beast, undead, ogre, or

demon is a boss to be fought, never routed around or reduced to a containment puzzle." That sentence

is real and I found it — in HL_0080's ninth-sitting amended text, not HL_0121. Live

HL_0121 reads: "Genuinely malevolent beings grounded in a tradition are legitimate antagonists and

legitimate combat… routing around fightable mythological material to make a chapter safe is a failure

of the standard." The nearest verbatim sibling is HL_0122 ("A myth's own dragon, serpent,

sea-beast, or demon is a staged boss fight, never defanged, never routed around, never reduced to a

non-combat puzzle"). Cite HL_0080 + HL_0122; the doctrinal point is unaffected and is if anything

better supported by three rows than one.

Everything else in F-5 verifies exactly and it is correctly ranked the slice's highest-priority

finding. T3_Core L593 ("heavy, crystallized, dense with accumulated emotion rather than flowing

freely" … "but the father must consent"), L595 (the attribution lock), L609 (the drift warning,

verbatim: "A later authoring pass that reaches for a second source has drifted the canon and is

corrected against this line"), and CH_75 L27 all land. The single-field-absorbs-the-grief risk is

real and the fixture-from-the-canon's-own-warning move is the right shape.

F-6 + F-17 — must collapse into ONE law. The schema forbids two.

Both findings propose a native law for the fairy realm and neither cites the two clauses that govern

that:

realm)**."

three."

realm then lands as sovereign not because it adds a ninth novelty but because it is the place

where every law the road taught is simply the local weather."

F-6's mechanism (saturation-expels-negation, which is CANON — CH_PROLOGUE L30 and beat L109,

verbatim as quoted) and F-17's proposal (perception reads FLOW, not concentration) are two different

laws for one signature slot, and the second is a novelty at the node §2.6 rules must not add one.

F-17 hand-waves the tension in one clause ("the same physics read from the other side") and that is

not enough for a schema mint.

AMENDMENT, and it is available: author ONE signature row whose law is the saturation/flow field —

canon's own — with physical_expression = flow_state, and let the perception inversion be the

CONSEQUENCE of that law rather than a second law. That keeps F-17's genuinely strong legibility

argument (the contrast instrument saturates; CH75_B02 L93's ruled verb "Tell the two wounds apart"

has no instrument), keeps T-RLM-COMPOUND green, and keeps §2.5's "NOT A DELTA. The source" honest —

the source realm's one law is the law all nine road realms partially express (LAW 0, §2.2 L398-404).

F-17's V-adequacy-probe candidacy survives intact and is a genuinely good catch: V is four terms

today (conductivity, mass, structural_integrity, thermal_state, §2.3 L488-496, verified) and

flow_state is exactly the kind of bend it cannot express.

F-1 — citation defect; verdict downgrades HIGH → MEDIUM.

The quoted charter line does not exist where cited. REALM_DESIGN_PROGRAM §4.2 L1156-1162 reads

"Hallownest is a broken kingdom with a ruined king, a court, a fading folk and an INFECTION…" —

it attributes "a fading folk" to HALLOWNEST, not to our realm. The phrase "a sovereign non-human

court in decline" is not in the charter at all; it is at

docs/pipeline_review/REALM_ANALYSIS_LENS_GAPS_2026-07-27.md L137, the R-A14 aspect row (and

L327 of the same file says "a sovereign non-human realm in decline," not "court").

The finding's substance survives on the one clause that does the importing — "structurally our fairy

realm almost line for line" — but it is weaker than presented, because the SAME charter already

carries the corrective in §1.2 L239-255: "The full vril ceiling returns for the first time since the

Prologue, the single largest power beat in the game… Intensity PEAK… Vril CEILING," plus the

guardrail verbatim. An artist is briefed from §1.2 and the art bible, not from a lens charter line.

Recommendation (b) also needs a reuse correction: the board has two image panels, not three —

Panel C is "Not images. A text card set, five slots" (§3.2 L793). A FOURTH panel changes a

gate-enforced board shape that §6.3 already ruled ("no unmarked recommendation and no more than one

marked recommendation per panel, enforced by the deterministic gate"). The reuse-first home for a

before-state reference is LANE R (§3.3 L820: "rights-cleared real imagery assembled into plate

sheets, no model runs"), or a Prologue-height plate inside Panel A — whose options are already "each

a plate sheet rather than a single image" (§3.2 L766). Same instrument, no new board shape.

F-13 — under-scoped, and the correction RAISES it: the staleness is in LIVE CANON too.

The finding scopes pre-v3 numbering to FINALE_FAIRY_ENRICHMENT and T3_Core_Characters. Verified

(headings ✓, T3_Core L549/L593/L599 ✓, live fold CH_74 Yggdrasil / CH_76 The Two Courts / CH_77 The

Waterfall ✓). But it stops one layer short. The same numbering sits in two LIVE rows:

atmosphere at Ch 70 grief geography; nearby during Ch 71-73 Realm War without direct

engagement; direct Ch 75 Father Encounter." That is a CANONICAL-severity hard-line row.

"Prologue;_grief_geography;-_realm_war_proximity;Ch_75_father_encounter;Ch_76-Ch_77_ending_cutsc…"

Precedent exists for exactly this class: HOLLOW_EXPANSION C-9 (L1475-1480) briefed HL_0020's

stale chapter number as JB-4 and it was Josh-confirmed. MEDIUM → HIGH, and the item is a

hard-line-row touch, which is a different (heavier) lane than a proposal-doc re-key.

Separately, "the ONLY place in the corpus holding the fading-realm L2 substrate" is unsupported and

is contradicted by the draft's own positive control (fading|dwindl|waning → 119 occurrences across

58 files) and by L24 of the same document ("the fairy-realm-in-decline / the fading" in the §2 Ch-75

brief). Soften to "the curated L2 fading substrate for the fairy fold."

F-15 — premise overstated; recommendation survives and strengthens.

"The fairy realm is, across its three main nodes, experienced almost entirely under fire" is

contradicted by §1.2 L267-273: Ch 77 is "No combat, no fail state, player-paced dwell," intensity

standard. One of the three main fold nodes is already an unopposed dwell sequence. The beat-mode

counts are exactly right (I re-derived them: combat ×6, character ×4, discovery ×1, vril_management

×1, transformation ×1 = 13) but CH_75 L153 is the mode_variety line, not the intensity cell

(intensity PEAK is set at L330 via CVD §16.5). The asset-sizing recommendation is CORRECT and gets

stronger with Ch 77 added to the unopposed set: Prologue-at-height, Ch 77 waterfall, Epilogue

restored.

---

3. SURVIVING UNCHANGED (spot-verified)

royal communion"; usurper L80 + CH75_B07 L123-126; power-behind L29/L83). The zero-cost

record-the-pass recommendation is correct and its risk is real — CH_75 L254 already names

"giving the family's reach a face or a front here" as an at-risk failure.

164 rows; fairy → 4 (CHAR_0001, CHAR_0005, CHAR_0059, CHAR_0164); friend → 3

(CHAR_0002, CHAR_0019, CHAR_0021); healer → 5; court → 8; noble → 1;

jarl/griever/kindler/courtier → 0. I ran one further control the draft did not — elf

12 rows — and confirmed all nine extras are incidental substring hits ("self-named", "Gebreyesus"),

so the three-row count holds. ADD one piece of evidence the finding is missing and would have

wanted: CH_75 L241's scene row uses childhood_best_friend as a participant_refs value with no

registry target — a live unresolvable reference in a ratified spine table, which is stronger than

an absence argument.

silhouette_law, impossible_architecture_budget, signature_phenomenon, three concept-choice

slots); no decay law. The landed presentation amendment verified at §5.1 L1214-1222. Timing claim

(DR-2 freeze, P1-C) verified at §4.3 L1179-1184 and DESIGN_GAP_REGISTER L406/L461.

from the live row and the champion-clearing conclusion is correct. However the finding's second

half — "HOLLOW_EXPANSION … still carries the scope ambiguity to Josh as JB-1, and still sequences

worklist item X-9 behind that resolution" — is refuted by that document's own §11.1 header block at

L1556-1566: "THE ENTIRE JB SET WAS RULED BY JOSH AT THE NINTH SITTING (2026-07-27) … JB-1 the

hard line is REMOVED for the Hollow class … X-9 UNBLOCKED." The §2.6 interim-interpretation

passage the draft read is preserved for provenance, in the same pattern as §6.2's own superseded

Hollow bullet in REALM_DESIGN_PROGRAM (L1405-1415). No supersession pass is owed; priority HIGH →

LOW; the finding is a confirmation, not an unblock.

first rift sealed (the PROOF that the King always seals it)"; T3_Core L605 "The father is not the

villain of this chapter"; L557 "no blame in him and none arriving in the years that followed"). The

two-column art-bible rule is the cheapest artifact in the slice and I would raise, not lower, its

priority.

stub, node_kind = story_anchor, last_updated 2026-05-02; yggdrasil sibling node_kind =

mythic_realm, 2026-07-12. §1.2 L230's independent record matches. harness/apply_zone_catalog_columns.py`

EXISTS and is the ruled sole writer (§5.1 L1225-1227). The state-axis-over-two-rows recommendation

is the right call for the stated FK reason.

than flowing freely" derivation, the LAW 3 / LAW 5 citations (§2.2 L413-418, L423-426) and the

V-adequacy candidacy all verify.

---

4. PRE-REGISTERED DIVERGENCES AND SCOPE GUARDS — CLEAN

against it. None addresses how the player crosses into the fairy realm, realm-transition

presentation, or conveyance. F-11's "placement" is Hollow spawn placement, not crossing. The draft

pre-registers the boundary correctly at §7.2 (FORKC_ANALYSIS_DESIGN_PRECEDENT_2026-07-27.md owns

it). Confirmed no re-litigation.

DIVERGENT-BY-DESIGN ×17** (L238, exact match to the draft's claim). No realm or fairy entry among

them. No finding here re-litigates no-scoreboard, markerless navigation, or any other listed

divergence.

is already partly done — DESIGN_GAP_REGISTER v1.2 L444-448 mints "R-A1 through R-A20 … by

reference" and names the corruption-as-condition slice as a claiming lens. What is owed is the

VERDICT FOLD into those rows, not the mint.

---

5. CONTESTED — for the director

1. C-1 (the corruption's aperture) survives as a genuine fork and does NOT duplicate anything.

I read REALM_DESIGN_PROGRAM §6.2 in full (L1366-1428): three forks — crossing intensity,

hard-line-flag asymmetry, the Hollow's admission (closed) — plus a realm-program consequence

bullet. No fairy-corruption-mechanism fork. The draft's §7.3 worry is discharged. Note §6.3

L1435 delegates "every signature law and delta" and "the schema's field set and home" to the

DIRECTOR, so this is a director ruling, not a Josh brief.

MY POSITION: agree with (b), with one bound the draft does not state. CH_PROLOGUE L225 rules

that the Prologue catastrophe "has NO AUTHOR — no targeting, no intent, no recurrence, nothing

that invites the question of who sent it," and HOLLOW_EXPANSION L1047-1051 shows what happens if

that is violated ("gives the death an author and leaks the antagonist pattern roughly thirty-eight

chapters early"). Option (b) routes the aperture through the King's sealing faculty; it must carry

an explicit clause that the Prologue's catastrophe stays authorless and is never retro-read as the

first seep. With that clause, (b) is right and (c) is the timid answer the calibration forbids —

the draft's own anti-timidity paragraph on this is correct and should be preserved verbatim.

2. One-signature-per-realm. Does the source realm take a signature-law row at all, given §2.5's

"NOT A DELTA. The source" and §2.6's "not because it adds a ninth novelty"? The F-6/F-17 collapse

above is my recommendation; the call is the director's.

3. affliction_state: mint or derive? The live awareness_category and target_type strings on

BE_0075_S1..S5 already encode most of the turned/native-dark distinction in prose form. Deriving

is cheaper than a column; the draft never checked.

4. C-3 (type the Griever's win as restoration) is effectively already ANSWERED by

BE_0075_S2.win_condition = FREE;PACIFY and its care_tier "win-is-rest." Re-frame from "should

we type it" to "the typing exists — does affliction_state add anything win_condition does

not." C-2 is dead (see the kill list).

---

6. METHOD NOTES

anchors across CH_75, CH_PROLOGUE, T3_Core_Characters, REALM_DESIGN_PROGRAM,

HOLLOW_EXPANSION_INTEGRATION, REALM_ANALYSIS_LENS_GAPS, FINALE_FAIRY_ENRICHMENT,

creature-system.md, DESIGN_GAP_REGISTER, and four live registries. Line-anchor accuracy in

CH_75 and T3_Core is exact. The two mis-anchors (F-1's charter quote, F-5's HL_0121) are

attribution slips on real text, not fabrications. I found no fabricated quote in the draft; the

"nine-tenths splendour" line I initially suspected is verbatim at CH_PROLOGUE L11.

(reproduced exactly, plus one extra control the draft did not run); the HL_0121 wording search

across all registry CSVs with routing around fightable mythological material as the control hit;

the nine-tenths search with the ceiling FELT and SERVED as the control.

outcome of an honest limits section — but two of the three (F-11, F-12) were resolvable inside the

same file the draft was already reading. A Wave-2 author should treat HOLLOW_EXPANSION §11.1 and

REALM_DESIGN_PROGRAM §6 as mandatory reads before touching any of this.

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