PRE_5090_BUILD_PLAN_VOL2.md

programs/PRE_5090_BUILD_PLAN_VOL2.md

PRE-5090 BUILD PLAN — VOLUME 2 (the enhanced item source, reconciled)

Status: ACTIVE — the authoritative item source for the vertical-slice window going forward. SLICE

RESCOPED 2026-07-27 (Josh RULED, fifth + seventh sittings): the slice is PROLOGUE→CH 13, built

Ch 2-13 first then Prologue + Ch 1. §A.0 carries the ruled shape and WINS over any Ch 2-5 boundary

statement elsewhere in this volume; the Ch 2-5 statements that describe what was BUILT stay accurate

and are not retro-edited.

This volume supersedes docs/PRE_5090_BUILD_PLAN.md §§3–11 as the ITEM registry (the ledger,

gating model, critical-path narrative, and decisions §12 in that doc remain current and are pointed

to from here). Assembled from the 8 verified plan sections + both fresh-context critics

(tasks/wavckc3ou.output: citeCritic GO_WITH_FIXES · cohereCritic GO_WITH_FIXES), with all 4

citation fixes applied and all 13 coherence findings resolved under single-owner discipline:

every item has exactly ONE id, ONE section home, ONE wave; cross-references replace duplicates.

Self-contained: header (directives · density curve · canon-first item law · gating/tag law) →

RECONCILIATION LEDGER → the 8 reconciled sections → the rebuilt MASTER WAVE QUEUE (with the cut

line) → the DECISIONS REGISTER → WHAT-THIS-DOES-NOT-CLAIM.

---

0. THE DIRECTIVES (Josh, 2026-07-24 — verbatim intent, binding on every section)

1. Target = the CH 2-5 VERTICAL SLICE: full playthrough, cross-chapter progression, 100% complete

structure at max honest pre-5090 polish. The 5090 + book scans arrive in 2-3 weeks; post-5090 =

full generative passes (VFX/SFX/cinematics/HUD art/music) + the deep-read enrichment pass over

everything (weavings, culture, side quests, relationships).

2. The playable-game surface has gaps the plan must close (bindings, quest log, lock-on, hit

confirmation, per-weapon/ability presentation...). The plan needs SIGNIFICANT enhancement.

3. The living-world substrate is required, grounded in real regional biota/geology: fauna, herbs,

minerals, trades, the ability book, + unnamed members.

4. CANON-FIRST (Josh caught "new" labels on documented canon): every item carries a canon citation

or a positive-controlled absence proof. Prior work is NEVER re-invented.

5. FULLY AUTONOMOUSLY TESTABLE: the machine is the alpha/beta tester (WS-10 expansion QT-1..22).

6. THE DENSITY MODEL (ruled): a completion-hours CURVE — full game ≈100-200h main / 1-2k h side+

weave / 8-10k h 100%. Ch 2-5 = early-game, deliberately lighter, seed-rich, opening the game.

Josh's in-conversation refinement SUPERSEDES the CANON_AUDIT #7 uniform-density reading.

The density curve (Josh RULED — recorded verbatim)

"THE DENSITY MODEL (ruled): a completion-hours CURVE — full game ≈100-200h main / 1-2k h side+weave /

8-10k h 100%. Ch 2-5 = early-game, deliberately lighter, seed-rich, opening the game. Josh's

in-conversation refinement SUPERSEDES the CANON_AUDIT #7 uniform-density reading."

What the ruling does NOT touch (LOCKED, still binding): the T0_Chapter_Index topography gate

(no-3-consecutive-same-type, mode-variety) and CVD §16/§4.4/§12.2 pacing — those govern *shape within*

the curve, not a flat amount. The curve is a refinement of #7, not a repeal of the gate. Made

measurable by the per-chapter content-budget + test-coverage scorecard (check_slice_scorecard.py;

spec in the PROGRAM FRAME §A.1 and the WS-10 §11.5.5 scorecard family).

The canon-first item law (binding on every item in every section)

Every item carries either a real canon citation (CVD / tier doc / registry row / prior plan id /

ratified proposal) OR a positive-controlled absence-proof (the search that WOULD have found it, run,

returning zero). Prior work is NEVER re-invented (directive #4). An "absence-proof" is a

first-class citation: a zero is only trusted with its positive control (memory:

*a-search-that-cannot-match-reports-zero*). The critics' phantom-fold corrections are standing

precedent, already applied, not re-litigated: lock-on/target-switch → P2.2; input set/dodge/interact/

wheel → P2.3 (+Q3.2); player HUD/UMG → P2.4; gather/harvest → P2.10/P2.5; journal → Q3.6; Grimoire →

Q3.8; char-progression UI → P2.8; LINK_0186 EXISTS (AB-13 killed); Boss_Encounter already speccs

BE_0002..0005 (16 rows — consume, never re-author); Creature_Roster = 132 rows (only numeric cols

blank); MV-1 free-travel is a phantom (Vimana-gated Ch 55, WGM §7 — the slice = ruled travel-beats

forward + free-roam within regions).

The gating model + TAG LAW (binding)

The RTX 5090 gates ONLY heavy generative INFERENCE (book deep-read/triage, photoreal 3D/asset gen,

music/voice/SFX generation, cinematic-video gen, VLM-OCR). **TAG LAW: only generative INFERENCE is

5090-gated.** Hand/mocap animation, hit-stop, damage numbers, camera shake, UMG, the color-independent

SHAPE-telegraph channel (an L1 accessibility FLOOR, T99_Translation_Combat §2.6) = BUILD-NOW.

Presentation lands in tiers: T0 greybox (BUILD-NOW) → T1 placeholder anim/SFX (BUILD-NOW /

GATED-EXT for mocap) → T2 final art (GATED-5090). The slice ships on T0+T1; T2 drops in

post-5090 without a re-wire. Tags: DONE · BUILD-NOW · CA (canon-author-now) · GATED-5090 ·

GATED-SCAN · GATED-EXT · BLOCKED-ON-BRIEF (waits on a Josh ruling).

---

RECONCILIATION LEDGER — the critics' 17 findings, resolved

Applied verbatim from tasks/wavckc3ou.output. Nothing below is re-litigated; each row is the

single-owner outcome the rest of this volume is built on.

The 4 citation fixes (cite-critic, GO_WITH_FIXES)

*Cite-critic note (not a defect):* "DT_SceneAnchor 439 rows" (ws56_av/ws3_prog) is the **game-repo

build count**; canon T0_Scene_Spec_Registry has 16 rows. The drafts label 439 game-side, so it is

not a canon-cite error — retained as-is.

The 13 coherence resolutions (cohere-critic, GO_WITH_FIXES) — single-owner discipline

Naming canonicalized: the boss HP=0 defeat ruling is D-DEFEAT-AXIS throughout (aliases seen in

drafts: D-BOSS-DEFEAT-AXIS, D-HP0-AXIS — same brief).

---

---

SECTION 1 — THE PROGRAM FRAME + THE 5090 HANDSHAKE

RECONCILED (Vol2): §A.4's original master queue is SUPERSEDED by this volume's MASTER WAVE QUEUE (orphans W-AI/W-PRES/W-MOVE/MG/CN now homed; audio-spine/input/backbone re-ordered). §A.5 D-ECONOMY-MEDIUM corrected to JOSH-PENDING (was wrongly 'ruled — director-applied'). D-HP0-AXIS renamed D-DEFEAT-AXIS (alias). §A.0–A.3 + §B (DR-1..8 handshake) remain current.

THE PROGRAM FRAME + THE 5090 HANDSHAKE

Front-of-plan restatement for the enhanced PRE_5090_BUILD_PLAN. Governs every section below it.

The named queue A/B/C/D is CLOSED (Josh can press Play and be inside Chapter 2, game 303e9f2). This

frame re-ranks the 2-3-week window to its real target and lays the drop-in surface the 5090 lands on.

---

A. THE PROGRAM FRAME

A.0 What this window is — the PROLOGUE→CH 13 vertical slice

RESCOPED 2026-07-27 (Josh RULED, fifth + seventh sittings). The pre-rescope text of this section read

"The whole pre-5090 program has ONE target: CH 2-5 as a full playable vertical slice" — that sentence is

now FALSE and is superseded here. The Ch 2-5 statements elsewhere in this volume are historically accurate

about what was BUILT and are not retro-edited; where any of them asserts the slice BOUNDARY, this section

wins.

The pre-5090 program's target is PROLOGUE → CH 13 as a full playable vertical slice (15 nodes) —

start-to-finish playthrough, cross-chapter progression carried across 13 regions, a 100%-complete

*structure* at maximum honest pre-hardware polish (Josh directive 2026-07-24 #1, scope re-pointed by the

2026-07-27 ruling). "Complete structure" ≠ complete fidelity: every beat, site, boss, fork, gather-node,

persona and weave-anchor is *present and playable* on placeholder assets; the final art/audio/cinematics

and full-fidelity enrichment ride the post-5090 passes (§A.3).

• Built today (the Ch 2-5 leg): flores_island (Ch2-3), bali (Ch4), sumatra_java + angkor_khmer

(Ch5 excursion) — chapter→region confirmed from spine headers + Zone Catalogs (ch2_5_slice_map.md).

The MASTER WAVE QUEUE's "CLOSED" statuses are correct for this leg and INCOMPLETE for the ruled slice.

• The Ch 6-13 leg (eight new regions): builds next, in order, and inherits the generalization already

paid for (the --slug/--chapter drivers), so terrain and staging are extension, not re-architecture.

Its eight region pages do not exist yet and are the slice's longest ungated pole (nine counting the

fairy_realm page, which sits on the Prologue+Ch 1 leg and has no template).

• The Prologue + Ch 1 leg (fairy_realm): builds LAST, once, at slice-end — after 11 earthly chapters

have matured the concept-art pipeline, the presentation ladder, the W-SPACE grammar and the QA belt.

Its terrain is invented (coordinates: N/A (dimensional)), so the deterministic DEM chain has no input

and the authored-terrain path is prototyped during the Ch 6-13 window before slice-end consumes it.

• The slice closes at Ch 13 with the healer's child joining and the departure, the amulet passing to

her at that departure — the to-be-continued act-break, and the Early Access base in full (no early cut).

• The Josh Gate fires twice: Gate 1 at the Ch 2-13 P3-exit (his verdict DIRECTS the fairy realm's

design, while it is still designable); Gate 2 at the full-slice exit (the origin judgement, and the gate

that opens the Steam listing).

RESCOPE RE-DERIVE — the slice-DEFERRAL class (2026-07-27). The precedence rule at the head of this

volume covers two classes of Ch 2-5 statement: BOUNDARY statements, where this section wins, and

descriptions of what was BUILT, which stay accurate. A third class needs naming on its own, because its

rows are live instructions to the factory whose PREMISE the rescope inverts — each was decided while

Ch 2-5 WAS the whole slice, and the ruled slice now runs to Ch 13. For every row below the premise is

invalidated and the new answer is deliberately NOT written here: each re-derives at its own item, against

the ruled 15-node scope, before it is built or skipped. Treat an unre-derived deferral in this class as

un-decided, never as inherited.

• §11.6.5 trade build order — it reads "the slice is entirely pre-first-trade-home (earliest ancient home

is Ch7, Blacksmith+Herbalism, S.India)". Ch 7 is INSIDE the ruled slice, and the execution brief counts

four trade homes among the Ch 6-13 system debuts. Re-derive the introduction order and which trades

reach a real home in-slice.

• TR-6 (Neophyte-gate quest ROWS for the soonest trade homes) — its "not slice-blocking (no Ch2-5 trade

home)" note now rests on a false premise. Re-derive its wave placement.

• TR-7 (crafting-station runtime stub) — "correctly deferred by beat-presence (no compound instantiates

in Ch2-5; first at Ch7 Forge+Apothecary)". The first compound is now in-slice. Re-derive.

• U11.18 (journal slice stub, empty-state by design), with the same premise carried by the empty-state

bullet in the screens section and by the slice-honesty paragraph — Q3.6's Ch-13 activation was OUTSIDE

the old slice and is now its TERMINAL node, and the brief's played-versus-told table lists the join and

the journal beginning at Ch 13 as PLAYED. Re-derive whether the journal ships as a correctly-empty

shell or as an activating system.

• SL-1 (death/respawn graduated persistence) — its death-learning fallback rests on "slice is pre-journal

(journal activates Ch 13)". Re-derive once U11.18 re-derives.

• CN-9 (DT_SceneDispatch path-conditional resolver, slice-DEFERRED), with the same premise in the

ws56_av wave table — path-invariance was measured over Ch 2-5's forks. The ruled slice carries fifteen

forks, and the Prologue + Ch 1 leg carries the choosings. Re-derive path-invariance over Ch 6-13 and

Pro/Ch 1 before inheriting the deferral; the Josh #22 schema question is unaffected either way.

• UW-2 (mastery ladder component, "all Ch 2-5 = Neophyte") and the ability-book sparseness note (~11

abilities across Ch 2-5) — whether the player is still Neophyte at Ch 13 is an open scope question now

rather than a settled one. Re-derive the in-slice tier span.

A.1 The density curve (Josh RULED — supersedes the uniform-density reading) + the scorecard spec

The density model is a completion-hours CURVE, not a uniform bar. Josh's in-conversation ruling

(2026-07-24, recorded verbatim in the directive pack):

"THE DENSITY MODEL (ruled): a completion-hours CURVE — full game ≈100-200h main / 1-2k h side+weave /

8-10k h 100%. Ch 2-5 = early-game, deliberately lighter, seed-rich, opening the game. Josh's

in-conversation refinement SUPERSEDES the CANON_AUDIT #7 uniform-density reading."

This RECONCILES the one genuine gap Dive #3 surfaced in the density cluster (owners=12, absent=1): *"the

spoken 'starting chapters less dense' ramp is unwritten AND conflicts with #7's uniform-density bar —

needs Josh to reconcile."* It is now reconciled: density is a ramp, Ch 2-5 sit at the low, seed-rich

end. What the ruling does NOT touch (LOCKED, still binding): the T0_Chapter_Index topography gate

(no-3-consecutive-same-type, mode-variety) and CVD §16/§4.4/§12.2 pacing — those govern *shape within*

the curve, not a flat amount. The curve is a refinement of #7, not a repeal of the gate.

The curve made measurable — the per-chapter content-budget + test-coverage scorecard (SPEC). Each

chapter carries a scorecard row; the early-game target is deliberately the lightest honest floor. Columns:

The scorecard is a gate artifact, not prose: check_slice_scorecard.py reads it, and no chapter is

"slice-complete" until its row hits target AND test-coverage = 100% of its main-path beats. This is the

density curve as an enforceable law rather than a feeling.

**SLICE-RESCOPE DEPENDENCY (2026-07-27) — the bands must EXTEND before this gate is armed past Ch 5, and

they are deliberately NOT set here.** The table above is a Ch 2-5 early-game band, and the ruled slice now

spans Prologue→Ch 13:

• Ch 6-13 sit FURTHER UP the same curve. The curve is monotone-rising early→mid, so running the Ch 2-5

band over Ch 6-13 either passes vacuously or reports a chapter as thin against the wrong floor. Those

eight nodes need their own band values on the same ramp.

• The Prologue and Ch 1 sit OFF the curve entirely. The Prologue is a ceiling TUTORIAL (a presentation

reference, not a content-budget node) and Ch 1 is declared out of corpus scope. They need DECLARED

SENTINELS — an explicit exemption the gate reads — not band values. Without them the monotonicity tooth

fires a FALSE regression the moment a 15-node slice is scored, which is the same defect class already

caught for Ch 77's exempt demand_ordinal.

• Why no numbers are written here: inventing band values would cross the tuning firewall (this program

gates shapes, relations and floors; it never asserts an absolute is "correct"). The band/sentinel pass is

its own item, and it is the precondition for arming QT-19 on any node past Ch 5.

• State check, positive-controlled: check_slice_scorecard.py does not exist in either tree today — a

name search returns zero while the same search returns hits for other Tools/*.py, so the zero is real

and not a search that cannot match. The band work therefore lands WITH the script, not as a retrofit.

A.2 The canon-first item law (binding on every section)

Josh caught "new" labels stuck on documented canon. The law, now enforced plan-wide: **every item in

every section carries either a real canon citation (CVD / tier doc / registry row / prior plan id / ratified

proposal) OR a positive-controlled absence-proof (the search that WOULD have found it, run, returning

zero). Prior work is NEVER re-invented** (directive #4). The critics' phantom-fold corrections are the

standing precedent and are already applied, not re-litigated: lock-on/target-switch → P2.2; input

set/dodge/interact/wheel → P2.3 (+Q3.2); player HUD/UMG → P2.4; gather/harvest → P2.10/P2.5; journal →

Q3.6; Grimoire → Q3.8; char-progression UI → P2.8; LINK_0186 EXISTS (kill AB-13); Boss_Encounter

already speccs BE_0002..0005 (16 rows — consume, never re-author); Creature_Roster = 132 rows (only

numeric cols blank). An "absence-proof" is a first-class citation: a zero is only trusted with its

positive control (memory: *a-search-that-cannot-match-reports-zero*). Where this frame reconstructs an

item from a 0-byte recovery file (§B, DR-2/DR-8), the 0-byte file IS the absence-proof and is flagged.

A.3 The post-5090 sequence (what the window is building the runway for)

The 5090 + finished book scans arrive in ~2-3 weeks and gate ONLY heavy generative inference (§0 gating

model; TAG LAW: only generative INFERENCE is gated — hand/mocap anim, hit-stop, damage numbers, camera

shake, UMG, the SHAPE telegraph accessibility floor are BUILD-NOW). Post-hardware runs in a fixed order,

and every stage lands as drop-in onto the surface this window leaves (§B):

1. DEEP-READS — full-corpus book triage + deep-read of the scanned library, VLM-OCR of non-Latin

scripts. The input substrate. (Gated: 5090 inference + Josh's scans.)

2. THE FULL ENRICHMENT PASS — over *everything*, grounded in the deep-read output: weavings, culture,

side quests, relationships, and every living-world row (FA/FL/MN/TR) taken from slice-placeholder to

full fidelity — pharmacology, ecology, folklore attribution, significance lenses, persona depth. This

is the pass that makes the seed-rich early chapters deep. It writes onto the stable ids + weave anchors

the DR handshake planted; it does not re-place or re-author structure.

3. GENERATIVE PASSES, PER SURFACE — VFX / SFX / cinematics / HUD art / music / 3D geometry / PBR

textures / voice. Each pass writes its generated asset path back onto a stable id through the

asset-path drop-in contract (DR-2). No pass re-stages the world; it fills the slots.

The whole point of the pre-5090 window is that stages 2-3 are pure fill, never rework. That is what

the handshake (§B) guarantees.

A.4 The re-ranked master wave queue — SUPERSEDED by this volume's MASTER WAVE QUEUE (retained below for provenance only)

The named queue A-D is closed. This re-ranks the remaining surface for the window, consuming every

section's item families. Ranking follows the directive: **E-residue first · nightly-soak + feel waves

early · terrain long-poles parallel from day 1 · canon-author lane on the canon side throughout.**

Two-repo discipline holds: game repo (C:/dev/Humanity/Humanity) and canon repo

(C:/dev/humanity-forgotten) are one-writer-each, so the persistent canon + soak lanes run *in parallel*

with the ranked game waves.

Persistent lanes (run the entire window, parallel to everything):

Ranked game waves (one writer, game side):

Note: the DR staging legs (DR-3/4/5/6) are construction rules applied during W1/W5/W6 — an NPC,

gather-node or dwelling is placed carrying its stable id + asset handle at placement time — then AUDITED

in W8. They are not a separate late build; W8 is the proof, not the placement.

A.5 Decisions this frame carries

Per the standing decision protocol: never a bare question — alternatives, deep-dive, adversarial pass,

recommendation, strongest objection. Creative-within-vision → ruled and applied (show the path);

genuine cardinal/system-canon → briefed for Josh.

**✅ 2026-07-24 — Josh ruled ALL FIVE live briefs in-conversation ("I agree with all your

recommendations"): D-VO-POLICY = PARTIAL · D-STEAM-FORMAT = EARLY ACCESS · D-PRICING = P4 w/ the

$20-permanent-discount-off-$69.99 direction recorded verbatim (tier: AAAAA/Quintuple-A) · D-ECONOMY-MEDIUM =

Option B ruled · D-DEFEAT-AXIS = brief-at-the-feel-wave confirmed. The ship-side three live in

docs/ROADMAP_POST_5090_TO_SHIP.md's register (collapsed same landing); memory:

ship-rulings-vo-steam-pricing.**

D-ECONOMY-MEDIUM — ✅ RULED Option B (Josh 2026-07-24). EC-medium + EC-vendor UNBLOCK; the per-region exchange designs are LANE-CANON author items (cited to each region's real documented systems). The *model* exists

(T1_Trades_Master [ACTIVE v1.0] §9.2 "Trade Economy at System Level" — trade-interdependence, two

economies); the *medium* (coin/price/unit) does not, and the TR/loot/vendor loop is live in Ch2-5.

• Options. (A) Universal coin from Ch2 — REJECTED: breaks §9.2's immersive-teaching intent and is

ahistorical for the era/regions. (B) Historically-real mixed exchange per region — barter +

weight-value + prestige-goods, NO universal currency early — RECOMMENDED. (C) No economy in the slice,

defer entirely — REJECTED: the gather→craft→trade loop is active content in Ch2-5.

• ✅ RULED (Josh, 2026-07-24: "historically real mixed exchange per region, no universal coin early - agreed with your recommendation"). Option B for the slice: per-region

mixed exchange, Neophyte-floor only, L2 treasure placed but sparse (early-game curve). Fits §9.2 +

immersive-teaching. Full vendor economy + balance ride post-playtest.

• Strongest objection. Per-region exchange multiplies vendor-UI complexity in the slice. Rebuttal:

the slice needs only the Neophyte-floor exchange surface + L2 placement, not a full market sim; Josh

backstops via playtest. The only cardinal fork inside this (if it ever surfaces) is a

universal-coin-vs-never *canon* call — that alone would flag; the medium mechanics do not.

D-DEFEAT-AXIS (alias D-HP0-AXIS) — the boss defeat-axis design gap (C:/dev/Humanity/Humanity/TODO_CANON.md §"Design gap"

L194-207; QA image-critic v3/v4 recorded Boss HP 0/100 while Composure 100/100 and the fight continued).

Blocks CF-10 defeat presentation + PG death/respawn polish.

• Options. (A) Health IS the defeat axis; HP-0 → death/execution — conflicts with the

Composure-break-to-Withdrawal thesis and the zero-enemy-stat-delta headline. (B) **Composure-break-to-

Withdrawal is the sole defeat axis; boss Health is a non-lethal pressure/stagger track** — RECOMMENDED,

matches Pillar 8 + the observable-thesis invariant. (C) HP-0 triggers a stagger/execution *window* that

feeds Composure — hybrid; more feel, more surface.

• Recommendation. Option B, with a thin C-flavored stagger read for feel (HP-0 → brief stagger, no

death). This touches the combat pillar (system-canon) — briefed for Josh, recommendation ready; it

is Phase-5M-adjacent and the greybox has no death state to wire until it is ruled.

• **✅ RULED (Josh, 2026-07-24, at the W2 brief — OVERRULING Option B): "Rated M game. Things

will die. Health 0 is death. If they're a specific boss or combatant that we don't kill for some

specific canonical reason because it'll gate story progression then they'd have to withdraw at 1

health."** HP is a REAL lethal track (Option A + the canonical clamp): HP-0 = death for killable

enemies — BE_0002's own row carries an explicit death outcome the Option-B analysis MISSED — and

the protected class floors at 1 HP → forced WITHDRAWAL (no-dead-end made mechanical), assigned

per-boss from registry data (wave G's win_condition back-fill), never a global rule.

Composure-break→Withdrawal remains the signature mechanic ALONGSIDE lethal HP. W2 builds the full

defeat state (both paths + tests) under this ruling; the H13 split is retired.

NG+ / prestige forks — NOTE ONLY (out of slice). New Game+ carry-over across "The Forgotten One"

(mastery, familiars, inventory, world-state, the un-pruned seventh vril-seat — decision #28) stays

proposal-tier and pending Josh. **Not in the 2-3-week window; recorded here so the save/progression

waves (PG-6, W6) leave the carry-over seam addressable, nothing more.**

---

B. THE 5090 HANDSHAKE (DR-1..8)

The drop-in doctrine. The post-5090 passes (§A.3) must be *fill, not rework*. That is only true if the

pre-5090 window leaves every surface addressable: a stable id on every actor, a single documented

contract for how a generated asset attaches to an id, placeholder-actor legs already placed for every

generable category, weave anchors planted for the enrichment pass, and a clean substrate for the

deep-read. DR-1..8 are that readiness pass. (Source note: the DR recovery agent's output file

tasks/a641d9b80d0fed63c.output is 0 bytes — the StructuredOutput-cap death class, disk work survives

but the return did not; DR-1..8 are reconstructed from the authoritative task directive + standing canon

homes + the RAG/substrate memory index, and each reconstructed item is flagged with its grounding.)

Handshake exit condition. The window closes 5090-ready when: every staged actor passes the DR-1

stable-id audit (W4.9 DataValidation green); the DR-2 asset-path drop-in contract is ratified and a

placeholder mesh/SFX round-trips its five-state machine; DR-3..6 staging legs are placed and audited

(W8); DR-7 weave anchors sit on the slice's weave-bearing content; and DR-8's 3 RAG fixes are

gate-verified. At that point stages A.3.2 (enrichment) and A.3.3 (generative) are pure fill — the drop-in

doctrine holds and the 2-3-week investment converts to finished game without rework.

---

SECTION 2 — WS-1/2 COMBAT FEEL · MOVEMENT · TARGETING · ENEMY-AI

RECONCILED (Vol2): CR-CHAKRA → cross-ref to AB-5 (the single Chakra-repair owner, LANE-CANON/W-LW1, CA). AU-10 → cross-ref to ws56_av AU-10 (audio home). CF-10's boss-defeat STATE is UNBLOCKED — D-DEFEAT-AXIS RESOLVED 2026-07-24 (Josh overruled: rated-M, HP 0 = death; per-boss non-kill via the pip model; executed by R1 + R-DATA — see the FUN-REBUILD absorption map + the register). §G waves map to the MASTER WAVE QUEUE: W-FEEL1→W2 · W-TGT→W3 · W-AI→W5 · W-MOVE→W5 · W-FEEL2→W9 · W-PRES→W9.

WS-1 / WS-2 ENHANCEMENT — Combat Feel · Movement & Traversal · Targeting & Enemy-AI Breadth

The CH 2-5 vertical-slice enhancement of the combat/player lanes. The trunk built the data layer

(185 abilities, 72 weapons, the 16 slice BE rows, the 42-var worldstate, the §9.5 evaluator) and the

channel layer (C1.16 GameplayCue, native combat tags, the DamageExecCalc, the BE_0002 StateTree)

— but not the FEEL layer: every cue routes to a no-op notify stub, and every feel-read lives on

the debug Canvas HUD. This section closes that. Tag law (binding, §0 + the two dive critics): the

feel primitives — hit-stop, floating damage numbers, target flash, camera shake/punch, hit-react

states, the color-independent SHAPE telegraph channel, UMG — are deterministic BUILD-NOW; only the

final Niagara VFX / generated meshes/textures are GATED-5090, and final mocap/anim is GATED-EXT

(an external-service/hand-auth axis, never generative inference). Presentation lands in tiers:

T0 greybox feel (BUILD-NOW) → T1 placeholder anim/SFX (BUILD-NOW / GATED-EXT for mocap) → T2

final art (GATED-5090). The slice ships on T0+T1; T2 drops in post-5090 without a re-wire.

Positive-control (dive #1 repoCritic verification_done, both critics GO_WITH_FIXES): grep of game

Source/ returns 0 for CameraShake, HitStop, DamageNumber, HitReact, Flinch, Stagger,

Knockback, LockOn, UUserWidget/CreateWidget. Read of InputConfig (4 of 8 verbs bound; no

crouch/dodge/interact/lock-on/gamepad), HumanityCharacter/BossCharacter/GameplayAbility (lock-on =

"P2.2 deferred"; boss header "does not locomote"; targeting = nearest-in-800uu stub). The gaps are real

in CODE; the phantom-folds below are already-tracked plan item ids (fold ledger, §E).

---

A. The hit-confirmation stack + camera feel — *the single highest-leverage new wave*

Josh's named gap ("how the player KNOWS an attack landed"). Rides the DONE DamageExecCalc + C1.16

cue channel — zero new canon deps; the highest "does it feel like a game" delta per engineer-hour.

Verified-by (W-FEEL1): the WS-10 loop (QA-1/2/5/6 LIVE) — a functional test asserts global

TimeDilation dips within the hit frame and a DamageNumber widget spawns; **image critics read a

before/after capture pair** on the C1.18 rig (the standing visual-verification directive) and PASS the

BREAK-moment frame. Runs parallel to wave E (no shared file with the Ch-2 staging).

---

B. Enemy reactions, telegraph & boss-attack variety

Verified-by (W-FEEL2): a functional test asserts a HitReact state is entered on damage and that P2

adds (not swaps) P1's tasks; an image critic PASSes the telegraph shape in greyscale (the L1

accessibility floor — legible with color stripped). Sequence W-FEEL2 with / just after wave G (G's

sub-bosses + Ebu Gogo consume the react + attack-shape substrate).

---

C. Lock-on & targeting — the P2.2 build spec (the four dive-ids fold here; do not re-mint)

Josh's named gap. CF-3 / MV-6 / TG-1 all restate plan P2.2 ("soft lock-on + toggle +

target-switch"); the input action IN-10 rides P2.3 (UI/input section). This section supplies the

build spec for the P2.2 item; the key binding is cross-section (P2.3), the reticle UMG is

cross-section (P2.4). One net-new sub-item survives the fold: TG-2, the targetable-set abstraction.

Verified-by (W-TGT): a functional test acquires the nearest ITargetable, cycles to the next, and

asserts the camera yaw-blends; a second asserts a non-Composure fauna actor (CR_0001) is a legal

lock-on target. This wave makes multi-enemy combat (W-AI) playable — sequence it before W-AI's

N-enemy encounters.

---

D. Per-weapon / per-ability presentation tiers

The presentation-tier model, grounded in canon: **T1_Combat_System_Spec §5.2 (Per-Weapon Canonical

Structure)** — every weapon carries 4 abilities + a Mastery ability at Proficiency 8 + a Legendary form

at 99; and the cast_pattern enum (`instant | charge_hold | sustained_drain | channeled | passive |

ambient`, §Damage Formula) that the 185 abilities already populate. The engine reads none of it.

Verified-by (W-PRES): an image critic confirms ≥3 visually-distinct cast primitives across

cast_pattern classes; a functional test asserts the light/heavy/combo timing window differs by

weapon_class; a grep asserts Chakra = 0 in player-facing ability labels post-repair.

---

E. Multi-enemy combat, enemy locomotion, spawn framework & non-boss AI

Today the enemy surface = one stationary 1v1 boss dist-checking GetPlayerPawn(0). The slice beats

need pursuit, N-enemy fights, fauna/NPC/construct/companion actors, and legible spawns. **All

deterministic BUILD-NOW; only meshes/anims are art-gated** (GATED-EXT/GATED-5090). Spawns obey the

CVD Pillar-1 Hybrid policy — three declared systems ONLY (CVD L168-170: *Aggro Magnet · Rift

Instability Zones · Antagonist Hunter Triggers*; executable test: any random encounter outside the three

= pillar violation). The faction director raises a weight the three systems read; it **spawns

nothing itself**, and the LLM never spawns (T99 Step 8; A.3-A.6). Living ecology is RULED

(migrate/regrow; progression-critical entities immune). The 16 slice BE rows (BE_0002..BE_0005_S4) are

canon — consumed, never re-authored; patrols owned by the runtime executor.

Verified-by (W-AI): the CH02_SliceWalk asserts an enemy pursues the player across navmesh

(TG-3) and an N-enemy encounter deals per-target damage (TG-4); a spawn test asserts no random spawn

fires outside the three declared systems (the Pillar-1 gate). **W-AI is the prerequisite substrate for

wave G** (sub-bosses + Ebu Gogo) — land W-AI at/before G.

---

F. Movement & traversal mechanics — *the slice window (HL_0079 sits inside Ch 2)*

A 173 km island with a 2 352 m volcanic spine, ringed by the live FloresOcean plane, is crossed today

only by the DEV M-key teleport. The Ch-2 canopy agility course is a HARD LINE inside the slice. This

section owns the traversal mechanics; the key bindings (crouch/dodge/interact) ride P2.3

(UI/input, cross-section); site-to-site fast-travel (MV-1) rides the transitions section.

Verified-by (W-MOVE): the Ch-2 Flores canopy course completes headlessly via the WS-10

AI-playtest InjectInputForAction path (12 pods collected, no stop) — the HL_0079 course is the wave's

formal exit; a shoreline test asserts MOVE_Swimming engages at the water volume; a footfall test

asserts the jungle modifier applies over a footfall_surface=jungle zone; a VERTICAL exit test

asserts the climb mode ascends a tagged vertical surface (v1.3 fold entry 54 — the canopy course is

horizontal, so without this the wave closes green with no vertical capability proven; if the schedule

cannot absorb the CMC climb mode, ship the fallback instead: traversal_capability_pending tags on

vertical zone rows so the slope teeth stop certifying sites the player cannot traverse).

---

C-brief. D-BOSS-DEFEAT-AXIS — the boss HP=0 defeat model *(open brief, recommendation ready)*

The gap (TODO_CANON): the greybox has no death/stagger/defeat model for HP=0 — it blocks CF-10 / PG

presentation polish. Not a bare question — the alternatives, deep-dive, and recommendation:

• Alt 1 (status quo): HP=0 → death ragdoll. *Con:* canon-wrong for a FREED boss — BE_0002 wins by

the polo's hold breaking and Rangku freed (CH_02.md L230), not a corpse; violates §17.1 (the host

is a coerced victim) and the RULED win-condition enum (T99 §5.1 REDUCE "never a pure DPS check").

• Alt 2 (RECOMMENDED): "defeat" = the win_condition is met (Composure fully broken → the

win_shape resolves), NOT HP=0. On win-condition met → a boss defeat/withdrawal state plays the

canon resolution (FREE → the hunter is freed and exits; PACIFY → yield; PURGE → the manifestation

purged). HP=0 is retained only as a fallback safety valve (the Ch-2 HP fallback at 15/18),

routed to the same withdrawal state, never a ragdoll for a FREE/PACIFY/PURGE win. Aligns with

no-dead-end (§2.5) + the RULED enum + the freed fiction.

• Alt 3: per-boss authored defeat asset. *Con:* out of slice scope; folds into T2 post-5090.

Strongest objection to Alt 2: the win-condition→defeat mapping needs the win_condition/win_shape

columns populated on the 16 slice rows (declared in T99 §4, back-fillable from the phase prose) —

so the brief carries a small data dependency (16-row back-fill), not just an engine choice. **Rec:

Alt 2**, with the 16-row win_condition/win_shape back-fill riding the same commit as the CF-10 shell.

Josh backstops via playtest; the choice is reversible (a defeat-state swap), so erring toward the

canon-honest FREE model is the cheap error.

---

G. Wave assignment — the re-ranked queue for the 2-3-week window

Six new feel/traversal waves, slotted against the E/F/G/H residue. Ranked by leverage (feel-core

first — it is the largest "is this a game" delta and has zero new deps):

Residue-wave interlock: W-AI lands the enemy substrate wave G (BE_0002_S1/S2 + Ebu Gogo) needs —

G's Ebu Gogo pursuit = TG-3/TG-11, BE_0002_S2 = the TG-9 construct class; sequence **W-TGT → W-AI → G →

W-FEEL2. The T2 final Niagara/VFX art** for every CF/CF-* cue rides wave H / GATED-5090 (the

polish riders) — the T0/T1 feel ships now, T2 drops in without a re-wire. W-MOVE depends on the

P2.3 binding layer (UI/input section) for crouch/dodge/interact keys and on E/F for the walkable Flores

terrain.

Test-coverage rider (WS-10 / QT expansion): the QT-1..22 autonomous test-org gains three combat

packs this program adds — QT-feel (hit-stop frame + damage-number spawn + BREAK-moment image pair),

QT-target (acquire/cycle/yaw-blend + non-Composure targetable), QT-traverse (headless canopy

course + swim volume + fall model). The nightly soak (starting this week) exercises the SliceWalk with

the feel + AI waves live, so a feel regression surfaces as a failed capture the morning after it lands.

---

E-ledger. Phantom-fold & cross-section map — *nothing here is re-invented*

Applied verbatim from the binding critic corrections; every fold points at an already-tracked plan id so

no consumer builds a duplicate:

• lock-on / target-switch / camera-lock → P2.2 (folds CF-3, MV-6, TG-1; input action IN-10 →

P2.3). Presented above as the P2.2 build spec, not new items.

• player combat HUD / UMG (CF-11: Composure bar, node pips, lock-on reticle, hit markers) →

P2.4 (UI section). This section fires the cue/feel events; P2.4 owns the widget home.

• input bindings (IN-1..IN-12, MV-10) → P2.3 (+ six-option/dialogue wheel → Q3.2). This

section owns the MECHANIC each binding drives (dodge motion, crouch state, interact dispatch); the

binding layer is P2.3's.

• combat/UI/movement SFX (AU-3/9/10/11) → WS-6 (A6.5/A6.8). This section names the cue timing

hooks; the SFX banks ride the audio lane.

• boss-phase cinematic trigger (CN-7) → WS-5 (CN5.6). CF-9's phase pushes emit the event the

cinematic listens for; the Sequencer side is WS-5's.

• fast-travel / site-to-site (MV-1) → transitions/PG section; vehicle/mount (MV-11) → N8.9

(out of the Ch2-5 slice window).

• character-engagement verb (the keyless-verb flag, IN-1) → the verb exists (C1.6 DONE, the

standalone 8th Composure node, D-COMBAT-TD R4); its binding = P2.3; its win-payoff (frees Rangku) =

the CF-10 defeat-axis brief. gather-at-water / recharge-as-checkpoint (Ch-2 first-water) →

P2.10/PG-5 (transitions/PG section).

---

SECTION 3 — WS-11 UI/UX & PRESENTATION LAYER

RECONCILED (Vol2): SHAPE-telegraph floor re-cited T99_Translation_Combat §2.6 (was CVD §2.6; cite-critic C2). U11.32 Chakra repair → cross-ref to AB-5 (single owner). The 11c INPUT cluster (U11.7–U11.13) rides the early INPUT wave (master W3), NOT the late UI wave (cohere H8) — W-MOVE/W-AI depend on those binds. HUD reads U11.1–U11.4 land in W2 FEEL (the CF-11 home); book screens/menus in W10.

WS-11 — UI/UX & PRESENTATION LAYER *(the playable-surface presentation layer; folds P2.4 + P2.5 + the dive UI/IN/MAP/PH/PF/ST/LOC clusters; ~100% BUILD-NOW)*

Why a workstream, not a P2 sub-row. The mappers found the playable surface is real underneath

and invisible on top: the ONLY HUD is HumanityDebugHUD (Canvas DrawText), the ONLY input path

is an 11-action IMC with 4 of 8 verbs keyless, and there is no map / pause / settings / rebinding /

subtitle / book / photo / loading surface anywhere in game Source/ (positive-controlled: dive-1

repoCritic grepped UMG · minimap · pause · photo-mode · loading = 0). P2.4/P2.5 named the HUD and

the loc-scaffold; §12 #26 bundled the rest as "no specs today." This WS is that bundle, specced and

itemised against the built runtime it renders (UQuestSubsystem Q3.3 DONE · GAS actor C1.1 DONE ·

WorldState T0.10 DONE · cue channel C1.16 DONE).

Tag law (binding, dive critics): UMG, damage numbers, hit markers, camera-shake reads, the

color-independent SHAPE telegraph channel (an L1 accessibility FLOOR, T99_Translation_Combat §2.6) are all BUILD-NOW —

only generative INFERENCE is 5090-gated. The only gated residue here is final HUD/icon art

(+5090-final) and localized non-English string/VO content (GATED-SCAN); every runtime, widget,

and placeholder ships now. The dive tag over-gates (CF-11/CF-1/CF-6/CF-7 tagged +5090-final) are

corrected to BUILD-NOW here.

Two disciplines this WS enforces by construction:

- Reveal-gating (HL_0028). Substrate-perception (vril-signature/polarity reads) unlocks only at

Adeptus-Minor tier-6 (Ch-15 onset). The ability book, Grimoire, and any perception HUD MUST filter

by tier/arc so no late-arc name, description, or tier-6 read surfaces for a Neophyte in Ch 2-5. A

screen that cannot hide is not shippable — the gating IS the feature.

- Empty-state by design. The journal (Q3.6) and Grimoire (Q3.8) are hard-anchored to Ch-13/38/57;

in the Ch 2-5 slice they ship as *empty shells that are correctly empty*, not as stubs to be filled

later. The Ch-3 healer's-child seed is the first thing the journal is allowed to write.

[RESCOPE RE-DERIVE — see §A.0, with U11.18: Ch 13 is now the slice's terminal node.]

11a. HUD / UMG core *(folds P2.4 · CF-11 · the debug-HUD replacement)*

*Verified-by:* image critics read the captured HUD frames (QA-6 loop, standing visual directive); a

functional test asserts every bound read tracks its WorldState/attribute source (no hard-coded HUD ints).

11b. Menus · flow · settings *(folds ST · AU-6 audio tab · PG-3 front-end surface)*

11c. Input · bindings · control-scheme UX *(folds P2.3 · IN-1..12 · MV-10)*

Authority is P2.3 + code — no T99_Translation_Input exists (the translation family holds

AssetGen/Audio/Cinematic/Combat/Terrain/UE5_Build, not Input/UI; U11.31 authors the UI member). The

IMC today maps 11 KBM actions, 4 of 8 combat verbs bound, bCanCrouch=true with no crouch key.

*Verified-by:* AI-playtest via InjectInputForAction exercises every new binding in PIE + packaged

(QA loop); a functional test asserts all 8 verbs and every screen open-key are reachable.

11d. The book screens · quest log · fork UI *(folds AB-* · Q3.6 · Q3.8 · the landed quest/fork panel)*

*Verified-by:* a reveal-discipline functional test asserts the ability book at a Neophyte state exposes

NO tier-6/late-arc row (HL_0028 teeth, mirrors the reveal-ladder gate); image critics read the empty

journal/Grimoire frames to confirm they are *correctly* empty, not broken.

11e. Map · compass · waypoint · travel UI *(folds MAP · MV-1)*

11f. Localization · subtitles · fonts *(folds P2.5 · LOC · CN-4)*

11g. Photo · codex-capture · loading UX *(folds PH · PF)*

11h. The teaching layer (FTUE) — implement the RULED spine design

**Teaching canon owners (recall; the teaching-onboarding recall agent errored, so these are cited

directly against the docs):** T1_Story_Spine [ACTIVE v1.0] # CHAPTER 2 (the mode_variety line owns

the tutorial beats: *"combat: T (tutorial bear forced-run)"*, fire-making Type-6, inscription-locked

Return/Backtrack, vril-mechanics introduction, vril-sight awakening as the arc's first-transformation

moment); docs/spine/CH_02.md (the diegetic Attunement design — *"None of this is spoken. It is felt,

through the hands, at the water"* L28; the first-recharge learning beat CH02_B04; the cold-air reveal

cue L63; attunement waking at the Neophyte vril-trace register, gated below tier-6 per HL_0028 L21);

T0_Hard_Lines HL_0028 (the capability gate the teaching must respect); CVD Pillar-8 + T99_Translation_Combat §2.6 (the

multi-channel telegraph, an L1 accessibility floor). Nothing here is invented — it is RULED spine

design with no UX spec; §12 #26 names "tutorial/onboarding" with zero coverage.

*Verified-by:* AI-playtest asserts the forced-run cannot be "won" and the traverse gate opens; image

critics confirm the Attunement cue renders diegetically (no tier-6 perception overlay); the reveal test

(U11.14) doubles as the HL_0028 teeth for the teaching layer.

11i. Contract doc + canon repairs

11j. Folded dive-item crosswalk (nothing lost; phantom-folds recorded)

11k. Wave rollup + verified-by (this WS in the re-ranked queue)

• Wave UI (the bulk — HUD, menus, input, book screens, loc, teaching-spec, T99_Translation_UI,

Chakra repair): U11.1–U11.20, U11.23–U11.25, U11.30–U11.32. *Verified-by:* every widget frame read

by image critics (QA-6); reveal-discipline + all-verbs-reachable + no-hard-coded-HUD-int functional

tests; AI-playtest drives all bindings in PIE + packaged; nightly-soak from this week.

• Wave FEEL/UI seam (the teaching layer, co-req the combat FEEL primitives): U11.27–U11.29.

*Verified-by:* forced-run-unwinnable + Attunement-diegetic + shape-telegraph-rendered assertions.

• Wave TRANS (map/travel/loading UX, co-req the cross-map transition backbone): U11.21, U11.22,

U11.26. *Verified-by:* a Flores→Bali map-change soak carries ws.* across and masks the hitch; the

map screen reads real Zone-Catalog site rows.

*Slice honesty:* the journal (U11.18) and Grimoire (U11.19) are empty in Ch 2-5 by canon, not by

under-build; the ability book (U11.14) is deliberately sparse (Neophyte, ~11 abilities across Ch2-5) [RESCOPE RE-DERIVE — see §A.0, with UW-2];

the map (U11.21) shows within-region roam, not cross-region fast-travel (Ch-55 Vimana). None of these

is a gap — each is the slice implementing the locked structure.

---

SECTION 4 — WS-3+ QUEST / PROGRESSION / TRANSITIONS / SAVE

RECONCILED (Vol2): CN-1..CN-9 → cross-refs to ws56_av (the single scene-runtime home; this section retains the quest-side binding view only). This section's PG-1/PG-6/CN-2/SL-2 are the backbone owner and land EARLY; PG-2's region-map handoff rides AFTER terrain (master W8) — ws4_regions RX.3 is its regions-side cross-ref (cohere H1). MG-1..4 are homed in master W8 REGIONS (cohere H9). D-DEFEAT-AXIS REC(a) == ws1_feel Alt-2 == the register's Composure-break-to-Withdrawal axis; Josh-pending.

WS-3+ ENHANCEMENT — Quest / Progression / Transitions / Save (the Ch 2-5 vertical-slice spine)

The built Q3.3 UQuestSubsystem executes ONE chapter (CH_02) as data (game bc356dc; produce-before-consume

DAG over DT_Beat, fork present/choose over DT_Fork, save/restore on the SaveGame seam, the 42-var

WorldState baked). The slice target is a four-chapter playthrough — Ch 2→3→4→5 across four region maps —

with cross-chapter progression, save UX, the unlock web, the integrity-band read, the scored-interaction

runtime, and dialogue/scene presentation all standing up over that seam. Nothing below is hardware-gated

except dialogue VO audio, final cinematic renders, and integrity-appearance final meshes (the DR handoff, §E).

CANON-FIRST: every item carries a canon citation or a positive-controlled absence-proof (grep=0 in game

Source/). The dive's PG/CN gap ids are carried as the item ids; new runtime clusters use MG-/UW-/IB-/SL-.

The one binding correction this section executes

There is NO free-travel gap. MV-1 ("173 km island crossed only by a DEV M-teleport; no world-map/fast-travel")

is a phantom for the slice: free cross-region travel is Vimana-gated at Ch 55 (WGM §7 L764 "Vimana enables

free travel between all locations visited"; Vimana *acquired* Ch 55, WGM L85/L985/L997). Ch 2-5 is deep pre-Vimana

ground-travel arc (WGM §11 "Travel Methods by Arc" L989-999: Act 1-2 = walking/riding/river+coastal sailing). The

slice therefore implements ruled directional travel-beats forward + free-roam within the loaded region, never a

fast-travel menu. Intra-region locomotion (the 150 km site spacing, climb/swim/vault) is MV's scope; this section

owns only the chapter-advance + region-map handoff along the ruled onward line (PG-2).

---

A. The item table

*Cross-section refs (owned elsewhere, wired here):* PG-11 difficulty-as-UX → C1.11 (combat/feel); PG-12 inventory-minimum → P2.6/economy (Ch 2 chapter_anchored_weapon_add=none, slice-LOW); Journal → Q3.6; Grimoire → Q3.8; input binds (crouch/interact/wheel/lock-on/skip) → P2.3; the qualitative-delta→integer map IB-1/CN-3 read is decision #11.

---

B. Wave queue (re-ranked for the 2-3 week window; each wave has its verified-by clause)

Nightly-soak (starts THIS week, current machine): enroll the CH02→05 SliceWalk + QT-7 + every scored-beat

Bronze run + a save→transition→Continue round-trip as a nightly autonomous soak (plan §11.5 QA loop; the machine is

the tester). Any red beat/fork/reachability failure files a typed bug record repro-first.

---

C. Per-chapter content-budget + test-coverage scorecard (this section's slice of the density curve)

The density model is a curve (directive 6): Ch 2-5 = early-game, deliberately lighter, seed-rich. This scorecard

makes the progression/quest budget measurable per chapter — counts read directly from the spine CSVs + slice map.

Scorecard law: every durable write in all four chapters is grep-verified ∈ the baked 42 (no new var); every fork

path in the QT-7 column must reach the chapter terminal (softlock-immune); every scored beat is Bronze-reachable and

non-gating. Early chapters carry FEWER scored beats by design — the curve steepens post-slice (full game ≈100-200 h main /

1-2 k h side+weave), so the slice validates the structure, not the volume.

---

D. Open decisions carried as briefs (recommendation + options — never bare)

• D-DEFEAT-AXIS (boss HP=0 defeat model) — blocks SL-1/CN-7 presentation polish. TODO_CANON (game) L199-206:

the Composure/Withdrawal fight consumes no HP=0, so the boss neither dies nor staggers. *Options:* (a) Health is a

non-lethal pressure track, defeat is always Composure-break→Withdrawal; (b) HP=0 opens a **stagger/execution

window that accelerates Withdrawal; (c) per-boss win_condition axis (some HP, some Composure). REC: (a)** for the

Pillar-8 slice — it keeps "smarter not stronger" honest and the greybox needs only a defeat shell; the axis choice

is a Phase-5M tuning ruling, not a slice blocker. *Strongest objection:* a pure-pressure boss reads as un-killable to an

action-RPG player expecting an HP payoff — mitigated by the Composure-break MOMENT (CF cluster) as the visible kill-beat.

• Qualitative-delta → integer map (#11) — IB-1 + CN-3 need the per-class integer weights for ws.integrity_score

fork writes (band phrases like +substantial-generative). The band map exists in T1_Integrity_Paths §3.2; this is a

data-import mechanic, not new canon. REC: derive from the §3.2 band midpoints (the illustrative map already banked

in docs/INTEGRITY_DELTA_MAP.md), under the tuning firewall.

• NG+/prestige carry-over (#28) — out of slice; note only (mastery/familiars/world-state/un-pruned seventh vril-seat).

E. 5090 handoff touchpoints for this section (the DR drop-in contracts)

Only the generated assets ride the 5090; every RUNTIME/selection/scene seam above is BUILD-NOW and consumes a

placeholder now, the generated asset later, via a stable id (the DR contract):

• Dialogue VO (Q3.2/CN-4) — the selection engine + subtitle box build now against silent/placeholder SoundWave;

Kokoro/ElevenLabs VO drops in on V_<voice_id>__<dialogue_line_id> (A6.6). Gen 5090/service-gated.

• Scene final cinematics (CN-1/CN-7/CN-8) — the lightweight beat-triggered runtime + greybox cine cuts build now;

MetaHuman/MRQ/final-render assets swap in behind the same scene_anchor_id. Gen 5090-gated.

• Integrity-appearance meshes (IB-1) — SelectAppearanceVariant(score,ageStage,pathLock) returns a variant id now

(P2.12); the render layer consumes the final gated mesh later. Selection ungated; mesh gen 5090-gated.

• Everything else — the DAG, transitions, save, unlock gates, band derivation, scoring spine, softlock autotest — has

zero hardware dependency and ships on the current machine.

---

SECTION 5 — WS-11 LIVING-WORLD SUBSTRATE & RUNTIME

RECONCILED (Vol2): LW-6 SFX-registry cite corrected to 12 rows (cite C3); LW-9 DOC_MAP line corrected to L127 (cite C4). AB-5 is the SINGLE Chakra-repair owner (CR-CHAKRA / U11.32 are its cross-refs). LW-4 (era-swap runtime) folds into ws4_regions RX.1 (the Ch-4 chapter-integration owner); LW-3 (Pawukon clock) + LW-12 (ws keys) stay here as the primitives RX.1 depends on (cohere H5). §11.6.4 is the AUTHORITATIVE pending-Josh stance for D-ECONOMY-MEDIUM — EC-medium/EC-vendor BLOCKED-ON-BRIEF; EC-inventory/EC-treasure proceed.

11.6 WS-11 — Living-World Substrate & Runtime *(new workstream; the Ch 2-5 slice's living skin)*

Josh's directive-3: the slice must carry the living-world substrate grounded in each region's real biota/geology — fauna, herbs, mineral veins, trades, the ability book, unnamed members. Dive-2 mapped six domains (FA/FL/MN/TR/LW + EC) over the four slice regions (flores_island Ch2-3, bali Ch4, sumatra_java+angkor_khmer Ch5). This WS is the honest split of that map: the data that is CANON and needs a structured home now vs the engine that makes it observable in PIE. It is CANON-FIRST per directive-4 — every row cites the owning page/tier-doc or carries a positive-controlled absence proof; nothing is re-invented (the completeness critic already caught the mis-graded "missing" registries: T0_Creature_Roster HAS 132 rows, T0_Boss_Encounter HAS 16 speccing the slice bosses, Cross_Cultural_Link_Graph HAS 184 incl. LINK_0186 — consume, never re-author).

The gating law (binding): only generative INFERENCE is 5090-gated. Cited rows, struct regen, deterministic spawners, gather/drop dispatch, day/night/weather/calendar drivers, placement wiring = BUILD-NOW. Generative skins (creature meshes, PCG density, pharmacology curves, atmos) ride the 5090. The RNG number magnitudes ride the tuning firewall (Phase-5M audit); the SHAPES and BINDINGS build now on illustrative constants.

The two load-bearing canon spines this WS wires to:

• The loot/yield/spawn spine — the Sigma constraint. HL_0061 / CVD §17.13 RNG hard line / T1_RNG_Sigma_Constraint_System: all RNG-based drops constrained to ±1σ — anti-streak FLOOR at −1σ (cannot drop before), mandatory CEILING at +1σ (must drop after), resets on drop; "foundational across drops, spawn rates, and RNG events." Engine: UHumanitySigmaRNG/FSigmaConstraintState are BUILT (game 1dabf50); P2.13 adds UHumanityDropComponent over DT_RngDropTable (0 rows today — author first). The integrity layer is IPW (T1_Integrity_Paths_Worldstates_Master §13, the mirrored 5-band per T1_Combat_System_Spec §-damage): the integrity_band_field modifies drop/exchange the same way it modifies damage — PLAYER-only, no enemy-stat path.
• The living-ecology spine — RULED. docs/RUNTIME_GENERATIVE_LAYER.md §"Living ecology + the immune set": over-gather/over-hunt one area → resources/fauna MIGRATE → the area REGROWS (tunable per-resource); HARD RULE — anything gating an achievement/progression/100% is tagged progression_critical on the canon graph = IMMUNE (never depletes/migrates/vanishes). The sim can never produce a non-progressable state. Rift/ambient spawns obey CVD Pillar 1 (Random Encounter Policy: Hybrid, Gate 1) — the random layer is confined to exactly three declared systems (Aggro Magnet · Rift Instability Zones · Antagonist Hunter Triggers); anything outside = pillar violation. The runtime LLM never spawns (deterministic-first).

---

11.6.1 CANON-AUTHOR-NOW lane *(canon-side; runs parallel to the trunk NOW; no engine dependency; each lands proposal-tier → normalize-doc critic → fidelity re-baseline)*

Tag key: CA = CANON-AUTHOR-NOW. Wave W-LW1 unless noted.

---

11.6.2 BUILD-NOW engine lane *(game-side; the substrate made observable in PIE; folds onto existing WS-2/WS-4/WS-8 seams — no new engine home invented)*

Tag key: BN = BUILD-NOW · BN+5090 = placeholder now, generative-final rides the 5090.

---

11.6.3 FL-1 — the flora-registry fork *(director-delegated; ruled and APPLIED, path shown per CLAUDE.md autonomy — this is EXECUTION-tier schema/tooling, not a Josh flag)*

The fork. Give gather-input flora a structured home as either (A) a NEW T0_Flora_Registry, or (B) an overload of T0_Creature_Roster with species_class=flora (the deep-book precedent — the creature schema already carries a species_class discriminator).

The steelman for B (species_class precedent). One registry, one struct, one spawner already exists (132 rows, 52 cols); the deep-book contract could route plant seeds through the same entity_kind machinery; flora and fauna share a spawn/placement/region-native shape, so FL-3/FL-4 node-placement could reuse N8.7 wholesale.

Why A wins (the grounded pick — APPLIED). The creature schema is tameable/familiar/husbandry-shaped: spawn_rate, tier_classification, bonding-window, tameable, breeding, and the 8 mesh_/generation_ cols (FA-2). Gather-input flora needs a disjoint field set — part, prep, mode, effect, active_compound, poison_class, dose (FL-2/FL-7) — none of which exist on the creature row. Overloading would leave ~30 creature cols null on every flora row AND force husbandry semantics (taming/breeding a plant) onto the discriminator, which the completeness pass would read as schema rot. A purpose-built ~14-col flora schema is the cleaner diff, and the gen_row_struct.py pipeline makes a second USTRUCT free (FL-12) — the precedent that matters is the *pipeline* (one USTRUCT per registry), not the *table*. Ruling: mint T0_Flora_Registry (path A). Consequence: FL-8 amends the deep-read contract to add a flora entity_kind → the new registry (not to Creature_Roster), and the ecology sim (LW-ECO) treats flora nodes and fauna as parallel simulatable classes over their own tables.

---

11.6.4 D-ECONOMY-MEDIUM — the transaction-medium brief *(carried as a BRIEF; recommendation + options, never bare)*

The gap. The trade MODEL is canon — T1_Trades_Master §9.2 (player-driven economy rests on trade interdependence; scarcity is real because materials are region-specific and require physical presence) + §9.1 (spellcraft and consumables are two separate economies) — but the medium (what the player pays with, how a vendor prices, where L2 treasure sits) is undefined. The slice needs *some* exchange the moment the player has gathered materials + drops with nowhere to spend them.

The options.

• Opt-1 (RECOMMENDED) — historically-real mixed exchange per region, NO universal coin early. Barter + region-appropriate proto-currency (rice/cloth/metal-weight where the real 1200CE region used it); vendors price in the local medium; L2 treasure is σ-bounded drops (HL_0061) placed by RngDropTable + gather-node yield. *Pros:* fits §9.2 regional scarcity + §9.1 two-economies + immersive-teaching (the economy IS a culture lesson); no anachronistic gold-standard; extends cleanly to the Tier-C MMO cross-player economy §9.2 names. *Cons:* more per-region authoring; needs a clean "value" abstraction so cross-region trade resolves.
• Opt-2 — single abstract "value" unit (universal coin) now, reskin per region later. *Pros:* fastest to build; one vendor UI. *Cons:* contradicts §9.2 "scarcity is real / physical presence" and the immersive-teaching pillar; a universal early coin is exactly the ahistorical flattening the mandate rejects.
• Opt-3 — no vendor economy in the slice; gather/craft/use only. *Pros:* zero new system. *Cons:* leaves gathered materials + drops with no sink; the slice reads incomplete against the "100% complete structure" target.

Recommendation direction (per the pack): Opt-1 — historically-real mixed exchange per region, no universal coin early. Land it as: a medium_of_exchange token on Region_Index (region → accepted media), vendor mechanics as a P2.6 inventory-component consumer (barter table + local-medium pricing), and L2 treasure placement = the σ-drop spine (LW-WIRE) + gather yields. This is proposal-tier pending Josh (economy medium is a cardinal-canon surface: it sets a world rule) — brief presented, not applied.

Economy items (EC) the brief spans *(the EC lean-recovery file returned empty — positive-control: tasks/ac6bd01d4145bf2c6.output and …accfbed6289b3e6b5.output both exist-but-empty; economy is grounded from the D-ECONOMY-MEDIUM brief + §9.1/§9.2 + P2.6, NOT re-derived):*

• EC-inventory — the material/drop sink already has a home: P2.6 UHumanityInventoryComponent + drop-table pickup (in-plan). Gathered flora/ore/hide + σ-drops land here. BN, W-LW2.
• EC-medium — the medium_of_exchange region token + value abstraction (Opt-1). BRIEF→CA on ruling, gated on Josh.
• EC-vendor — trade-NPC barter/pricing mechanic over EC-medium; folds onto N8.2 persona runtime. BN+5090, W-LW3, gated on EC-medium.
• EC-treasure — L2 σ-bounded treasure placement = LW-WIRE + gather yields; no separate system. BN, W-LW2.

---

11.6.5 Trade substrate build order + the deferrals *(from the trades map)*

The slice is entirely pre-first-trade-home (earliest ancient home is Ch7, Blacksmith+Herbalism, S.India — T1_Trades_Master) — [RESCOPE RE-DERIVE — see §A.0: Ch 7 is inside the ruled PROLOGUE→CH 13 slice, so this sentence's premise is invalidated; the order below re-derives against the ruled scope]; every trade Ch2-5 touches is *substrate seeding*, not a mastery climb. The trades-map recommended structuring/introduction order for the pre-5090 window:

Tailoring > Engineering > Blacksmith > Herbalism > Mining > Inscription — with Alchemy + Culinary DEFERRED (they are consumer trades: Alchemy requires Herbalism botanical inputs, Culinary requires the vril-food substrate-perception unlock — both HL_0082 Apothecary/Atelier compound trades that need their feeder trade structured first).

This order drives which trade rows/gates (TR-2/TR-3/TR-6) and which compound wiring (MN-11 Forge, FL-11 Apothecary) get authored first. It is a build-priority recommendation, not a canon ranking — the 10-trade/12-tier/120-gate numerical canon is untouched (185 abilities, 10 trades, 22 familiar slots all intact; no item mints an nth+1).

---

11.6.6 Living-World wave rollout *(each wave with its verified-by clause; slots into the global re-ranked queue as the "living-world" wave, W-LW2 the engine core, W-LW1 parallel in the canon lane, W-LW3 folding into the Ch3-5 region + transitions waves)*

---

11.6.7 Binding-correction & adjacency notes *(apply; do not re-litigate)*

• Anti-phantom kills (from the dive-2 critics): LINK_0186 (Bali leyak → Khmer krasue/ap) is already present in the live Cross_Cultural_Link_Graph (row 184) — the raw AB-13 "grep=0" was a cannot-match miss; do not re-author (AB-13 belongs to the ability-book WS regardless; noted here only as the phantom-kill). The slice bosses BE_0002..BE_0005 EXIST as canon (16-row Boss_Encounter) — consume. Creature_Roster (132) + RngDropTable both EXIST (the latter empty, not missing).
• Tag-law fixes applied: RngDropTable row-authoring (FA-5/FL-5/MN-4) is CANON-AUTHOR-NOW; the σ-drop engine wire is the separate BUILD item (LW-WIRE). MN-9 (declare the biota/mineral/geology deep-read lane) re-tagged CANON-AUTHOR-NOW (a contract amendment, same class as FL-8).
• Reveal discipline: the ambient-fauna/flora/festival surface carries no late-arc leak; AB-5 is the only ability-facing item here and it is a rename-in-place under HL_0028/reveal-timing (no name/description of a below-tier or late-arc ability surfaces early).
• Adjacent consumers owned by other workstreams (this WS wires the substrate; it does NOT re-own these): familiar-bonding slice surface → N8.6 (T0_Familiar_Registry 22 rows carry CH_02-05 bonding windows: Shell→Kira→Khan across flores/bali/sumatra — the slice's live familiar arc); integrity-path + reputation play surface → N8.4/P2.9 over T0_Reputation_Ladder (companion-loyalty + assimilation choice are slice canon); language/script in-world surface → Q3.7 InscriptionDecoder over T0_Language_Script_Registry (84 rows). The living-world persona/relational-web tier (LW-9/LW-10) is the shared read these consume.

---

SECTION 6 — WS-4+ THE CH 3-5 REGION ROLLOUT

RECONCILED (Vol2): RE.5 cite corrected — HL_0003 only (HL_0032 dropped; it is the 12-side-questline lock, cite C1). RX.1 owns the Ch-4 era-swap chapter mechanic (folds LW-4); its deps now explicitly include LW-3 + LW-12 + the LW-4 fold (cohere H5). RX.3 → cross-ref to PG-1/PG-6 (the ws3_prog cross-map backbone is the single owner; RX.3 is the regions-side verified-by, cohere H1). The 9 RB.1 bbox fills are authored in LANE-CANON.

WS-4+ — The Ch 3-5 Region Rollout *(the CH_02 pipeline generalized to 3 regions / 4 maps; terrain builds are the long poles)*

Ch 2 is the built reference: FloresIsland.umap imported (W4.1, wave D) + dressed + CH_02 staged, BE_0002 StateTree compiled. This section rolls that ONE proven pipeline across the rest of the slice — CH_03 → flores_island (terrain ALREADY BUILT), CH_04 → bali, CH_05 → sumatra_java + angkor_khmer (excursion) — by generalizing three flores-hardcoded drivers, running the deterministic GIS chain against the three un-built regions, filling 9 cited zone bboxes, and CONSUMING the 16 already-authored slice boss rows (BE_0002..BE_0005 in T0_Boss_Encounter_Registry, verified — never re-author). Nothing here needs a generative model: DEM/GDAL fetch, Landscape import, weightmap paint, StateTree build, and data staging are all deterministic. The long poles are the three terrain builds (bali · sumatra_java · angkor_khmer) — network-bound GIS fetches that parallelize, gating only the STAGING of their maps, not each other and not the CH_03 work.

Density register (Josh's curve): Ch 2-5 = early-game, deliberately lighter, seed-rich. The per-chapter budget the rollout stages is the OPENING band — ~7-9 sites, 12 beats, 4-5 encounters, 1-2 forks each — not the completionist density of mid/late game. The rollout builds the STRUCTURE at max honest pre-5090 polish (greybox geometry, real geo-placement, real StateTrees, real forks); the deep-read weave/culture/side-quest enrichment and all generated finish (VFX/SFX/music/MetaHuman) ride the post-5090 pass.

Chapter → region → map (verified: spine headers + Zone Catalogs + Saved/Terrain/)

Positive controls (re-verified this pass, both repos read-only): Saved/Terrain/ holds only flores_island_* → bali/sumatra_java/angkor_khmer confirmed un-built. The 9 empty zone_bounds_wgs84 cells read directly from build/zones/{slug}.csv (flores CH_03: compang_lingko_fields, wae_water_spring, mama_klaras_waterfall, todo_royal_village, lingko_forest_edge; bali: village_balian_ward, wayang_performance_ground; angkor_khmer: kra_traverse, bayon_taprohm). T0_Boss_Encounter_Registry has exactly 16 rows CH_02-05 (BE_0002 ×3, BE_0003 ×4, BE_0004 ×4, BE_0005 ×5) — the slice bosses EXIST as canon.

---

J.1 — Generalize the three flores-hardcoded drivers *(one pass unblocks bali + sumatra_java + angkor_khmer staging)*

The terrain-derive (derive_region_bbox.py --slug) and GIS-fetch (terrain_fetch_preprocess.py --params) drivers are ALREADY region-agnostic (verified: both take a slug/params, derive FAILS LOUD on zero anchored zones). The StateTree builder (build_boss_statetree.py --boss) is ALREADY boss-parameterized. The three that hard-code flores_island/FloresIsland/CH_02/BE_0002 are the import, dress, and stage drivers — generalize each to --slug/--chapter, deriving the boss(es) from DT_BossEncounter by chapter_id, and every terrain number from Saved/Terrain/<slug>_terrain_meta.json (the single-source-of-truth pattern is already in-code).

Verified-by: run all three against --slug flores_island --chapter CH_02 and assert byte-identical placement to the built FloresIsland.umap (the generalization must reproduce the reference exactly before any new region runs). QA capture leg reused (-HumanityTerrainShot=), image-critic on the frame.

---

J.2 — The 9 cited bbox fills *(the rider pattern: apply_zone_catalog_columns.py, all-or-nothing, --emit-baseline same commit)*

Fill the 9 empty zone_bounds_wgs84 cells so the staging step resolves 7/7 · 7/7 · 9/9. This is canon-side execution (region geometry within the locked structure — apply and show the path, no Josh flag): real named places get a gazetteer geo-citation; composite/named-in-canon sites get a placement sub-box within a cited real locus, explicitly flagged. Landed via the T0.1 single-writer as one rider per region page + its registry mirror; refresh fidelity_baseline.json in the SAME commit (the standing baseline-shift rule).

Absence-proof discipline: RB.1c/d/e/f/g are placements (no canonical point exists) → each ships flagged in TODO_CANON with the cited real locus it sits inside, exactly as CH_02's two un-bbox'd sites were flagged. Verified-by: check_region_page.py + a re-run of stage_region_chapter.py resolving 7/7 · 7/7 · 9/9; the fidelity gate "FAIL set" grows by exactly these 9 cells + baseline refresh.

---

K/L — The three region terrain chains *(GIS fetch → import → dress → stage; the long poles; deterministic, parallelizable)*

Each region terrain derives its region_bbox from the FILLED zones alone (union + pad) — so the un-built regions build even with empty zones (bali 5/7, sumatra_java 4/4, angkor_khmer 3/5 all derive clean, verified). The bbox fills (J.2) gate STAGING of those specific sites, NOT the terrain build. All three GIS fetches can run concurrently the moment RG.1-3 land — they are network/CPU-bound, no GPU, no cross-dependency.

Verified-by (per region): derive_region_bbox emits a clean region_bbox (fails loud if not); terrain_fetch_preprocess emits absolute-metre heightmap + 7 WorldCover weightmaps at valid vertex size; import_region_landscape read-back asserts component count = ⌈X/quad⌉×⌈Y/quad⌉ and bounds match the meta JSON (the FloresIsland 132-component round-trip pattern); a dressed-coastline capture goes to Josh (frame-claim honesty: numbers stay honest, the sea/rake fix makes the frame TRUE before it ships). W4.3 hardening (absolute-metre encoding + void-fill fallback) is a hard prerequisite — un-hardened it manufactures false craters/tile-seams across the multi-tile SEA regions.

---

F′ — W4.8 biome material, generalized + applied per region *(the F-wave fold — currently flores residue)*

W4.8 (weightmap-driven auto-material + PCG biome binding) does NOT yet exist in the game repo (absence-proof: no M_Landscape_Auto, no weightmap-paint driver — the 7 flores WorldCover maps are imported as LayerInfo DATA only, per the 2026-07-24c ledger; the paint pass is flagged residue). Generalize it once, apply per region as each terrain lands.

Verified-by: a lit oblique capture per region shows painted biomes (not a grey/white slab), image-critic PASS; weightmaps confirmed pixel-aligned to the elevation grid (the W4.5 same-AOI resample). PCG population + Nanite/HLOD batch bake is the 5090-gated tail; the material/paint harness is not.

---

Per-chapter STAGING + ENCOUNTER waves

Staging = stage_region_chapter.py places site markers + PlayerStart + boss(es); encounters = build_boss_statetree.py --boss <id> per slice boss (an EXISTING parameterized tool — the roster is consumed, not authored). The new work is the win-condition verb BEHAVIORS beyond DODGE/STRIKE (the care-doctrine appease/balance/endure seams), shared with C1.6 (ability set) + C1.13 (environmental-boss protocol) + the HP-0 defeat-axis ruling (open in TODO_CANON, blocks presentation polish but not the greybox StateTree).

---

RX — The chapter-specific mechanics + the transitions backbone

Verified-by (RX.3): a scripted CH_02→CH_03→CH_04→CH_05 walk asserts every handoff key survives the level change (Shell bonded at Ch3-in, +Kira at Ch4, +Khan at Ch5; community marks accumulate; integrity band carries; octopus/Sanur familiar stays deferred/gated). RX.1 verified-by: the stratum swap is reversible and writes no new ws var. RX.2 verified-by: the excursion doubles back to the anchor with the anchor's beat-state intact (no re-open); FORK_CH05_A/B resolve on the anchor return (integrity-gated lethal route offered on the two corrupt operatives ONLY — datu + court official — never the communities, per CH_05.md L276).

---

Wave assignment — the 2-3-week window (terrain builds as the long poles)

The rollout parallelizes hard: the generalization pass + data fills land day-1, the three GIS fetches kick off immediately after and run concurrently in the background (network-bound), and CH_03 (built map) completes end-to-end while bali/sumatra_java/angkor_khmer terrain bakes.

Long-pole note: the three terrain builds are the schedule spine, but they are network/CPU-bound deterministic fetches that overlap fully — start all three the moment J's RG.1-3 land; K and L are gated on their terrain finishing, not on each other. The AI QA loop (WS-10) wraps every wave: capture → image-critic → functional/regression → the repro-first bug ledger; Josh's feel-pass is the slice playthrough, never bug-finding.

Care/reveal riders carried forward (already in the spine, verified — the rollout must not leak them): no House-of-Velheim name/family/pattern player-facing across CH_03/04/05 (the family first appears Ch 8; the datu's FAMILY_BRIEFED awareness reads only as unnamed unease); all perception at the Neophyte register (inscription/reverence/lore tells, never a tier-6 vril read — HL_0028); the Khmer Rouge modern trauma held OUT of the ~1200 CE era (standing high-care flag, never spectacle); the CH_05 lethal fork falls only on the two corrupt operatives.

---

SECTION 7 — WS-5/6 AUDIO FRAMEWORK + CINEMATIC EVENTS (the AV layer)

RECONCILED (Vol2): ws56_av is the SINGLE home for CN-1..CN-9 (ws3_prog references) and for AU-10 (ws1_feel references). The audio spine AU-8→AU-1→AU-2 rides master W2 FEEL (early), not W7 — so W2 hit-confirm audio (AU-10) has its spine (cohere H2). CN-1..8 are the master W7 SLICE-PRESENTATION wave; CN-9 parked (Josh #22). AU-13 correctly cites the SFX registry at 12 rows.

WS-5/6 ENHANCEMENT — Audio Framework + Cinematic Events (the AV presentation layer)

Dive-1's AU/CN maps found the same shape twice: **the channels + data are built, the feel + the

consumer are not.** SFX ride 3 hardcoded PlaySoundAtLocation calls that bypass the C1.16 cue

channel entirely; the island is silent; DT_SceneAnchor's 439 imported rows have zero consumer

and Build.cs carries no LevelSequence/MovieScene/MetaSound deps. WS-6 §8 (A6.1–A6.8) and

WS-5 §7 (CN5.1–CN5.9) scoped the *generation* pipeline; this block adds the greybox-tier runtime

that plays a scene and makes a hit audible NOW — the part Josh's Ch2-5 slice needs before any 5090

asset exists. TAG LAW (binding critic correction): only generative *inference* is 5090-gated —

the MetaSound graphs, the SoundClass mix, the scene-presentation subsystem, letterbox, cine-camera,

footstep-by-surface routing are all deterministic BUILD-NOW; only the final generated

WAV/VO/Niagara *asset* drops in later. Every +5090-final below gates the asset, never the seam.

Care: cinematic + audio honor hard_line_relevance / reveal: HELD (the Ch2 polo is a materialized

superstition, never a dukun; no Grand-Sage/Cassius thematic leak in any cue or scene) [SRC:

T99_Translation_Cinematic §2.4, Step 1; T99_Translation_Audio §2.7; HL_0046 gate #25 grand_sage_silence].

Item-ids carry the dive prefix (AU-/CN-); the Fold column names the §8/§7 plan item each expands

(never a net-new duplicate) or marks NEW. +DA = generated-asset drop-in is GATED-5090; the runtime is BUILD-NOW.

---

WS-6 — Audio Framework (the AU items)

The whole lane hangs off one routing decision: today cue→sound is two disconnected paths (the

C1.16 GameplayCue channel fires notify-stubs while placeholder SFX play in PARALLEL via hardcoded

calls — dive AU-2). AU-1 + AU-2 + AU-8 are the spine: build the subsystem, register the cue

manager, link the module. Every other AU item routes through them.

---

WS-5 — Cinematic Events (the CN items — a LIGHTWEIGHT scene-anchor runtime)

WS-5 §7 (CN5.1–CN5.9) is the FULL generative Sequencer/MetaHuman/MRQ/dispatch pipeline. Josh's slice

wants the greybox tier: play a scene by scene_anchor_id, cut a camera, frame a dialogue box,

render the Ch2 fork AS a scene, letterbox, skip. Dive-1's recommendation stands: build a

runtime presentation subsystem, NOT full LevelSequence assets — the 439 DT_SceneAnchor rows

become playable now; the MRQ/MetaHuman half parks on the 5090. CN-1 is the spine every other CN hangs

off. T99_Translation_Cinematic is the authority for the eventual full pipeline; these items are its

platform-runtime *greybox* realization.

---

The QA audio-capture tie-in (QT-3/QT-4 → WS-10)

The WS-10 loop (§11.5) verifies by screenshot — UHumanityQACaptureSubsystem captures named shots

• a scene-luma floor + auto-exit. It has no audio-side assertion, so the machine cannot catch the

two defects this block fixes (the silent island AU-4, the bypassed cue channel AU-1/AU-2) — a black

frame is caught, a silent frame is not.

---

The 5090 drop-in handshake (the AV generation lanes — DR-style contracts)

Per §0 + §13, all AV generation is GATED-5090; the runtime seams above are the pre-wire so a

generated asset drops in with zero runtime edits. The deliverable NOW is the id + path contract

each lane fills — the drop-in point is already named in the T99 docs. Do NOT re-plan these as buildable;

plan them as the contracts the placeholders honor.

---

Wave assignment + verified-by

Re-ranked into the 2-3-week window; the AV items thread through the new waves (feel / UI / QA-org /

slice-presentation) parallel to the combat-feel and Ch3-5-region waves.

---

SECTION 8 — WS-10 AI QA / ITERATION LOOP

RECONCILED (Vol2): the QT-1..22 org IS LANE-SOAK in the master queue (nightly, this week). QT-AU (audio-capture leg) is owned by ws56_av. Every QT harness folds into its content wave's verified-by clause — no separate build wave; it is the definition-of-done handed to every wave.

11.5 WS-10 — AI QA / Iteration Loop *(EXPANDED 2026-07-24: the QT-1..22 autonomous test organization; the loop that wraps EVERY workstream)*

The full architecture is docs/proposals/AI_QA_LOOP_ARCHITECTURE.md (PROPOSAL-TIER, critic-fixed; ENGINE-class) — that doc OWNS the role/item table (QA-1..QA-14) and the per-pipeline hook map (§3). This section is the execution organization: the QT-1..22 map turns the four AI roles into a standing test ORG — the four senses, seven bug-class harnesses, the three-tier cadence with a nightly soak that starts THIS WEEK on the current machine, auto-triage, the beta player, the 5090 fleet contract, the coverage-scorecard family (the density CURVE made measurable), and the honest human-only list. Every QT item folds into a QA-x id (extends it, never re-mints it) and inherits QA-8's self-testing-gate law + QA-9's repro-first discipline. It invents no canon; it is ENGINE infrastructure sitting below the pillars, consuming CANON (hard-line floor, reveal windows, §17.1 collective-care, the reference bars, the per-chapter content budget) strictly as data. DOC_MAP row rides on integration (consumer = the QA-org wave + every workstream's verified-by clause).

LIVE TODAY (2026-07-23e; game f25d8a3+, canon 0f365c5+): QA-1/2/5/6 DONE — the loop ran end-to-end on the C1.18 greybox rig and caught five real bugs day one; Tools/run_qa_pass.py reads verdicts from the TEST REPORT never the exit code (self-test 9/9); the typed bug ledger is at 11 records under QA/bugs/; QA-8 teeth and QA-9 repro-first are already operating (every fixed record carries a regression_test_ref). Ground truth the org is built on: the rig captures 8 named shots (greybox_spawn, _hud_wellspring, _hud_forgotten, _boss_closeup, _quest_panel, _direct_broken, _phase_p2, _feel_swing) and has ALREADY hit a soak-class failure — a Automation RunTests hang, TESTS: TIMEOUT after 900s / tests_exit=124, capture-leg green — which is exactly what the nightly heartbeat (QT-13) and auto-triage (QT-14) exist to catch.

11.5.1 The four senses — how the machine perceives the build

The beta player perceives through four channels; its hands (input injection, QT-15) are the actuator, not a sense. Each sense already has a live seam; the org completes and generalizes it.

• SIGHT — rendered frames. FScreenshotRequest/HighResShot capture (QA-1) + image-reading critics (QA-6). LIVE on the greybox rig (8 shots). No play-term claim ships without a backing frame (the standing visual-verification directive).
• STATE-SENSE (proprioception) — the game's own internal readout: UCombatTelemetrySubsystem JSON (Saved/Telemetry/<s>.json), UWorldStateSubsystem reads (42 ws.* vars), and the DebugHUD enemy-stat hash that makes zero-enemy-stat-delta VISIBLE. Scored by score_greybox.py (SHAPES/RELATIONS/FLOORS only). LIVE.
• HEARING — the audio event stream: every cue_id/sfx_id_ref that should fire on its trigger. Today needs-build (WS-6 §3 = needs-build; no cue→trigger matrix in harness/). Folds into QT-10.
• FAULT-SENSE — the log/crash/assert stream: Saved/Logs/*.log, Saved/Crashes/**/CrashContext.runtime-xml, ensure()/check() lines, and the automation-timeout signal. Partially live (run_qa_pass reads report markers); auto-triage (QT-14) completes it.

11.5.2 The seven bug-class harnesses (each DT-driven, each QA-8-toothed, each QA-9 repro-first)

Not one hand-authored test per surface — one harness per defect class, data-driven off the DataTables so it scales to every row without re-authoring. Each ships a must-fire fixture (a known-bad row it MUST flag) and a must-not-fire fixture (a known-good row it must pass) per QA-8, refuses to run if its own teeth break, and lands its repro BEFORE any fix per QA-9. Every listed bug class is grounded in a defect the loop already caught or a §3 hook row.

The care/reveal harness (QT-9) enforces the §17 hard-line FLOOR mechanically — the one absolute the machine judges on its own. Everything ABOVE that floor it does NOT scrub: it opens an enhancer intake (QT-20/QA-13) for a RENAME/REFRAME, never a removal or a rated-G defang (care doctrine, CVD §17.1, delegated-bolder calibration — Josh is the backstop).

11.5.3 The three-tier cadence + the nightly soak (starts THIS WEEK)

The beta-player fidelity ladder is a three-rung gate — a rung never skips the one below it (scripted gates agent, agent gates soak).

• Tier 1 — scripted (BUILD-NOW, live): the QT-5..11 harnesses run headless via Automation RunTests Humanity.* + -TestExit; deterministic pass/fail through RequestExitWithStatus.
• Tier 2 — agent (BUILD-NOW): the AI beta player (QT-15) — screenshot → decide → InjectInputForAction (packaged-valid; EnhancedInput is Runtime). Exploratory input agent-chosen; regression input scripted.
• Tier 3 — soak: the nightly on the current machine (QT-13, BUILD-NOW) scaling to the 5090 Gauntlet fleet (QT-16, GATED-5090).

SOAK SCOPE UNDER THE RESCOPE (2026-07-27). The soak matrix grows with the slice — 15 nodes, not 4 —

and it is the one leg that does NOT fan out: it is wall-clock, and a 3-4× larger matrix costs 3-4× the

wall-clock at the same instance count. The QT-16 Gauntlet fleet is exactly what the 5090's parallelism buys

here, and it is the one item where the hardware buys THROUGHPUT rather than capability. Two riders the

build order forces: (1) the companion/ally runtime debuts in Ch 13, the slice's TERMINAL chapter, so it

gets one chapter of soak where everything else gets twelve — build and soak it against a STAND-IN early so

Ch 13 is its integration and not its birth; (2) the Prologue and Ch 1 enter the matrix LAST and carry the

slice's only never-run terrain path, so their first soak nights are also that path's first evidence.

QT-13 — THE NIGHTLY SOAK, this week, this machine (BUILD-NOW). An unattended scheduled overnight pass over the greybox + Ch2-5 rig (extending per region leg as Ch 6-13 land, and to the full 15 nodes at slice-end): (1) the full accreted Tier-1 regression suite; (2) a Tier-2 agent walk of the current playable surface; (3) a capture set the morning critic reads; (4) a hang heartbeat (the current on-demand run_qa_pass ALREADY hit TIMEOUT after 900s — the nightly wraps each leg in a MarkHeartbeatActive-style watchdog so a hang opens a record instead of stalling the run); (5) an auto-triaged bug delta + a scorecard delta waiting each morning. This is NOT 5090-gated — only the *volume/parallel* soak is (TAG LAW: only generative inference gates on the 5090). Absence-proof that it is genuinely new: run_qa_pass.py is invoked on-demand in the workflow logs, there is no scheduled-task/cron entry and no Saved/QA/nightly/ tree today.

11.5.4 Auto-triage, the beta player, and the 5090 fleet

• QT-14 — auto-triage → typed bug records (QA-11, BUILD-NOW). Ingest Saved/Logs/*.log + Saved/Crashes/**/CrashContext.runtime-xml + the automation-timeout signal; ensure()/check()/TIMEOUT lines auto-open a typed record (§5 schema) with artifact refs pre-filled and a severity guess. The 900s hang is record #1 this class writes.
• QT-15 — the AI beta player (QA-4, BUILD-NOW). The Tier-2 agent proper; the four senses feed it, InjectInputForAction is its hands. NeoStack PIE Lua is the works-now in-editor bridge; the packaged path is the InjectInput action-level route.
• QT-16 — the 5090 soak fleet (QA-14, GATED-5090). UGauntletTestController + RunUAT RunUnreal on packaged builds; many parallel long-soak/perf/crash-collection instances. Drop-in contract: QT-13's nightly harness IS this harness scaled down — the fleet is the same driver × N instances when the workstation lands. Parked, not re-planned.

11.5.5 The coverage-scorecard family — the density CURVE made measurable

Per-chapter content-budget and per-chapter test-coverage are ONE scorecard family (coverage_scorecard.json per node, rolled up 79-wide) that the gates watch. This is the density ruling made observable: CANON_AUDIT §7 RULES every chapter a dense woven experience (a few sub-bosses culminating in one main boss; combat/exploration/discovery/puzzle/mystery/lore/trade/gathering/side-quest/folklore all woven) — and Josh's in-conversation refinement (2026-07-24) supersedes any uniform-density reading with a CURVE: the completion-hours target is ≈100-200h main / 1-2k h side+weave / 8-10k h 100%, and the early chapters (Ch2-5) are deliberately LIGHTER, seed-rich, opening the game. The scorecard's per-chapter TARGET is therefore a rising band, not a flat floor.

• QT-17 — content-budget scorecard (QA-7). Per node, the measured budget vector read from the DTs: sites (DT_Site 376), beats-by-type (DT_Beat 891), encounters + sub-bosses (DT_BossEncounter/DT_BossPhase), forks + options (DT_Fork 90/DT_ForkOption 305), minigames (DT_Minigame MG_), weave-links (Cross_Cultural_Link_Graph 184 + CROSS_CULTURAL_WEAVE.md), significance-lenses (SIGNIFICANCE_LENSES.md, 33), personas (T0_Character_Index 163), threads-woven (DT_ThreadGrid 664 / 22 threads), reputation surfaces (T0_Reputation_Ladder + reputation-exalted-prestige-nemesis.md), and §16 citations (DT_ResearchCorpus 1356). Grounded reference shape (the slice, read directly): CH_03 = 7 sites / 12 beats (4 combat·4 character·1 puzzle·1 learning·1 discovery·1 transformation) / 1 main + 3 sub bosses / 1 fork × 4 options — the early-game floor of the curve.
• QT-18 — test-coverage scorecard (QA-7). Per node, what fraction of that content is under an actual QT harness: beats with a reachability test (QT-7), bosses with an invariant test (QT-5), forks with a completability test (QT-7), minigames with a runtime test, care/reveal surfaces with a validator (QT-9), audio cues fired (QT-10).
• QT-19 — the unified family + the gate (QA-7). One scorecard, one gate watching BOTH axes. It gates SHAPES/RELATIONS/FLOORS only (tuning firewall): (a) content ≥ the chapter-band floor — the SHAPE is monotone-rising early→mid, so a mid-game chapter under the early-game floor fails; (b) test-coverage ≥ its band floor; (c) the RELATION coverage-tracks-content (a heavy chapter with thin coverage fails). It NEVER asserts an absolute count is "correct" — the exact per-band targets stay illustrative until Josh tunes; the curve's monotonicity and the coverage relation are what is armed. worklist_next names the thinnest chapter each run.
• **RESCOPE BLOCKER on QT-19 (2026-07-27) — the family may NOT be armed past Ch 5 until the bands extend

and the off-curve nodes carry declared sentinels.** The slice is now Prologue→Ch 13, so the gate would be

asked to score eight nodes whose band does not exist and two nodes that are not on the curve at all: the

Prologue is a ceiling TUTORIAL (its job is presentation, not content budget) and Ch 1 is declared out of

corpus scope. Clause (a)'s monotone-rising SHAPE is exactly what mis-fires here — a false regression, the

same defect class already caught for Ch 77's exempt demand_ordinal. The fix is the band/sentinel pass,

which is its own item and lands WITH check_slice_scorecard.py (which does not exist yet); no band values

are written into this spec ahead of it, because inventing them would cross the tuning firewall this very

item exists to defend. Until then QT-19 arms over Ch 2-5 only, and its scope grows band-by-band.

11.5.6 The human-only list — what the machine does NOT test (honest)

The machine owns floors; Josh owns the ceiling. These are explicitly NOT gated and route to Josh (the J-cadence dailies packet) or to the enhancer intake (QA-13), never to a bug record:

• Fun / feel / does-it-land — is the fight satisfying, does the beat move you, is the region a place you want to be. Gates judge floors; humans judge ceilings.
• The care backstop — content Josh judges genuinely too insensitive or brutal from a REAL playthrough. Routes to the enhancer as a RENAME/REFRAME edit (never removal/defang); the mechanical §17-floor is the only care call the machine makes alone.
• The reference-title bar SEEDING — which shipped titles set the AAA bar per surface. Director seeds within-vision; Josh adjusts at playtest. The audit machinery (QA-12) that measures against the bar is ENGINE; the bar itself is a human call.
• Cardinal-canon rulings — chapter/numerical/naming/hard-line/§17/system-canon forks go to Josh as a full brief, never a bare question (the machine flags, never rules).
• Aesthetic taste on generated assets — "does this read as AAA, not slop"; and the dignity gut-check ("does this culture read with respect") beyond the mechanical §17.1 collective-care validator.

11.5.7 Accretion + CI — the standing spine

• QT-21 — regression accretion + graduation (QA-9/QA-5, LIVE). Every fixed bug's failing repro stays in the accreted suite permanently (the 11-record ledger already enforces fixed ⇒ regression_test_ref); "is this bug back" is the test going red, never an argument; a class caught twice graduates from a per-run catch into a standing gate exactly as reveal_discipline and boss_axes did.
• QT-22 — CI wiring + gate integration (QA-10/QA-8, BUILD-NOW). The QT harnesses ride run_gates.py (canon-side, unpiped, green-before-land) + RunUAT BuildCookRun … -run=DataValidation (game-side); non-zero exit on any Tier-1 or DataValidation failure blocks the wave; every harness self-tests (QA-8) and refuses to run if its own teeth break.

11.5.8 The QT-1..22 map — id · fold · tag · citation/absence-proof · depends · wave

Verified-by (the QA-ORG wave's own clause, and what WS-10 supplies to every OTHER wave): the QA-ORG wave is verified when (1) each of QT-5..11 ships green must-fire + must-not-fire fixtures and refuses to run toothless (QA-8); (2) the nightly (QT-13) completes one unattended overnight pass on the current machine, producing a triaged bug delta + a coverage-scorecard delta with the hang-heartbeat armed; (3) run_gates.py stays green unpiped and the coverage gate (QT-19) is armed against the curve, not an absolute. WS-10's product is the verified-by-playing clause it hands to every content wave — FEEL/UI/Ch3-5-REGIONS/TRANSITIONS/LIVING-WORLD each inherit their QT harness as their definition-of-done, so "it works / it reads / it's fun" never ships without a backing frame, a green repro, and a scorecard.

---

THE MASTER WAVE QUEUE (rebuilt — zero orphans, dependency-correct)

**STATUS AS OF 2026-07-26 EVENING — THE WAVE QUEUE IS CLOSED (the ▶ PROGRESS LEDGER is the

per-landing authority):** W1-W8 ✅ · W9 ✅ (3 passes) · W10 ✅ (the shipping surface + the

executable reveal table) · W-SPACE ✅ ALL regions · the COMBAT ADDENDUM TAIL ✅ (foundry /

status / arena-mutation flagship / presentation seam) · Ch-5 staged ×4 maps + PG-2 seams ✅ ·

LANE-SOAK ARMED ✅ (first run green) · the slice is STRUCTURALLY END-TO-END on a real player

HUD with 54/54 tests and every space/state/reveal tooth green. Remaining pre-5090: the

canon-residue batch (named in ledger 2026-07-26g), polish riders, the nightly soak cadence.

The earlier per-wave detail block below is superseded by this line and the ledger.

*(The block below reflected mid-day 2026-07-26:)* **W1 ✅ · W2 ✅ · W3 ✅ · W4 ✅ (true weightmap

tier; BUG-0016/0015 closed) · W5 ✅ · wave-G ✅ · W6 ✅ (foundation: material-spine runtime +

B04 recharge) · W7 ✅ (foundation: scenes + ambient beds audible) · W8 ✅ (Ch-3 staged + seam;

Ch-4/Bali staged; PG-2 cross-map = open) · W9 ◐ IN PROGRESS (passes 1-2 landed: parry, attack

library w/ punish windows, counter-class law, caci, hitzones/part-breaks, the weakness-read,

spawn-Z; pass 3 in flight: S2/S3 mechanics + state isolation; the tail = the

COMBAT_PROGRAM_ADDENDUM's six systems) · W10 ⬜ NEXT (the shipping surface — all its canon

inputs are ready) · W11 ⬜ · LANE-CANON ✅ for the slice (the table CLOSED: strings ×4, bboxes

0-empty, material spine, fauna 150, festivals, SFX, the element matrix, clue lane, pip schema;

residue = personas/TR-3/TR-6/rune-registry reconciliation) · LANE-SOAK ⬜ arming next ·

NEW WAVES ADDED THIS SESSION: W-SPACE ✅-on-Flores/◐-remaining-regions (the audit→sculpt→

teeth factory stage) + the COMBAT ADDENDUM tail ⬜ (foundry/status/combo/healing/ladder-seam/

arena-mutation + the Rune Forge).**

Supersedes synth_program §A.4. Covers every item in all 8 sections (217 unique items after

dedupe). Ranking follows the directive: **E-residue first · nightly-soak + feel + audio-spine early ·

input/targeting before the movement + AI that need the binds · terrain long-poles parallel from day 1

· W-AI before the region sub-bosses (wave-G fold) that consume it · canon-author lane on the canon

side throughout.** Two-repo discipline: canon (C:/dev/humanity-forgotten) and game

(C:/dev/Humanity/Humanity) are one-writer-each, so the persistent canon + soak lanes run in PARALLEL

with the ranked game waves. Each wave inherits its WS-10 QT harness as its definition-of-done.

Persistent lanes (run the ENTIRE window, parallel to everything)

Ranked game waves (one writer, game side)

*Note: the DR staging legs (DR-3/4/5/6) are construction rules APPLIED during W1/W6/W8 — an actor is

placed carrying its stable id + asset handle at placement time — then AUDITED in W11. W11 is the proof,

not the placement.*

FUN-REBUILD ABSORPTION MAP (2026-07-24 — the Ch-2 player-fun review folded in)

The step-back playthrough review (four PXR sections + both fresh-context critics, GO_WITH_FIXES)

produced eight rebuild threads — R1-R8 + R-DATA, specified in docs/FUN_REBUILD_PLAN.md (the

detailed item specs live there; this map is the queue authority for WHERE each thread lands). **No new

waves**; the three re-scopes below AMEND the W6/W9/W10 rows above in place. Two new obligations only: a

runtime clue-state addition (WS-C lineage, feeds W10) and the canon-lane DT_Clue table + R-DATA columns

(the latter APPLIED 2026-07-24).

W6 re-scope (AMENDS the W6 row): ADD the Ch-2 keystone-teaching consumers — the vril-site

rapid-recharge interaction (B04; the Vril bubble visibly fills in the spray — CVD §18.4

substrate-as-real; spec L43 channel-c) on the vril-site/LW-WIRE seam, and the fire-making gather loop

(B02) on the FL-3 gather-node + TR-1 harvest subsystem. Verified-by adds: B04 recharge visibly fills

the bubble in a Ch-2 QA pass; the B02 fire loop yields on a real gather→strike, no N-press.

W9 re-scope (AMENDS the W9 row): from "boss-attack variety" to **the puzzle-boss chain-meter

template** (R2 + R1 depth + R8 refusal-pick). ADD: the NEW parry primitive (a timed deflect — does not

exist; dodge does); per-attack counter-class tagging (dodge-only / parry-only / elemental-neutralize on

cast_pattern); the pip-meter UI (5/7/9+, pip_reveal_threshold-driven appearance); thermal-only

elemental-neutralize at Ch 2; monotonic de-escalation (vary the READ, never re-tighten the window).

Consumes R-DATA's columns (live). Verified-by adds: pips fill on a correct counter and WIPE on a

mistake; the freed-path pips appear at Rangku ≤25% HP and NOT before; the guardian eases monotonically

with no window re-tighten; a parry deflect whiffs a parry-tagged strike; kill-instead docks integrity +

closes a fork. Honest caveat carried: the serpent read-channel (anim) is T2-deferred — W9 ships the

logic, not the "template" feel.

W10 re-scope (AMENDS the W10 row): build the shipping surface FROM the R7 translation table, never

the debug draw, and host the discovery surface: ADD the field-sense re-read affordance + the

clue-staged objective display + the "sense updated" toast (the pre-Ch-13 stand-in for the Ch-13

journal, HL_0070); ADD integrity-as-appearance/voice (no integrity_score integer) and

difficulty-as-diegetic (no d_local/lever names/hash on any player surface). CONSUME the R7.1

spoiler-safe player-objective/option fields + the R7.2 site display_name/chapter-title strings.

Verified-by adds: the reveal-discipline test (no B06/fork-opt-2 pre-tell reaches the player surface);

the field-sense panel gains an entry only after a clue fires, never pre-stated; no

mode-taxonomy/raw-ID/number leaks in any shipped widget frame.

THE CUT LINE — honest about the 2-3-week window

The window's TARGET is the full slice STRUCTURE playable on placeholders (directive #1). "Complete

structure" ≠ complete fidelity. Three explicit buckets:

SHIPS IN THE WINDOW (built now, on placeholders):

• MUST-SHIP core: W1–W8 + both persistent lanes — Ch2-5 playable end-to-end, cross-chapter

progression across 4 maps, feel + audio spine, input + lock-on, traversal (incl. the HL_0079 canopy

course) + enemy AI, the living-world runtime, the scene layer, and the three staged regions with

save/progression. This is the honest "100% structure" deliverable.

• POLISH tail (ships if the window holds; else first post-window work): W9 (reactions / boss

variety / per-weapon+ability presentation), W10 (full UI + book screens + loc + economy UX), W11

(handshake audit). These deepen feel and surface, not the playable spine.

HANDED TO THE 5090 (drop-in, NOT built in the window — the window builds the SOCKET):

• T2 final art: Niagara VFX, generated meshes/PBR textures, final HUD/icon art (every CF-* cue's

final VFX rides GATED-5090; the T0/T1 feel ships now).

• GATED-EXT: final mocap/animation (external service, never generative inference).
• GATED-5090 generation payloads via the DR contracts: music (DR-AU1), voice/VO (DR-AU2), SFX

(DR-AU3), MRQ dailies + MetaHuman roster (DR-CN1) — the seams (AU-3/5/9/10/11, CN-1/4/7) build now.

• The FULL ENRICHMENT PASS over the seed-rich early chapters: weave/culture/side-quest/persona depth,

pharmacology/ecology fidelity, significance lenses — writes onto the DR-1 stable ids + DR-7 weave

anchors this window plants. Pure fill, never rework — that is the handshake's guarantee (§B).

BLOCKED-ON-BRIEF (Josh gates — status as of 2026-07-24):

• D-ECONOMY-MEDIUM — RESOLVED (Josh ruled Opt B: historically-real mixed exchange per region).

EC-medium + EC-vendor unblocked.

• **D-DEFEAT-AXIS — RESOLVED (Josh overruled: rated-M, HP 0 = death; protected-only 1-HP withdrawal;

per-boss pips at threshold).** CF-10 boss-defeat STATE, CN-7 resolution scene, SL-1 unblocked;

executed by the R1 wave + R-DATA (see the FUN-REBUILD absorption map above + the register).

• CN-9 DT_SceneDispatch → Josh #22 schema ratification (slice does not need it; parked — still open).

Realistic-window honesty: at the team's ~5–10× execution rate the in-window deliverable is the full

slice STRUCTURE (MUST-SHIP core, and likely the polish tail); the density/fidelity climb is explicitly

the post-5090 curve work, per the seed-rich early-game band of the density ruling.

---

---

THE DECISIONS REGISTER (options · deep-dive · adversarial · recommendation · strongest objection — never bare)

Per the standing decision protocol. Creative-within-vision calls are ruled + applied by the director

(shown in the sections, not asked); only genuine cardinal / system-canon forks are carried here for

Josh. The 3 SHIP briefs are carried by reference to docs/ROADMAP_POST_5090_TO_SHIP.md (they are

post-5090 / ship-facing, not pre-5090-window decisions).

D-ECONOMY-MEDIUM — transaction medium + vendor mechanics + L2 treasure placement *(RESOLVED — Josh RULED Opt B, 2026-07-24)*

RESOLUTION: Josh ruled Opt B verbatim — "historically real mixed exchange per region," no universal

coin early. EC-medium + EC-vendor are UNBLOCKED and land per the Opt-B mechanization below

(medium_of_exchange on Region_Index; vendors price in the local medium; L2 treasure = the σ-drop

spine). [Historical note retained: an earlier draft wrongly pre-marked this "director-applied" before

the ruling existed — the flag-then-rule path was the correct one and it has now been ruled.]

The gap. The trade MODEL is canon (T1_Trades_Master §9.2 trade-interdependence + region-specific

scarcity; §9.1 two separate economies). The MEDIUM (what the player pays with, how a vendor prices,

where L2 treasure sits) is undefined, and the gather→craft→trade loop is live in Ch 2-5.

• Opt A — universal coin from Ch2. REJECTED: breaks §9.2's immersive-teaching intent + ahistorical

for the 1200 CE regions.

• **Opt B (RECOMMENDED direction) — historically-real mixed exchange per region, NO universal coin

early.** Barter + region-appropriate proto-currency (rice/cloth/metal-weight where the real region

used it); vendors price in the local medium; L2 treasure = σ-bounded drops (HL_0061) via

RngDropTable + gather yields. Land as: a medium_of_exchange token on Region_Index, vendor

mechanics as a P2.6 inventory consumer, L2 treasure = the σ-drop spine (LW-WIRE) + gather yields.

Fits §9.2 + §9.1 + immersive-teaching; extends to the Tier-C MMO cross-player economy.

• Opt C — no vendor economy in the slice, gather/craft/use only. REJECTED: leaves gathered

materials + drops with no sink; reads incomplete against the 100%-structure target.

Strongest objection to B: per-region exchange multiplies vendor-UI complexity. Rebuttal: the

slice needs only the Neophyte-floor exchange surface + L2 placement, not a full market sim; Josh

backstops via playtest. Blocked items: EC-medium, EC-vendor. Proceeding regardless: EC-inventory

(P2.6), EC-treasure (LW-WIRE).

D-DEFEAT-AXIS — the boss HP=0 defeat model *(RESOLVED — Josh OVERRULED the recommendation, 2026-07-24)*

RESOLUTION (verbatim intent): "Rated M game. Things will die. Health 0 is death. If they're a

specific boss or combatant that we don't kill for some specific canonical reason because it'll gate

story progression then they'd have to withdraw at 1 health." Refined same-day: "can't free everyone" —

non-kill paths are PER-BOSS, and for fight-first bosses the pip chain APPEARS only at a per-boss HP

threshold (~25%), revealing the alternative win. Executed by: the R1 wave (FUN_REBUILD_PLAN — the

defeat-axis derivation fix BUG-0014, the freed win-route, consolation surfacing) + R-DATA (the four

explicit Boss_Encounter columns has_nonkill_path/pip_count/pip_reveal_threshold/defeat_protected,

16 slice rows populated). Opt B below is retained for provenance only — it was NOT adopted; HP is lethal,

Composure is the boss keystone read, and defeat_protected (rare, canon-cited) is the only 1-HP clamp.

CF-10 boss-defeat STATE, CN-7 resolution scene, SL-1 death polish: UNBLOCKED.

*(The original brief, for provenance:)*

*(Aliases in drafts: D-BOSS-DEFEAT-AXIS, D-HP0-AXIS — one brief.)* Blocks CF-10's boss-defeat STATE,

CN-7's resolution scene, and SL-1's death polish. TODO_CANON (game): the Composure/Withdrawal fight

consumes no HP=0, so the boss neither dies nor staggers (QA image-critic recorded Boss HP 0/100 while

Composure 100/100 and the fight continued).

• Opt A — Health IS the defeat axis; HP-0 → death/execution. REJECTED: canon-wrong for a FREED boss

(BE_0002 wins by the polo's hold breaking and Rangku freed, not a corpse); violates §17.1 (a

coerced-victim host) + the RULED win-condition enum (T99 §5.1 REDUCE "never a pure DPS check") + the

zero-enemy-stat-delta headline.

• **Opt B (RECOMMENDED) — Composure-break-to-Withdrawal is the SOLE defeat axis; boss Health is a

non-lethal pressure/stagger track.** On win-condition met → a boss defeat/withdrawal state plays the

canon resolution (FREE → the hunter is freed and exits; PACIFY → yield; PURGE → the manifestation

purged). HP=0 retained only as a fallback safety valve, routed to the same withdrawal state, never a

ragdoll. Matches Pillar 8 + the observable-thesis invariant + no-dead-end (§2.5). *(ws1_feel Alt-2 ==

ws3_prog REC(a) == this — the drafts converge.)*

• Opt C — HP-0 opens a stagger/execution window feeding Composure. More feel, more surface;

recommendation folds a thin C-flavored stagger READ into B for feel (HP-0 → brief stagger, no death).

Strongest objection to B: a pure-pressure boss can read as un-killable to an action-RPG player

expecting an HP payoff — mitigated by the Composure-break MOMENT (CF-6) as the visible kill-beat, and

the thin stagger read. Small data dependency: the win_condition/win_shape columns need

back-filling on the 16 slice boss rows (declared T99 §4, back-fillable from phase prose) — rides the

same commit as the CF-10 shell once ruled. **This touches the combat pillar (system-canon) — briefed

for Josh; the greybox has no death state to wire until it is ruled.** Reversible (a defeat-state swap),

so erring toward the canon-honest FREE model is the cheap error.

D-BOSS-PIP-SCHEMA — the per-boss non-kill data model *(RATIFIED by Josh, 2026-07-24 evening: "ratify as applied")*

Same-message rulings (2026-07-24 evening): naga naming — the Ch-2 pair stays **"the Kelimutu

Guardian"** (naga reserved for Ch-3's naga tana; the Ch-3 retro-framing dialogue link adopted as

enrichment). The Ch-2 two-of-three counter-class scope stands (parry debuts at the Ch-3 caci;

the primitive builds now). The Vril Sight label fix ruled + applied (CH_02.md corrected per

the reconciled registry/T1/D-06 model). The elemental weakness chart is SUPERSEDED — Josh

AUTHORED a full matrix incl. hybrid schools (lightning/life/poison/mud/smoke/steam+, trade-skill

ties, consumption costs) in the GDD / original humanity docs, possibly Drive-only; recovery sweep

ran; NOTHING elemental is minted until it is recovered (Ch-2 thermal-only stands interim). See

docs/FUN_REBUILD_PLAN.md §4 for the annotated briefs.

What: four explicit Boss_Encounter columns — has_nonkill_path, pip_count, pip_reveal_threshold,

defeat_protected — landed on the built registry-extension rail (docs/registry_extensions.json

combat-encounter ownership; schema-dictionary §4.1 annotated; fidelity re-baselined), with the 16

Ch 2-5 slice rows populated. Why director-ruled, not flagged: the columns are the direct mechanical

encoding of Josh's two same-day verbatim rulings (the chain-meter spec; "can't free everyone… reduced to

25% or less health… for the pips to show up") — schema/tooling execution of a ruling already issued, per

the reframed cardinal rule; blocking the marquee mechanic on a re-confirm would be the over-conservatism

defect. Ratifiable shapes (adjust any value at leisure): Rangku 5 pips @ 0.25 reveal · Kelimutu

guardian 5 @ 1.0 (test-from-the-start) · Ch-4/5 encounters 7 · pips EMPTY on the three non-chain

encounters (polo refusal-pick, Chelvan information-layer, devaraja baray-puzzle) · Naga Padoha the one

defeat_protected TRUE (his row's own J10 cosmic-subdual-nonkill Batak-binding target_type). Remaining

265 rows populate per chapter at factory time; empty = not-yet-populated, NEVER kill-only.

Carried by reference — the 3 SHIP briefs (docs/ROADMAP_POST_5090_TO_SHIP.md Decisions register)

These are ship-facing / post-5090 Josh-tier forks, not pre-5090-window decisions. Full options +

recommendation + strongest objection live in the roadmap; referenced here so the window leaves each seam

addressable.

• D-VO-POLICY (roadmap P2.8) — does Ch 2-5 ship with spoken VO, and on which lane? Rec: partial

(generated hero/canonical lines only; bulk/ambient stays subtitle). The FText subtitle floor + the

CN-4 dialogue-box seam are BUILD-NOW now; the VO payload is DR-AU2 (GATED-5090).

• D-STEAM-FORMAT (roadmap P4) — Early Access vs standalone slice vs demo+wishlist. Rec: Early Access.

Josh's business call; no pre-5090 build depends on it.

• D-PRICING (roadmap P4) — price point + any EA-to-full path. Deferred to Josh's market read; the

loop is prohibited from setting it.

NOTE-ONLY (out of slice, recorded so the seam stays addressable)

• NG+ / prestige carry-over (#28) — what carries across "The Forgotten One" (mastery, familiars,

inventory, world-state, the un-pruned seventh vril-seat). Proposal-tier, pending Josh; NOT in the

window. The save/progression waves (PG-6, W8) leave the carry-over seam addressable, nothing more.

• The qualitative-delta → integer map (#11) — IB-1 + CN-3 read per-class integrity weights; the band

map exists (T1_Integrity_Paths §3.2, docs/INTEGRITY_DELTA_MAP.md). Creative-within-vision: derive

from the §3.2 band midpoints under the tuning firewall — ruled + applied, not a Josh flag.

---

WHAT THIS PLAN DOES NOT CLAIM

• It does not claim any generative surface is done. Every T2 / GATED-5090 payload (VFX, meshes,

PBR, music, VO, SFX, MRQ/MetaHuman) is a DROP-IN over a socket this window builds — proposed against

benchmarks that have not run. The AssetGen PBR texture leg is genuinely unsolved as of 2026-07-15.

• It does not claim complete fidelity in the window. The deliverable is the slice STRUCTURE playable

on placeholders (directive #1). The density/fidelity climb — the enrichment pass over the seed-rich

early chapters — is explicitly post-5090 curve work.

• It does not claim a ship date or chapter cadence. Waves gate on their verified-by clause, not on

weeks. The "2-3-week window" is Josh's hardware ETA, not a delivery promise; the factory throughput is

a range the first post-slice chapters must MEASURE.

• It does not rule the cardinal forks. D-ECONOMY-MEDIUM (world-rule) and D-DEFEAT-AXIS (combat

pillar) are Josh-pending with recommendations ready — the loop flags, it does not rule. The 3 ship

briefs (VO / Steam-format / pricing) are Josh's business calls.

• It does not invent canon. Every item carries a citation or a positive-controlled absence-proof;

the LOCKED pillars/vision/structural-skeleton/numerical-canon (185/72/22/120/33/10/163/42; 16 slice

bosses; LINK_0186) are consumed, never changed. Where an item rests on a 0-byte recovery file (DR-2,

DR-8), the 0-byte file IS the absence-proof and is flagged for confirmation against the live wiring.

• It does not claim zero orphans by assertion. The MASTER WAVE QUEUE was cross-mapped item-by-item

against all 8 sections (217 unique items after dedupe); the RECONCILIATION LEDGER is the audit trail,

and any item not appearing in a wave is a defect to be surfaced, not assumed absent.

• It does not claim the machine judges taste. WS-10 gates FLOORS (the §17 hard-line, reveal windows,

reachability, data integrity, the coverage curve). Fun/feel/does-it-land, the care backstop beyond the

mechanical §17.1 validator, and the reference-title bar are Josh's ceiling calls, routed to the

J-cadence packet, never to a bug record.

---

*End of PRE-5090 BUILD PLAN — VOLUME 2.*

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