SLICE_SCOPE_COST_2026-07-27.md

pipelines/SLICE_SCOPE_COST_2026-07-27.md

SLICE SCOPE / COST REALITY — the four shapes, measured

Analyst axis: SCOPE + COST per shape. Read-only pass over canon (C:/dev/humanity-forgotten) and game

(C:/dev/Humanity/Humanity). Every number below is read off a file, not estimated, unless the line says

ESTIMATE and carries its derivation.

---

0. STATE CORRECTION — the ruling has already landed (read this before using anything below)

The task framed Josh's question as "thinking aloud — NOT yet a ruling." It is now a ruling.

docs/spine/DECISIONS_PENDING_JOSH.md L3-26, ✅ RULED 2026-07-27 (fifth sitting):

"Option c. We will do 2-13 first then prologue and chapter 1 then will continue after vertical slice.

After ch 13 we will close with healers child joining and they're off on their travel — to be continued."

• THE SLICE = Prologue → Ch 13 (15 nodes). Build order **Ch 2-13 first (11 earthly chapters), THEN

Prologue + Ch 1**. Slice ENDS at Ch 13 on the healer's-child join + departure.

• The build-order collision the task named is already resolved by Josh himself: the fairy realm now

designs-and-builds ONCE at slice-end (after Ch 2-13 matures the concept-art pipeline) — explicitly

superseding the same-day realm directive's "after the nine mythological realms" ordering. His stated

reason: *"11 chapters is definitely strong enough to have our concept art pipeline down and enhanced

and enriched and optimized to then design that fairy realm."*

• Capacity noted in the same ruling: 5090 paid + being built; three Claude Max x20 subscriptions to

chain for the factory.

• The ruling itself orders the follow-on: *"the slice-scope workflow's analyses re-point to the execution

re-plan."*

So this document changes register: it is no longer a decision brief between four shapes. It is the

execution cost baseline for the ruled shape (c-with-d-sequencing), with the other three shapes costed

as the counterfactuals that make the ruled shape's price legible. Section 6 states what the ruled shape

costs and where it is most likely to hurt. Sections 1-5 are the measured surface it rests on.

---

1. THE MEASURED BASELINE — what Ch 2-5 actually is, and what it actually cost

1.1 The built surface (read off the spine + the game repo)

1.2 The cost precedent (the honest read of what Ch 3-5 cost)

Three of the four regions were built inside the pre-5090 window. PRE_5090_BUILD_PLAN_VOL2.md §6 is the

plan-of-record and the ledger is the outcome. What it shows:

• The pipeline generalization was a ONE-TIME cost, already paid. RG.1/RG.2/RG.3 turned three

flores-hardcoded drivers (import_flores_landscape, dress_flores_map, stage_flores_ch02) into

--slug/--chapter drivers. derive_region_bbox and terrain_fetch_preprocess were already

region-agnostic; build_boss_statetree was already boss-parameterized. Ch 6-13 inherits this free.

• Terrain is deterministic, network/CPU-bound, and parallelizes fully. No GPU, no model. Three GIS

chains ran concurrently. This is the cheapest leg per region and is NOT the long pole.

• **The long poles named in the plan were the three terrain builds; the long pole in practice was the

canon side.** The four region pages total 12,918 lines. flores pass-2 alone went 555 → 1,721 lines.

bali's consumption manifest bucketed 85 CONSUMABLE / 11 HELD / 9 CUT. That authoring is per-region,

serial per page, and critic-gated per section.

• W-SPACE was an unplanned discovery cost that is now permanent per chapter. The slice's own audit

found 6 of 11 staged sites sat on >15° raw slope; the fix (14 sculpted arena pads + the

approach→threshold→arena→payoff grammar + permanent slope/navmesh teeth) is now a standing per-chapter

factory leg (ROADMAP_POST_5090_TO_SHIP.md P5). Every new chapter pays it.

• Calendar anchor, stated honestly as a measured fact and not a promise: the window from "trunk

closed + combat chain built" (ledger 2026-07-23d) to "slice structurally end-to-end, 54/54 tests"

(ledger 2026-07-26 evening) was four calendar days of director-run fan-out, on top of a foundation

laid 2026-07-03 → 2026-07-22. That four-day figure covers W1-W10 + W-SPACE + Ch3-5 staging across

4 maps. It is the only real throughput datum that exists, and the roadmap's own discipline

(ROADMAP_POST_5090_TO_SHIP.md §0, P5) forbids converting it into a schedule.

---

2. WHAT Ch 6-13 ACTUALLY IS — the enumerated build surface

2.1 Per-chapter, read off the spine entries

Delta over the current slice: sites +82% · beats +215% · forks +200% · boss rows +200% · regions +200%.

No SPREAD tags add regions beyond the 8 (verified per entry); Ch 13 carries a ratified SPREAD (Great

Zimbabwe + Zambezi) but it lands inside its own region page.

2.2 The map-count reality — the single biggest terrain cost driver, and it is superlinear

A UE map covers a contiguous DEM AOI. The slice already proved the split rule: **sumatra_java's four

sites span 1,142 × 1,255 km and needed THREE maps**, while angkor_khmer's 261 × 459 km needed one.

Spatial span measured from T0_Region_Index.coordinates (the anchor pairs recorded there; the spine site

lists reach further in several cases):

Derivation of the estimate: apply the observed sumatra_java rule (>~500 km separation between site

clusters ⇒ separate AOI) to each region's site list. Lower bound assumes aggressive clustering and

travel-card seams; upper bound assumes one AOI per disjoint cluster. **The honest statement is a band,

not a number — but even the floor is 2.3× the slice's map count and the ceiling is 4×.**

Two riders that make the ceiling more likely than the floor:

• Fork C (RULED 2026-07-27, second sitting) — tiered traversal: transitions are PLAYABLE navigable

areas, important crossings become staged or open DUNGEONS. That converts inter-cluster gaps from

"loading screen" into buildable content. It raises quality and raises cost; both are real.

• swahili_coast's dhow_open_sea site is an ocean AOI with no DEM analogue in the built pipeline.

2.3 The FIVE system debuts Ch 6-13 forces that the current slice never exercises

This is the finding that matters most, and it is not visible from a chapter count.

1. THE COMPANION RUNTIME — debuts at Ch 13, the slice's own closing beat.

Ch 2-12 continuity anchors read "no traveling companion" verbatim, in every one. Ntala (CHAR_0002)

formally joins at CH13_B01 under HL_0070. And Fork B (RULED same day) makes her load-bearing in

combat: she is THE RESCUER — the one who brings the downed player to the revival/recharge zone,

with the hold-on countdown as the failure clock. So the slice's terminal chapter turns on ally AI +

the downed/rescue loop + companion loyalty-on-integrity + (per the addendum) ally stat builds — a

system stack with zero prior in-slice exercise. It is live for all 13 of Ch 13's beats, not just

the finale, which helps; but it gets one chapter of soak instead of twelve.

2. THE EVIDENCE JOURNAL — debuts at Ch 13 (HL_0070; ws.journal_status: none → ACTIVE_HEALERS_CHILD).

A whole L2 collection/UI/authorship system, path-variable across the arc, with zero slice precedent.

3. FOUR TRADE HOME BASES — the entire trade-home system debuts in Ch 7-12.

Ch 7: TRADE_03 Blacksmithing (wootz crucible) + TRADE_04 Herbalism (Charaka/Sushruta pharmacopoeia) —

the arc's first two. Ch 8: TRADE_10 Fishing and Navigation (mtepe sewn-plank + monsoon navigation).

Ch 12: TRADE_01 Animal Communication (Ouidah python temple, cross-species coordination). The slice has

zero home bases — gather/harvest substrate only.

4. NAVAL / OPEN-SEA TRAVERSAL — Ch 8. dhow sailing, rigging part-breaks, an open-sea encounter

(dhow_open_sea), monsoon (kusi/kaskazi) navigation. New locomotion class + new map class.

5. STEALTH as a dominant mode — Ch 9. The fifth and only unbuilt mode (perception/detection/light

gameplay). Bounded, but genuinely new.

Secondary new-mechanic surface (cheaper, still real): Ch 7's pacify-by-yielding-into-respect with the

boss won as an ally; Ch 10's restraint-then-purgation boss where *"damage [is] impossible by

design"* — a no-damage encounter class; Ch 12's Dahomey Decision as a moral-choice boss with no

combat against the Ahosu or Agojie; Ch 13's evidence-journal activating on a boss's own record; and

8 distinct agility courses with 8 distinct hazard registers (fresco-wall, negative-space, ravine,

arboreal, sandstone cliff, alpine cold+rockfall).

2.4 What Ch 6-13 ALREADY has — the factory-readiness per node (this is the good news)

Calibration note that cuts against panic: the four slice regions were built on only ~24 tagged

sources between them (flores 10, sumatra_batak 7, bali 4, angkor 1, malay 1, mixed 1) and still produced

12,918 lines of cited region page. Substrate thinness routes work to deep-research agents (the ruled

default per BOOK_ACQUISITION.md), it does not block. But it does mean the Ch 8/9/11 pages lean on

research fan-out rather than on the P1 deep-read of scanned books.

2.5 The delta in one line

Ch 6-13 = **8 new region pages (0 exist), ESTIMATE 14-24 new UE maps (6 exist), 51 sites, 99 beats,

10 forks, 32 boss rows, 8 agility courses, 5 system debuts, 80-104 empty env-profile cells, and 8

missing substrate shelves** — on top of a spine layer that is already 100% complete.

---

3. THE PROLOGUE + CH 1 BUILD SURFACE — small in rows, unlike anything else in cost

3.1 What the rows say

On row count alone this reads as the cheapest pair of nodes in the game. That reading is wrong, for six

reasons, all of them structural.

3.2 The six reasons the fairy realm is the most expensive node-pair per row in the plan

1. coordinates: N/A (dimensional) — the entire terrain pipeline does not apply.

Every earthly region runs `derive_region_bbox → terrain_fetch_preprocess (COP30 DEM + WorldCover

weightmaps) → import_region_landscape → dress_region_map → stage_region_chapter`. That chain is

deterministic, ungated, parallel, and free of model spend — and it is **the reason the slice's terrain

was never the real long pole.** For fairy_realm there is no DEM, no WorldCover, no bbox, no

gazetteer citation. The realm's terrain must be wholly authored (sculpt / PCG / Gaea-from-nothing),

through a path the repo has never run once. This is the single largest unmodelled risk in the

whole slice, and no existing measurement bounds it.

2. A second, disjoint player kit, used for exactly one node. The Prologue's

abilities_unlocked gives the four SOVEREIGN faculties in full — Sovereign Vril Projection, Rift

Sealing, Temporal Sight, Familiar Communion — *"the ceiling tutorial ... STRIPPED at the Ch-1

rift-fall."* Ch 2-13 runs at the Neophyte floor across every capability (HL_0026), so the sovereign

kit shares nothing with the 185-ability tree the slice builds. It needs its own abilities, VFX, cues,

telegraphs, feel-tuning, and — under the Presentation Ladder (Josh NODDED, 2026-07-26: *"every spell

and ability and movement needs to be beautiful and fluid art"*) — its own top-tier animation/VFX

variants. Built for two nodes, at the game's highest presentation bar, because it IS the ceiling

reference the other 75 chapters climb toward.

3. Twelve luminous familiar species, all hero-grade, all for two nodes. The twelve royal communions

are the Ch-1 Child-Familiar selection pool. Fantastical (so generation-legal, no photogrammetry/rights

gate) — but twelve distinct creature designs, rigs, and communion behaviours, each of which the player

sees before choosing one.

4. Fully invented architecture at the grandest-display bar, with no shortcut path. Earthly regions

route real cultural architecture to reference-built / photogrammetry / licensed-kit

(T99_Translation_AssetGen §1 constraint 3). The sovereign hall and the flowering plain qualify for

generation — which means they route through the realm program's art choice-board pipeline

(REALM_DESIGN_PROGRAM.md §3: establishing board rules the grammar, accurate/enriched/surrounding

panels, four critics before Josh sees a board, boards versioned under docs/art_boards/). That

pipeline's pilots are Atlantis then the Duat — neither has run. Josh's own ruling is exactly this

point: build the earthly eleven first so the pipeline exists before the fairy realm consumes it.

5. The amulet integration is an 18-file canon edit before a single asset.

AMULET_PROVENANCE_BRIEF.md §10 enumerates it: CH_PROLOGUE (largest edit — header, continuity, body

prose, threads, a NEW town/settlement site row, boss scope note, beat-graph amendment, scene anchors,

mode-variety fold, ally NPCs, carry-forward, seam diffs, decisions), CH_01 (header, continuity, prose,

the dumb-fight and run beats inserted, ally NPCs, scene anchors, decisions), TRANSITIONS_PROLOGUE_01

(all four bullets re-authored), T3_Core §2.1/§6/§6.6 (+ the Ch-76 restraint note),

T0_Character_Index (a NEW mother row + a CHAR_0005 amendment), T0_Boss_Encounter_Registry

(BE_PROLOGUE scope + possibly a new catastrophe row), the scene rows (SC_CHPRO_the_birth participants

• new anchors for the catastrophe, the run-back, and the mother's death — the highest-care scene row

in the chapter), T0_Chapter_Index (mode_variety_check + beat counts + narrative cells for both nodes),

T0_Equipment_Registry (the amulet's row — the registry is header-only, zero rows),

T0_Ally_Behavior_Policy (the ward's not-aggro-eligible / not-downable rule),

T1_Threads_22 + T0_Thread_Grid (Thread 20 gains the amulet; Thread 6 gains a first taking),

CH_02 (reconcile the no-inventory start state against a worn unidentified object), CH_13-or-later

(the gift beat), DOC_MAP, SPINE_LIVING_REVISION_LOG, fidelity_baseline. None of it is applied;

the brief states of itself that nothing enters the spine until the reconciliation pass runs with its

own critic and its own reveal-discipline verification.

6. The beat graph GROWS and the chapter changes type. §4.2 of the brief: CHPRO_B07 amended (the

mother present, the amulet passing), plus new beats for the realm-service sandbox arc (help,

rebuild, maintain the town, close further rifts — *"the first place the Prologue becomes a sandbox

rather than a corridor"*), the catastrophe, the playable run-back, and the arrival at the dying mother.

The brief states the consequence plainly: *"the chapter gains combat-mode beats it did not have, and

its beat count rises past ten"* — so the Chapter_Index projected-five and mode_variety_check cells both

move. ESTIMATE: Prologue 10 → 15-17 beats, Ch 1 10 → 12-13 beats (derivation: 4-5 named new

Prologue beats, 1-2 named new Ch-1 beats, plus the amended ones).

3.3 What the fairy realm ALREADY has (the counterweight)

• The richest substrate pool of any node in the game: celtic_fairy 80 + british_isles 74 +

fairy_literary 3 ≈ 157 tagged sources — against flores's 10. Whether they are acquired/OCR'd is a

separate question, but the shelf is deep.

• quest_seed_queue/fairy-literary.md is 313 lines — the second-largest seed shelf in the repo

(flores 504, sumatra-batak 294, bali 271). Seeds for the realm are already harvested.

• Both spine entries are complete, typed v3.1, with substrate_sources cited (Mag Mell, Tír na nÓg,

Preiddeu Annwfn, aos sí / daoine sídhe, féth fíada, the changeling / F321 / F377 folklore).

• Zero forks, zero ambient threat, no protected cultural surfaces to gate — the care load is the

lightest in the arc (§17 floor is not in play; a parent's death in a dimensional accident is nowhere

near it, per the brief §9).

• FINALE_FAIRY_ENRICHMENT.md exists and already feeds the sovereign_hall site row.

3.4 The net

The fairy realm is row-cheap, substrate-rich, care-light, and production-expensive. Its cost lives

almost entirely in three places that no existing measurement bounds: the invented-terrain path (never

run), the sovereign kit (a second player kit for two nodes), and the art choice-board pipeline

(never piloted). Josh's build-order ruling is a direct, correct attack on the third of those.

---

4. THE FOUR SHAPES, COSTED

Delay is expressed in phase-equivalents against P1/P2/P3 of ROADMAP_POST_5090_TO_SHIP.md, not weeks.

That roadmap's own §0 binds: *"Do NOT time-box to weeks; the phases gate on exit criteria, not calendar."*

Where a band is given it is dependency-ordered and carries its derivation.

Shape (a) — STATUS QUO: Ch 2-5

• Delta over today: zero. Structure is done; P1 (deep-read + enrichment), P2 (generative surfaces),

P3 (QA at ship bar) run over 4 regions / 6 maps / 46 beats / 16 bosses / 27 sites.

• P1 scope: 4 region deep-read batches (already the named pilot in

DEEP_BOOK_PROCESSING_CONTRACT.md §4.2); 264 UNROUTED seeds typed; personas on ~30 slice entities.

• P3 soak matrix: 6 maps × 5 difficulty presets × the QT-5..11 harnesses; 54 tests accreting.
• P3-exit delay class: BASELINE (0).
• The honest problem, which is a design problem and not a cost problem: the slice ends mid-arc with

no companion, no journal, no trade home, no amulet, and no origin. It is a competent 4-chapter demo of

a 79-node story whose emotional spine Josh authored this week and which none of it contains. Josh's

instinct in the question is correct on the merits.

Shape (b) — Ch 2-13 earthly-only

• Content delta: +8 regions, +ESTIMATE 8-18 maps, +51 sites, +99 beats, +10 forks, +32 boss rows,

+8 agility courses, +5 system debuts, +8 region pages (ESTIMATE 12,000-42,000 lines at the observed

1,367-5,296 lines/page), +8 seed shelves, +8 civ-mark shelves, +80-104 env-profile cells.

• What the 5090 gates vs what is authorable NOW: the split does not move. Gated: the book

deep-read + VLM-OCR, photoreal 3D/asset/material generation, music/voice/SFX generation, cinematic-video

generation, roster-scale MetaHuman, the per-region headless PCG/HLOD/Nanite bake, the Gauntlet soak

fleet (QT-16). Ungated and buildable today for all 8 regions: every region page (canon authoring),

every Zone Catalog + bbox, the full GIS→Gaea→Landscape→dress→stage chain, PCG authoring, all StateTree

encounters, all beat/fork runtime, the W-SPACE audit + sculpt, every QT harness, Niagara authoring, HUD.

**This is the load-bearing scheduling fact: ~70-80% of shape (b)'s work is 5090-independent and can

start before the hardware lands.**

• QA surface growth: the QT-5..11 harnesses are DataTable-driven by design — one harness per

defect class, scaling to new rows without re-authoring. So the *test authoring* cost is near-flat.

What grows is the soak matrix: maps 6 → ESTIMATE 14-24, beat-reachability nodes 46 → 145,

boss-invariant cases 16 → 48, save/map-handoff key-carry paths 4 → 12 chapter seams. **ESTIMATE: soak

wall-time scales ~2.5-4× with map count** (each map is an independent load + walk). The five system

debuts each need their own must-fire/must-not-fire fixtures — additive, one-time, not amortized.

• P3-exit delay class: LARGE. Derivation: the per-region legs that scale linearly (terrain chain,

staging, region page, deep-read batch, W-SPACE audit) go ×3 on region count; the map-bound legs (soak,

captures, PCG bake, image-critic sweeps) go ×2.5-4; the five system debuts are additive one-time costs

with no slice precedent to amortize against. Against the measured 4-day structure window for Ch 2-5,

the structure tier alone is a 2.5-4× multiple, and the P1/P2/P3 tiers scale with node count on top.

Naming it in the roadmap's own register: **shape (b) is a full second pass of P1+P2+P3 at roughly

three times the surface, not an increment on the first.**

• The two named long poles that do NOT parallelize cheaply (ROADMAP_POST_5090_TO_SHIP.md P1/P3

honest-risks): deep-read ratification (per-record for high-care) and the QA soak window (N clean

soak-days is a wall-clock floor no amount of fan-out shortens).

Shape (c) — Prologue → Ch 13 complete ← THIS IS THE RULED SHAPE

• Content delta over (b): +2 nodes, +4 unique sites (1 new town/settlement site from the amendment),

+ESTIMATE 27-30 beats (post-amendment), +2 boss rows (+1 conditional catastrophe row), +1 region page

(fairy_realm, which has no template precedent), +14 env-profile cells, +the 18-file amulet integration.

• Content delta that does not show in rows: the invented-terrain path (never run), the sovereign kit

(a second player kit), 12 hero familiar species, the art choice-board pipeline's first non-pilot

consumer, and the "grandest display" bar applied to a node-pair the player sees FIRST.

• P3-exit delay class: LARGE + a HIGH-VARIANCE TAIL. The tail is not large in expectation; it is

large in *variance*, because three of its legs have zero measurement behind them.

• The build-order ruling is the correct variance control and Josh reached it himself: doing Ch 2-13

first means the art pipeline, the presentation ladder, the W-SPACE grammar, and the QA belt are all

mature before the one node-pair that has no fallback path consumes them.

• The collision the task named is resolved, not open. The morning realm directive ("fairy realm after

all nine mythological realms") is superseded IN PART by the fifth sitting: the fairy realm designs and

builds once, at slice-end; the nine mythological realms (Ch 66-74) stay in the post-slice factory

era; the endgame fairy-realm return (Ch 75-77 at the sovereign ceiling) extends the slice's build —

which is a genuine efficiency: the sovereign kit and the realm's art bible get built once and are

consumed twice.

Shape (d) — STAGED Early Access

• This is already the ruled ship model, not an alternative to it. `D-STEAM-FORMAT = RULED (1) EARLY

ACCESS` (Josh 2026-07-24): *"The slice frames honestly as the start of the larger game; chapters ship as

updates; the EA feedback channel feeds the care backstop + density curve."* Pricing direction: ~$20

permanent discount off a $69.99 anchor, AAAAA framing.

• So shape (d) is not "instead of (b)/(c)" — it is the question of where the FIRST EA drop cuts.

Costed honestly, staging moves no total cost; it moves three gates earlier:

• the Josh Gate (P3-exit) fires on a smaller surface, so his first contact with the running game

arrives sooner and his care/feel backstop starts working sooner;

• the P5 factory cadence range — which the roadmap explicitly refuses to fabricate and insists the

first post-slice chapters must MEASURE — gets measured on real shipped chapters instead of estimated;

• real player feedback starts feeding the density curve and the care backstop.
• What staging costs: each drop pays its own P4 chain (Shipping cook + DataValidation + Steamworks +

store update + day-0 patch path). The Steamworks integration itself is one-time. Per-drop overhead is

real but small relative to a chapter's P1-P3.

• What staging risks: shipping Ch 2-5 first means the first public impression is the chapter band that

is *deliberately* the lightest on the density curve, with no companion, no journal, no trade home, and

no origin. That is an honest-framing problem, and EA framing is exactly the tool for it — but it is a

marketing risk, not a cost risk, and it is Josh's call.

• Delay class: NEGATIVE to first-ship, NEUTRAL to total.

Added shape (e) — SHIP EARLY, LAND THE ORIGIN AS THE EA FINALE

Offered because the analysis supports it and the examples-not-limits rule applies. This is (d) applied to

the ruled (c): **cut the first EA drop at Ch 2-5 or Ch 2-8; content-drop Ch 6-13 forward; land

Prologue + Ch 1 as the EA finale**, which is exactly the build order Josh ruled.

• Why it is coherent with the ruling and not a competitor to it: the ruling fixes the SLICE

(Prologue→13) and the BUILD ORDER (2-13, then Pro+1). It does not fix the SHIP boundary. Under EA,

slice ≠ ship-unit. Shape (e) keeps the ruled slice and the ruled build order intact and only asks

whether the first store push waits for all 15 nodes.

• What it buys: the Josh Gate fires on a 4-6 chapter surface instead of 15; the P5 cadence gets

measured for real; the origin's reveal lands as a *finale* rather than a prologue the player reads

first — which is dramatically stronger and is also how the Prologue is already designed

(REALM_DESIGN_PROGRAM.md §1.2: *"built LAST and read FIRST"*).

• What it costs: the store page must honestly frame a mid-arc entry point, and the Ch-13 act-break

("to be continued") lands as a content update rather than as the launch shape.

• **This is a genuine open question the ruling does NOT foreclose, and it is worth putting to Josh as a

short brief with a recommendation** — not as a bare question.

Added shape (f) — the one shape the analysis DOES NOT recommend

A "Prologue-lite": a cinematic or short-corridor Prologue at the front for narrative comprehension, with

the full realm built later. Rejected. It contradicts the Prologue's whole function — the vril CEILING

demonstrated *and felt* so the 75-chapter climb back to it lands (REALM_DESIGN_PROGRAM.md §1.2), and the

sovereign kit is the thing that must be *played*, not watched. Building it twice costs more than building

it once. Recorded here so it is not re-proposed.

---

5. THE CROSS-CUTTING CONSEQUENCES THE SHAPE CHANGE FORCES

These are downstream facts, not opinions. Each one is a plan document that now says something false.

1. The amulet-gift staging is FORCED to Ch 13 by the slice boundary.

AMULET_PROVENANCE_BRIEF.md §7 lists five candidate stagings and **recommends "after the first real

scare"** (chapters past the join), with the Ch-13 join as the *fallback*. If the slice ends at Ch 13,

"the first real scare" is out of the slice, and the arc's emotional payoff — *the protagonist hands

over the thing that cost their mother her life, on purpose, knowing exactly what it cost* — either

stages at Ch 13 or does not ship in the slice at all. **The recommendation the brief carries is

collapsed by the ruling.** This needs re-presenting to Josh, because the brief's own reasoning against

Ch-13 staging (*"the player gives away the object before they have spent much time fearing for the

person"*) is a real craft objection that the slice boundary does not answer.

*Note also:* Josh's question conflated the gift with the join. **The JOIN is Ch 13 (HL_0070, CHAR_0002,

CH13_B01) — his guess is right. The GIFT is a separate, unruled beat.**

2. ROADMAP_POST_5090_TO_SHIP.md P1 entry criterion is now wrong. It reads *"the slice's regions

(flores_island, bali, sumatra_java, angkor_khmer) have their books ingested and triage-scored."* Under

the ruling that is 13 regions, and the DEEP_BOOK_PROCESSING_CONTRACT §4.2 pilot scope moves with it.

3. PRE_5090_BUILD_PLAN_VOL2.md §A.0 is now wrong. *"The whole pre-5090 program has ONE target: CH

2-5 as a full playable vertical slice"* — superseded. The MASTER WAVE QUEUE's "CLOSED" status is

correct for Ch 2-5 and incomplete for the ruled slice.

4. The Josh Gate's surface changes from "AAAAA-polished Ch 2-5" to "AAAAA-polished Prologue→13". The

gate itself is untouched; what it gates is 3.75× larger.

5. The density scorecard bands must extend. check_slice_scorecard.py and the QT-17/18/19 family

currently carry Ch 2-5 early-game targets. Ch 6-13 sit further up the curve; Prologue/Ch 1 sit off it

entirely (the Prologue is a ceiling tutorial and Ch 1 is declared out of corpus scope). Both need

declared sentinels, or the monotonicity tooth fires a false regression — the same class of defect the

realm program already caught for Ch 77's exempt demand_ordinal.

6. T0_Equipment_Registry is header-only with zero rows. The amulet is the first equipment row in

the game. Its magnitudes stay ordinal under the Phase-5M firewall.

7. Fork C (tiered traversal) now applies across 12 earthly regions, not 4. 78 seams type T0-T3; the

ruled "important crossings become staged or open DUNGEONS" doctrine turns inter-cluster gaps in the

spread-heavy regions (rift_valley, congo, west_africa, south_africa) into buildable content. This is

quality-accretive and cost-accretive in the same stroke, and it is already ruled.

---

6. THE BOTTOM LINE, FOR THE RULED SHAPE

**The ruled shape (Prologue→13, built 2-13 then Pro+1) costs roughly 3.75× the current slice in nodes and

ESTIMATE 2.5-4× in maps, and it is the right call anyway** — because the current slice contains none of the

systems (companion, journal, trades) and none of the emotional spine (amulet, mother, origin, the gift)

that make the game the thing Josh authored this week.

Where it will hurt, ranked by risk, not by size:

1. The invented-terrain path for fairy_realm. No DEM, no bbox, no precedent run. The single largest

unmodelled item in the plan. Mitigation: prototype the authored-terrain chain EARLY (it is ungated and

costs no model spend), during the Ch 6-13 window, so the path is proven before slice-end consumes it.

2. The companion runtime debuting in the slice's terminal chapter. Twelve chapters of soak for

everything else; one for ally AI + the rescuer loop + the downed countdown. Mitigation: build and soak

the ally runtime EARLY against a stand-in, so Ch 13 is its integration and not its birth.

3. Eight region pages that do not exist. The measured long pole of the slice was canon-side page

authoring, not terrain. 12,000-42,000 lines, serial per page, critic-gated per section. Mitigation:

this is 100% ungated — start it now, in parallel, on the three-subscription capacity Josh named.

4. The map count, which grows superlinearly with spatial spread. Four of the eight regions are

sumatra-class or worse. Mitigation: run the AOI-clustering decision per region EARLY (it is a cheap

deterministic analysis) so the true map count is a measured fact before P2 sizing depends on it.

5. The soak window, which is wall-clock and does not fan out. N clean soak-days on a 3-4× larger

matrix. Mitigation: the QT-16 Gauntlet fleet is exactly what the 5090's parallelism is for; this is the

one item where the hardware genuinely buys throughput rather than capability.

The one genuinely open question the ruling does not foreclose, worth a short brief: shape (e) — under

the already-ruled EARLY ACCESS format, does the FIRST store push wait for all 15 nodes, or does it cut at

Ch 2-5 / Ch 2-8 with Ch 6-13 as content drops and Prologue+Ch 1 as the EA finale? The ruled build order

already produces exactly that artifact ordering for free.

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