FORKB_RESURRECTION_BRIEF_2026-07-27.md

pipelines/FORKB_RESURRECTION_BRIEF_2026-07-27.md

FORK B — THE DECISION BRIEF

The direction (Josh, verbatim intent, 2026-07-27): *"after death, allow the user to choose to

revive at last checkpoint, or wait to see if one of their support companions who may have been

trained or taught resurrection may be able to revive the user, but this can only happen out of

combat."*

Synthesis of five read-only deep dives (canon-integration · mechanics/economy · presentation ·

engine/schema · adversary), 2026-07-27. Per the standing decision protocol: this brief presents

HOW, not whether. Analyst disagreements are preserved as named FORKS, never averaged.

Source files (scratchpad): forkb_canon_integration.md · forkb_mechanics_economy.md ·

forkb_presentation.md · forkb_engine_schema.md · forkb_adversary.md.

---

0. THE FOUR FACTS ALL FIVE ANALYSTS AGREE ON

These are not proposals. They are positive-controlled findings, converged from independent passes,

and they reshape the question:

1. Resurrection is ALREADY RATIFIED CANON — it was never absent. T0_Ability_Tree_Registry

carries A_A_009 Breath of Life (T5, L1, Air/Heart), A_C_003 Vril Revival (T5, L1/L2,

Ether-Crown), A_C_004 Vril Mass Revival (T7, L1, revival_aoe), all reading *"restores a

downed ally to combat consciousness with reduced vitality."* Collective Surge (T7) explicitly

relays a Mass Revival across a formation from any linked ally's position — canon already

presupposes allies hold revival-class casts. The pre-Claude GDD authored the whole ladder, and

the antagonist family has revival too (Death's Scythe: *"opponents defeated with this weapon

cannot be revived by antagonist family means"* — proving revival-DENIAL is already a designed

counter-lever). The only missing arrow in a four-way symmetric system is ally→player. Josh

asked for the one hole filled. Any framing of this as "adding resurrection" is factually wrong.

2. Josh's "trained or taught" speaks the corpus's own vocabulary, one-to-one. *Taught* =

Lane B, the quest-gated transformative tier (T0_Bonus_Pool_Registry.unlock_gate_quest_ref —

the column already exists). *Trained* = Lane A, the proficiency-depth ladder + INT + bond +

spec purpose. Nothing needs inventing at the acquisition layer.

3. The out-of-combat constraint is derivable from canon, not arbitrary — and it is genre-correct.

The seven infinite-vital-energy devices bond to the protagonist's vril-seat equipment only,

so a companion has no Enhanced Ability Mastery and cannot hold a Crown-seat channel under combat

pressure. Independently, the rule maps 1:1 onto the WoW combat-rez / out-of-combat-rez split.

Josh arrived at the correct line himself; the loop's job is to state it in-fiction as capability,

never as a UI rule.

4. The runtime is 0% built and the ally runtime is ZERO. Positive-controlled across

C:/dev/Humanity/Humanity/Source: \bAlly|Allies → 0 hits; Companion → 10 hits, all data

plumbing; Checkpoint → 0 hits; bInCombat|IsInCombat|CombatActive → 0 hits. No ally

pawn, no ally ASC, no ally AI, no party frame. THE FAMILIAR TRAP: ~20 Familiar hits in

Source are all the boss composure node (Ability.Mode.Familiar, Composure_Familiar,

UHumanityFamiliarAbility) — not one is a creature. Flag this to every downstream agent.

Also: "revive at last checkpoint" has no checkpoint to revive at. The shipped path is

RespawnAtNearestSite() → nearest AHumanitySiteMarker by squared distance. Josh said

*last*; the code does *nearest*. That is a semantic fork, §2.7.

---

1. THE SHAPE OPTIONS — four coherent parameterizations

Each shape is a complete, internally consistent answer across four axes: the choice surface ·

what the wait costs · where you come back · how the capability is acquired. They nest:

A ⊂ B ⊂ C. D is the cost-axis alternative.

SHAPE A — THE DIEGETIC HOLD (minimum coherent; condition-only uncertainty)

• Choice surface: no menu, no modal, no two-button prompt. HP 0 → the camera inherits the fall

and settles low (arena-biased yaw, never the player's last look direction); a single quiet

hold-to-let go is available from the first frame to the last. Not letting go IS the wait —

the player never "picks" wait, so there is no second UI object and no greyed tease. No input =

wait (the accessibility floor); the checkpoint is an active press.

• What the wait costs: the world-advance clock keeps running (the ruled Hades doctrine, already

in force, made visible) + the ally's vril, refilled only at the vril site.

• Where you come back: in place, at reduced vitality (canon's own word, three ability rows).
• Acquisition: taught (a Lane-B quest gate on a support pool) + trained (INT + spec purpose +

proficiency depth). Deterministic on conditions — no roll.

• PROS: zero canon amendment under the "let go / hold on" reading (§2.1); satisfies ruled Fork A

by construction (no cut the game imposes); natural-voice-native; no-dead-end by construction

(Option A never fails); the smallest thing that is honestly Josh's direction; buildable in ~4-5

waves + one mint without any ally runtime (the wait resolves truthfully to "no one could

reach you").

• CONS: until party-combat-continues-after-you-fall exists, the out-of-combat constraint is

nearly FREE (enemies de-aggro from a body within seconds) — the constraint is *correct* but not

yet *meaningful*, and must be reported that way. The wait can be dominated (§3.1). In the

Ch 2-5 slice the branch is inert.

SHAPE B — THE HOLD WITH TEETH (A + the enemy side) ← RECOMMENDED

Everything in A, plus:

• downed_target_policy on T0_Creature_Roster (rides the missing aggro block, gap #1):

disengage (generous default) · execute (finish the body after a visible windup) ·

guard (hold over the body, refuse to leash — combat never ends, so the wait cannot resolve) ·

ignore.

• Allies do not initiate new engagements while the player is downed — they finish what is on

them and come. One rule; closes the AFK-farm-from-the-floor exploit; improves the fiction.

• PROS: execute is the single highest-value value in the whole analysis — it makes the wait

genuinely able to end *worse* than letting go, it is register-correct for rated-M, and it creates

the feature's best moment (your ally peels the executioner off you) with no die roll. It also

closes the "enemies wander off ⇒ free revive" leak that makes A's constraint hollow.

• CONS: promotes DESIGN_GAP_REGISTER #1 (non-boss engagement/aggro spec, class MISSING,

owner W-SPACE) from "highest-leverage" to blocking. The execute windup is authored content.

SHAPE C — THE THREE-ROUTE ECONOMY (B + path parity + measured balance) ← the destination

Everything in B, plus the three routes to one mechanic:

Plus a MAJOR pick cost on a support pool (taking it means not taking a damage/utility major),

a new SPEC_VRIL_REVIVE_ALLY row in T0_Spec_Purpose_Registry, and a **named per-path pick-rate

assertion** in the Phase-5M/5H build-space sweeps.

• PROS: kills the sharpest latent defect in shipped canon (§3.2 path asymmetry — all 8

heal/revival rows across 185 are L1 or L1,L2; zero at L3/L4/L5, while 13 L3 rows and 7 L4,L5

rows exist elsewhere, so the bands are expressible and the zero is real). The craft route is

bench-independent, so a DEPARTED healer narrows options without deleting the mechanic. Makes the

whole thing measurable rather than asserted: near-100% pick rate in one band = meta-collapse;

near-0% = a dead pick; both are defects, fixed by widening competitors, never by removal.

• CONS: three lanes to author, gate, and balance; the craft route risks becoming a

consumable-hoarding economy if the drop rates are wrong; the largest mint surface.

SHAPE D — THE CONSERVATIVE COST VARIANT (the revive returns you to the well)

Same surface as A/B, but a successful ally revive still relocates you to the checkpoint — what

the ally saves you is the world-advance delta, not the walk.

• PROS: completely kills the attrition exploit (§3.3) with one rule; smallest engine surface

(no in-place restore path); leaves graduated persistence untouched.

• CONS: it guts the fantasy Josh described — *your healer reaching you* — and replaces it with a

bookkeeping difference the player will never feel. Three analysts rate it weak.

• Recommended disposition: NOT the default. Keep it in the toolbox as a **per-encounter

override** (the ally_revive_policy column, §3.3) for the rare authored arena where in-place

revival would break a boss economy.

---

2. CANON-INTEGRATION RULING POINTS — the loop vs Josh

2.A GENUINELY JOSH'S (brief, never bare — presented here with recommendations)

**J-1 — THE CONSCIOUSNESS-WITHDRAWAL RECONCILIATION. This is the gating decision: every

player-facing word depends on it. ⚠ ANALYST FORK — kept visible.**

The ratified model is *"consciousness withdrawal to the last attuned recharge source"*

(BATCH1_RULINGS.md L50-51), deliberately a substrate event, not a UI event.

• **Three analysts (canon-integration, presentation, mechanics) read withdrawal as a PROCESS WITH A

DURATION** — so the wait is simply the interval before it completes, and Fork B "intercepts a

withdrawal in progress." Under this reading nothing is amended: the choice is *let go* vs

*hold on*, the timeout resolves to the checkpoint because the withdrawal completes on its own, and

no new metaphysics is minted.

• The adversary calls this a genuine contradiction, not a nitpick: if consciousness has already

withdrawn to the well, the choice is offered to someone who is not there and both options collapse

into one location. It offers three repairs — (1) make withdrawal explicitly graduated (amends

BATCH1 item 4), (2) introduce a tether the healer works on (mints a new vril concept), or

(3) reframe the options as "let go / hold on" (costs canon nothing).

• RECOMMENDATION: repair 3, ratified explicitly rather than assumed. It is the reading three

analysts already hold, it costs zero amendment, and it is the most emotionally loaded framing of

the two options. But the loop should NOT apply it silently: the ratified sentence does not say

"graduated," and writing that word into canon is a change to a Josh-ratified model. **Ask him to

bless one sentence, then everything downstream is execution.**

J-2 — THE ASYMMETRY / FALLEN-vs-DEAD LINE. ⚠ ANALYST FORK — kept visible.

• Canon-integration proposes THE ASYMMETRY RULE as a hard line: the revive reaches the

protagonist and nobody else, ever. Grounds: T3_Familiars_Named L919 already rules the Ch-58

reunion resolved *"not through resurrection"*; named/persona'd NPCs stay permanently dead

(community-ripple doctrine, DESIGN_GAP_REGISTER #13). Canon has already declined resurrection as

a narrative resolution mechanism.

• Three analysts require the opposite at the combat layer: canon's own Vril Revival targets

*downed allies* and Mass Revival targets the whole fallen team, so DOWNED must be party-wide or

the player's own ratified kit is meaningless. Engine's B2 recommends **allies go DOWNED, never

dead, in combat. The adversary draws the line at the community-ripple / persona-death event**:

before it fires you are *fallen*; after it fires you are gone.

• RECOMMENDATION — these reconcile, and the reconciliation is the ruling: *DOWNED is a combat

state and is symmetric across the party; DEAD is a story event and is asymmetric and permanent.

Permanent death is never a combat outcome for a persona'd character. The revive acts on DOWNED,

never on DEAD, and the boundary is the community-ripple event — a write that already exists.*

This preserves the Ch-58 ruling, the community-ripple doctrine, Mass Revival, and the gear-return

contract simultaneously. **It is §17-adjacent cardinal territory (a hard line about death

permanence) — recommend Josh ratifies the sentence rather than the loop applying it.**

Enforcement note: loyalty_state has no dead value and WS_028's event enum has no KILLED

(BG3-2). Fork B raises that gap's priority because the player will now visibly notice the

asymmetry.

J-3 — THE HEALER'S CHILD (D-2 F-STAR). The emotionally correct reviver — the Manggarai

healer-elder's daughter, the protagonist's primary human companion from Ch 13 — is canonically

barred: *"no combat ability, vril, or weapon is gained"* (T3_Core_Characters §3.4, per the D-2

F-STAR ruling 2026-07-19).

• Reading A (default, applies today): a vril revival is a vril capability; she cannot be the

reviver; Fork B lives on the other companions and the familiars.

• Reading B (the reframe): D-2 F-STAR's context is *combat* capability; an out-of-combat,

non-vril, Herbalism/medicine-grounded resuscitation is not a combat ability and is the exact

expression of her mother's tradition — reachable without granting her vril, a weapon, or any

combat presence. Her parallel-analytical way of knowing is already canon (Ch 16 Indus).

• RECOMMENDATION: hold A as the default, brief B for a ruling. B is the highest-emotion version

of this feature in the entire game. It is a characterization-canon call on a Josh-authored ruling

about a Josh-owned character — not a creative-within-vision apply.

J-4 — THE EVIL-PATH ANSWER (path parity). ⚠ ANALYST FORK on who rules it.

Mechanics calls minting the L4/L5 revival lane *"not a fork — an authoring task the loop rules and

applies,"* citing T1_Ability_Tree L848 which already declares the lane exists. Engine calls the

rows *"cardinal-adjacent (band-gated capability) — Josh's call on the routing."* The adversary adds

a third route entirely: HL_0076 already bands incoming healing (L1 +10%, L2 +5%, L5 +10%

vulnerability), so the companion comes regardless and the *magnitude* rides the band — which

routes cleanly around the §8.11.8 firewall (no condition_expr reads loyalty; the magnitude reads

the band exactly as all healing does).

• **RECOMMENDATION: the loop authors the lane (it is canon-declared and the shape is canon-supplied

— same ability, higher vril cost, karmic-debt residue per T1_Story_Spine L114), and Josh rules

ONE thing: whether path parity is delivered by a separate L4/L5 route, by HL_0076 magnitude, or

both.** That single choice touches the equally-weighted-endings architecture, which is his.

J-5 — CAN THE PARTY WIN A FIGHT WITHOUT YOU, AND CAN THEY FINISH A BOSS? The adversary names

this *"the largest single cost, invisible in the direction's phrasing."* Out-of-combat-only means

either the mechanic almost never fires (dead feature) or the fight must be **capable of ending

without the player**. If the party can finish a boss: who gets the win credit, the win_reward, the

mastery credit, the tell-reads?

• RECOMMENDATION: yes for ordinary encounters, and the chain-meter bosses answer themselves —

their win_condition is a player-performed read (5-7 counter-class reads in a row on a pip

meter that wipes on error), Rangku's FREE route is an integrity choice only the player can make,

and Naga Padoha's resolution is a cosmic subdual. **An ally cannot satisfy any of them, so those

encounters cannot end while you are down** and the wait resolves to attrition then the checkpoint.

The design disarms itself exactly where an ally-undo would do the most damage. Josh's call is the

scope statement, not the mechanism.

2.B THE LOOP RULES (creative-within-vision — apply and show the path)

• The diegetic-hold presentation, its input grammar, and its strings under the natural-voice

doctrine.

• The out-of-combat predicate's definition and its provenance semantics (§3.4).
• The downed_target_policy enum and which creatures carry which value.
• Capability-not-character authoring: which pools, which gates, which spec purposes, the

T0_Ally_Behavior_Policy rows, the T0_Familiar_Bond_Ability rungs.

• The Withdrawal-clock-runs-while-down anti-scum rule (canon-native, free).
• The zero-bond-movement anti-exploit rule (§3.6).
• The death-economy plumbing: the journal observation fires on both paths; the consolation line

is replaced on the revive path by the companion's own words.

• The unlock sequencing and its diegetic first beat.
• The reveal-discipline baseline entry for the death surface + its critic pass.
• Care ruling (already delegated, ruled here): **Fork B stays fully available in every high-care

chapter** — removing it would be exactly the care-tier inflation the doctrine names as a defect —

but the wake is routed off protected sacred ground (§5.7 already models sites as having their own

consent), the voice is culturally registered per region, and no borrowed rite is ever depicted.

2.C A CORRECTION THE PARENT BRIEF NEEDS ⚠ ANALYST CONTRADICTION — unresolved, director call

The parent brief's premise *"protected chapters carry 1-HP withdrawal"* is inverted per the

canon-integration analyst: defeat_protected / the 1-HP clamp is a BOSS column (the boss

withdraws at 1 HP; Naga Padoha is the slice's one clamped boss), not player-death protection.

The player can and does die inside a protected encounter.

But the engine analyst independently specifies a player-side clamp as build item A8 (*"missing

today"*) and QA test T10 (*"protected chapter clamps player to 1 HP; S2 never opens"*).

These cannot both be right. Recommendation: the canon analyst's read is better evidenced

(FUN_REBUILD_PLAN.md L372, PRE_5090_BUILD_PLAN.md L349-350 both scope it to the boss row) —

so A8/T10 should be struck from the build list unless the director's own re-read of the

deciding artifact says otherwise. Per the standing agent-frame-claims rule, the director re-reads

before landing.

---

3. THE EXPLOIT / BALANCE GUARDS THAT MUST SHIP WITH IT

Non-optional. Each is cheap; each closes a defect an analyst demonstrated.

Two firewalls the feature must not breach (Gate 29 check_ally_build.py B5 already enforces

them): (i) behavior quality is a BUILD property, never an integrity property — **no policy

condition_expr may read a loyalty ws key, therefore the revive policy may not be

loyalty-gated; (ii) policy_source is a closed enum with no player-authored member — the

player may not author the revive policy.** They build the ally; the ally decides. That is also §1.4's

existing familiar rule (*"the player cannot override this behavior"*).

⚠ ANALYST FORK — the loyalty question. Canon-integration's clean split says **WILL she? rides

THE BOND + integrity band** (and recommends a WAVERING companion still comes — the best beat the

fork can produce). Mechanics/engine say firewall (i) makes a loyalty-gated policy **structurally

illegal. Recommended reconciliation:** loyalty decides whether she is *still with you at all*

(the roster, upstream — departed/estranged cannot revive by construction), and it decides

nothing at the policy layer. Willingness below that is not bond-gated; a soft floor (a

low-bond ally tries and does not reach you in time) preserves the AAAAA no-dead-pick principle and

keeps the bond legible through play instead of through a tooltip. This keeps both readings' intent

and breaks neither gate.

---

4. THE PRESENTATION SHAPE UNDER RULED FORK A

**The headline: Josh's direction SETTLES Fork B by splitting it per-branch — an answer none of the

four escalated options anticipated, and strictly better than any of them.**

PRESENTATION_DOCTRINE.md §0.B offered four competing camera treatments. They become sequential

states of one machine:

The hard cut becomes PLAYER-AUTHORED, and that retires Fork A's strongest objection to Opt 1.

The objection was that *"making death the one routine hard cut trains the player to read a cut as

failure, which then poisons every other legitimate cut."* A cut the player pressed is not a cut the

game imposed — it reads as the player's own impatience, so it never transfers dread onto the

authored cuts. **Fork A's scarcity doctrine survives intact and the one exception it had to carve

becomes an exception the player writes.**

The six states. S0 THE FALL (camera inherits the fall, settles low, arena-biased yaw — a

build requirement, not a detail: a camera that keeps the player's last look direction will stare at

a wall for half of all deaths and the wait will read as broken) → S1 THE LOW HOLD (choice offered

frame one, world not paused, not desaturated; treatment at the EDGE — a slow closing

vignette that doubles as the diegetic timer — never from the centre of frame) → S2 THE WAIT →

S3 THE RISE (diegetic, camera rises with the body, so CC-3 is satisfied trivially) / S4 THE CUT

(fastest transition in the game, CP-6 unchanged) / S5 THE DISSOLVE.

Three canon-binding presentation facts fall straight out of the ability rows: the revive returns

you at reduced vitality (readable on the body — breath, walk speed, frame edges — never on a

bar); the revive is instant cast (no channel bar, no progress meter on a death surface); and the

dark path is not denied the revive, it gets a different one (*"alternate revival mechanics

through different element routing"* is an explicit instruction that the revive IMAGE forks on band).

Three things this buys that nothing else does:

1. THE WAIT MAKES THE INT AXIS WATCHABLE — its only real showcase. Low INT: the ally is still

swinging, or standing where it was; it does not read that the fight ended. Mid: it turns,

registers the body, comes. High: it breaks off early, clears or pulls the last threat off

your body first, and arrives with intent. Josh's INT-as-AI ruling (2026-07-26) is on the

comparator roster as a *"day-old ruling with zero teacher"* — better AI is normally felt as

ambient competence, the hardest thing for a player to attribute to a stat they bought. *"Your

healer noticed you go down, disengaged, made it across the arena, and got you up"* is the single

most legible possible payoff for an invested stat.

2. THE HONEST-FAILURE PRESENTATION — load-bearing, not polish. A wait that resolves into nothing

is indistinguishable from a bug, and *not being able to tell* is the failure. **Rule: waiting

always shows you something.** Ally still fighting → you watch them try and fail to disengage.

Cannot reach → the near-miss. No vril → they arrive, kneel, and cannot. Never learned it →

they arrive, kneel, hold on, and stay with you while the draw takes. Companion dead → empty

ground where they would have been; say nothing; the absence is the memory. **The "arrives and

cannot" cases are better presentation than a success, because they carry the relationship. If

these images are cut for budget, the "it's a slot machine" objection becomes correct and the

feature should not ship.**

3. The frequency problem solves itself. A fixed ceremony is the AAA mistake; a fixed fade is

invisible. Here the fast player presses once and takes the fastest transition in the game, and

the patient player gets a beat that is different every time — because the beat is *the live world

around the body*, and the live world is never the same twice.

Do NOT gate the OPTION on combat state — gate the ACTION. "Out of combat only" invites a static

availability check at HP 0 (enemies alive → hide the wait). That is the wrong build: the fight can

END while the player waits, most often because the companion finished it, and *watching your ally

kill the last enemy and then come for you is the single best expression of this entire system.*

The surface stays constant; the world decides. This also makes the rule invisible — the player

never learns there is a rule, they just observe that nobody kneels next to them mid-fight.

Availability is communicated by the CAMERA, not by a greyed menu item. If someone is out there,

you can see them. If nobody is out there, you see empty ground. Show-don't-tell, on the exact

surface where a UI availability flag would have been most tempting. **All analysts reject the greyed

tease.** (Residual fork: engine prefers "shown with a truthful diegetic read"; mechanics/canon

prefer the option simply not existing until someone is there. The camera-bias answer satisfies both

— the surface is constant, the frame is honest.)

THE DEFAULT ON NO INPUT MUST BE "WAIT" — three floor-level reasons: a player who is stunned,

distracted, away from the pad, or on assistive input is not punished; the timeout resolves WAIT into

the checkpoint anyway, so no-input can never softlock (CC-8 + the no-dead-end floor, satisfied by

construction); and it preserves the beat for the player who did not know there was one. **The

checkpoint is an ACTIVE press.** That ordering is what makes the machine both merciful and fast.

Attunements (ratified, judgment-free) carries a presentation default — pre-select wait/return —

governing which option is pre-selected, never availability, so it stays a preference and not a

content gate.

Accessibility floor (Attunements is ratified canon, so this has a home): options are TEXT +

distinct input, never colour-coded or icon-only; the companion approach needs **two independent

channels** (silhouette/motion in frame — which is what the S0 camera bias buys — AND audio); the

diegetic vignette timer needs an explicit duration-indicator backstop for low vision; **the dark

hold → bright vril close-up is a photosensitivity risk by construction and needs a

reduced-intensity variant — a hard floor item on the most-repeated surface in the game**; motion

sliders inherit; no hold-to-confirm-only, no twitch window, full remap, one-handed reachable.

Strings — the natural-voice gap is LIVE and shipped. Today's surface reads

"DEFEATED (no fail wall) -- %s" (HUD L548). Three defects in one greybox line: **"no fail wall" is

internal design vocabulary on a player-facing frame** (CP-3 — the player must never learn the game

has a *policy* about fail walls; naming the mercy destroys it); "DEFEATED" is system voice; and the

consolation payload is system voice too (death_reward = *"logs his stalk cadence for the retry;

unlocks a polo-corruption lore fragment"* — every load-bearing word is a machine's word). **The

2026-07-27 rewrite pass did not reach it** — it covered the 96 sidecar strings + boss display names,

and death_reward is neither. The highest-frequency player-facing text in the game is still in

machine voice, and the natural-voice rewrite makes it strictly *better* because it makes it teach:

*"He always steps left before he lunges. I saw it that time."* Same word count; the first promises a

log, the second is the log and hands the player the counter.

Recommended option pair (PROPOSAL-tier seeds; examples-not-limits): "Not yet." / "Go back." Two

opposites, both what a person actually thinks, neither naming a system, and — critically — **neither

promising rescue**, so the honest-failure cases do not read as broken promises.

Strings to refuse: anything naming the mechanism (withdrawal, vril site, consciousness);

anything naming the ability by system name ("Breath of Life", "resurrection" — fine as an internal

token, never on the surface); anything promising ("someone is coming"); and the literal phrase

"content to wait"/"content to watch", a live advisory signature in

harness/check_reveal_discipline.py WHO_SOFT.

**Reveal discipline — the death surface is the highest-frequency player-facing text in the game and

it fires from Ch 2.** Five risks, in severity order:

• R1 — the L3 theology leak. Any string that *explains* what death is (substrate, vril,

withdrawal, tether, consciousness) hands the player Layer-3 metaphysics on a surface they will see

fifty times. The return is shown, never explained. Canon models this perfectly: four revival

ability rows describe *what happens* and never *what it means*. Match that restraint exactly.

• R1b — the adversary's sharpest single finding. Every instinct a competent author has on a

death screen — a whisper, a pulse, a light, a sense of being held, the collective reaching for you

— is exactly HL_0046's Grand Sage communication vocabulary (visions, intuition pulses,

vibrational imprints), deployed on a surface visited dozens of times before Ch 76. HL_0044's

protection assumes hints are rare enough to police individually; **a death screen played 200 times

is a drip. check_reveal_discipline.py reads canon artifacts, not runtime presentation — this

is a MAJOR-class hazard the existing gates would not catch.** The death/wait surface needs its own

named entry in the reveal-discipline baseline.

• R2 — Thread 6 contamination. Dying-Rising God has authored anchors (Osiris Ch 15, Eleusis

Ch 23, the Duat Ch 69, the Realm Road Ch 70-74). A revive dressed in dying-rising imagery — light

from above, mythic register, resurrection language — pre-empts authored beats with a routine loop.

**Keep the revive vital and medical, never mythic. Someone catching you, not someone raising

you.** The resonance is allowed to *exist*; a player who read the Osiris material and then watches

someone gather her up will feel it. That feeling is the payload. Naming it destroys it.

• R3 — the boss name on the consolation. BE_0002 carries boss_name "Rangku Wera, the Hollow

Hunter" and display_name "the one who tracks me." The surface draws the diegetic name or no

name, never the registry name.

• R4/R5 — the Ch-38 antagonist-pattern leak (even invoked-and-negated agent-nouns leak) and

the Ch-58 HELD familiar reveal. The death-and-wake loop rhymes with the Ch-2 opening

(*"Get up. Climb toward the light."*) — a free presentation asset with a hard constraint: it must

never re-stage the familiar's death image or gesture at it.

• Astral framing is a trap. "You float above your body" is diegetically available only from

Ch 15 (Astral Projection) and leaks a Crown-seat capability the player does not have. Pre-Ch-15

the read is body-and-senses: the world greys, sound muffles, edges go soft. **Not a soul

leaving — a vessel emptying.**

• The companion's revive line is the single highest-risk string in the feature (a healer talking

about bringing someone back is one adjective away from theology). Relief, reproach and exhaustion

— never cosmology. It goes through the natural-voice pass with string_status tracked AND the

reveal gate.

• The journal is a SECOND artifact — do not double-present. The on-screen consolation is the

observation's first line; the journal holds the full entry. One authored string, two lengths, one

write. Timing gift: the journal activates at Ch 13 and Ntala joins at Ch 13 — both halves of

the mature death surface arrive in the same chapter, so the surface can be designed deliberately

as *the thing that grows*: bare and solitary early, a record at Ch 13, a witness who can reach you

later.

Doctrine deltas to land: PD §0.B is superseded, not amended (the four options become the

states; the fork's field stays as provenance); the §3.2 Death/defeat row is replaced;

CP-6 needs one clarifying clause — the WAIT state can outlast the arc-reveal hold, so it must be

declared a player-paced live-world state outside the hold-class ordinal ladder (closer to CC-9

dwell), or a future author will read a contradiction into a ruled doc; CP-2 (everything skippable)

now binds two exits; BUG-0017's single-framed-overlay invariant has an unruled collision —

dying during a held scene or with a fork pending — and needs a precedence ruling (recommended: the

defeat surface outranks both). T1_UI_UX_Spec is a 2-line empty stub; the death choice surface

has no canonical home and is its natural first real item, alongside the vril-polarity readability

floor already earmarked for it.

---

5. WHAT IS ACTUALLY REQUIRED — the honest build list

Sizes in this project's own wave idiom (a wave ≈ a few files + its own tests + its own commit and

gate ritual). ENGINE/CANON declared per the standing engine-reuse rule.

5.1 BUILD-NOW — no ally runtime required, fully testable today

BUILD-NOW total: ~4-5 waves + one mint. A1, A4, A5 and A7 are all things the slice needs

regardless of whether the revive branch ever ships — which is the strongest scheduling argument

in the brief. With the ally runtime absent, the wait branch resolves to *"nobody could reach you"* —

honest, no-dead-end, fully testable, and it becomes the real mechanic the moment allies exist.

Nothing is wasted and nothing is faked.

5.2 POST-SLICE — the ally runtime (the real cost centre)

5.3 5090-ERA / GATED

5.4 THE SCHEMA MINT — every mechanic to the row that must carry it

Canon edits (tier docs, critic-gated): T1_Ability_Tree §11.7 (the ally-cast/out-of-combat/

protagonist-target case + the no-EAM reason) and §5I.2 (the spec purpose); **T1_Integrity_Paths

§8.11 — the owning home** (the CAN/WILL/HOW-WELL split, the loyalty-state legal set, the

asymmetry/fallen-vs-dead hard line, the zero-bond-movement rule); §8.11.5 (the familiar grant

shape); T1_Combat_System_Spec failure states (the second door, the density coupling, the

out-of-combat condition); PRESENTATION_DOCTRINE §0.B + §3.2 + CP-6; T1_UI_UX_Spec (its first

real content); docs/DOC_MAP.md row + a named consumer for every new artifact.

5.5 QA TEETH

**Engine (UE automation, existing latent-command idiom) — 11 Tier-1 tests, each with a must-not-fire

twin: choice offered at HP 0 (and the 1.45 s auto-respawn timer must NOT be armed**) · checkpoint

return lands at the anchor, not nearest-marker-by-distance (a nearer marker must NOT win) · wait

blocked in combat over REAL ticks · wait resolves out of combat → revive at reduced vitality ·

wait impossible fails FAST (must NOT run a timer out) · boss-fight wait refused/unresolvable

(both killable and protected classes) · downed pawn contract (**an ALIVE player at 40% HP MUST still

satisfy AP_0001) · teardown does not fake out-of-combat** (extend GuardianTeardownOnEndPlay)

· ledger appends exactly once, idempotent under re-possession · strings resolve from data (extend

FHumanityCH02SpoilerSafeStringsTest) · the hysteresis test — the predicate must not fire during

a momentary LOS break (*the one most likely to be wrong and least likely to be noticed*).

Canon gates: Gate 29 check_ally_build.py (B3 predicate lint + B4 disposition legality +

B5 both firewalls) · NEW tooth — revive-capability provenance (every taught_capability_refs

entry resolves to a row with ability_class ∈ {revival, revival_aoe} and a satisfiable band lock,

positive-controlled so an unreachable resolver reports as unreachable, never as a clean zero) ·

NEW tooth — the evil-lane hole (if the L1/L2 lane is a live affordance and L3/L4/L5 is empty,

FAIL or WARN with a named worklist — the executable form of the no-dead-end floor) · **NEW tooth —

the asymmetry firewall** (no row may carry revival_direction_scope=to_protagonist without

out_of_combat_only=true — Josh's constraint made unspellable-to-violate in data) ·

Gate 26 check_build_space.py (enroll the spec + a dominance tooth: ~100% pick rate is a defect

symmetrically to a never-picked option) · Gate 9 check_no_dead_end.py (assert the

Option-A-never-fails invariant explicitly) · check_reveal_discipline.py +

check_grand_sage_silence.py scoped to the new death-surface strings, **with the surface given its

own named baseline entry** · check_registry_extensions.py · check_ws_value_form.py ·

registry_fidelity.py --emit-baseline in the same commit as any registry edit.

---

6. RECOMMENDATION

**Build SHAPE B, staged: land Shape A's BUILD-NOW block now, design the surface so B and C are data

additions rather than rewrites, and commit to C as the balance destination before the feature is

ever described as shipped.**

Concretely:

1. Now (pre-slice, ~4-5 waves + one mint): A1 checkpoint anchor · A2 the state machine and the

diegetic-hold surface · A3 the downed contract · A4 the combat-state authority with provenance ·

A5 the world-advance delta · A6 the mint · A7 the soak lane. **Four of these seven are things the

slice needs anyway. The WAIT branch ships present and honest**, resolving to "no one could

reach you," because no ally runtime exists.

2. Report it register-honestly. Until B2 (party combat that continues after you fall), the

out-of-combat constraint is *correct but nearly free* — enemies de-aggro from a body within

seconds. Ship it anyway (retrofitting it later is expensive) and never let a play-quality

claim ride on it pre-B2.

3. Post-slice: B1 → B2 → B3 → B4 → B5, at which point Shape B is real.

4. Before any "the revive is balanced" claim: Shape C's three routes + the per-path pick-rate

sweep assertion.

Why B over A: A's wait can be dominated and its constraint is hollow — execute and the

no-new-engagements rule are what make the wait a genuine risk rather than a delay with a menu on it,

and they cost one enum column and one behavior rule.

Why B over C now: C's three routes are the balance destination, but they are a mint-and-author

program whose correctness is only *measurable* at Phase-5M/5H sweeps. Shipping C's rows before the

sweeps exist is authoring against an unmeasured target.

Why not D: it trades the exact fantasy Josh described for a bookkeeping difference the player

cannot feel. Keep it as a per-encounter override only.

THE STRONGEST OBJECTION

**"This is a multi-wave engine program, a greenfield ally runtime, a blocking design gap, and a

17-row mint — bought for a branch that cannot fire once in the only thing Josh will play."**

It is true and it is the honest shape of the cost. Every analyst converged on it: Josh ruled

Ch 2-7 stays SOLO as built, the slice is Ch 2-5, the revival ladder is T5 Philosophus of a

twelve-tier climb, and the earliest plausible companion window is Ch 13 (Ntala) with the earliest

*functional* one later still. Fork B has zero surface in the Josh Gate slice. So the feature

Josh just directed is a feature he will not see in the build he is waiting to play — and if it is

mis-built as a greyed second option, it is worse than absent: it is a **taught-then-broken

affordance**, a promise the game cannot keep for eleven chapters.

The answer, and why it does not overturn the recommendation:

• Four of the seven BUILD-NOW items are slice requirements on their own merits. There is no

checkpoint system (grep Checkpoint → 0). There is no combat-state authority, and four ruled

canon rules (RAGE decay, STAMINA regen, the loadout-switch ban, every ally predicate) are

silently unenforceable without it. Nothing is written to world state on death, so the ruled

Hades loop is unbuilt. And player death has zero test coverage while 55 automation tests

assert boss defeat. Those get built either way; Fork B is the reason they get scheduled.

• The zero-surface fact is a design gift, not a defect. The death surface is *bare and solitary*

early, gains a record at Ch 13 (the journal activates), and gains a **witness who can reach

you** later. That is a true story about the arc — the world starts empty and fills — and the first

time someone can come for you is a story beat. Built deliberately, it is better than the mechanic

being available from Ch 2. Discovered as a bug report, it is a disaster.

• The alternative is worse. The timid answer here — keep the fade, revisit later — ships a

0.55-second greybox placeholder as the death presentation of a 30M-word game, which is exactly

how *"a greybox constant becomes shipped canon without anyone ruling it."* And it forfeits the one

surface where the party system, the INT-as-AI axis, the healing spec's apex, the loyalty canon's

only mechanical gain, and the Presentation Ladder all land at once. Per-frequency, the death

loop is among the highest-leverage presentation surfaces in the project.

Second-strongest objection, worth naming: *two options at death is one option too many* — the

player answers this question hundreds of times, and a decision made hundreds of times with the same

answer is a tax, not a decision. Answer: the no-input default is WAIT and the fast path is one

press, so neither audience is taxed. But this is the objection I would most want **measured in the

QA loop's feel telemetry rather than argued** — it is a wall-clock question about a repeated

surface, which is exactly the class the feel strips measure.

---

7. THE MMO BRIDGE

SCALES cleanly — the architecture is already there:

• The raid rez is ALREADY IN CANON. A_C_004 Vril Mass Revival (T7, revival_aoe, *"all downed

allies within range return to combat consciousness simultaneously"*) is a raid rez by

construction. And Collective Surge (T7, Ch 70-77) explicitly relays Mass Revival across a

formation from any linked ally's position — a raid-rez-relay primitive authored before anyone

asked for one.

• The role taxonomy needs nothing added. T0_Role_Axis_Registry rows carry arena=both;

§8.7 states *"One taxonomy, two arenas."* ROLE_SINGLE_HEAL / ROLE_GROUP_HEAL already span.

• The out-of-combat constraint maps 1:1 onto the WoW combat-rez / out-of-combat-rez split.

Josh's rule is the genre-standard rule, arrived at independently.

• §8.8's Enhancement and Protection team is, verbatim, the wipe-recovery role — *"reads

developing disasters and intervenes before they cascade into team wipes."* The role exists; the

mechanic it would use is what this direction builds.

STAYS SOLO — do not scale these:

• The modal CHOICE surface. 100 players cannot hold on one player's prompt. The MMO form is a

timed release-or-wait, non-blocking, world keeps running. **Design S2 surface-agnostic so

blocking-ness is a property of the surface, not of the machine** — this is a BUILD-NOW

architectural requirement, not a Tier-C afterthought.

• The INT-as-AI axis. There is no AI ally in a raid — the ally's brain is another player. INT

stays a Tier-A/party property.

⚠ ONE REAL CANON COLLISION — name it now. §8.7 rules raid roles path-agnostic: *"Integrity

band gates no role, no build axis, and no operational team. An L5 Pure Evil guild fields the same

role space as an L1 Truly Good guild."* But **every revival row carries integrity_band_lock = L1

or L1,L2. Under current data an L5 guild cannot field a rezzer at all** — a direct

contradiction of a ratified Tier-C rule.

Recommended resolution: the band lock is a Tier-A campaign catalog-access property (which is

what T1_Ability_Tree §12 says catalog access is *for* — path identity and replayability), and

Tier-C fields the L4/L5 alternate route so the role space stays equal. **Which promotes authoring

the evil-lane revival from a no-dead-end nicety to a Tier-C BLOCKER.**

Also flag: §8.8's four-team architecture has no failure economy at all — kill benefits are

canon, wipe/recovery rules are absent. Out of scope now; record it as a Tier-C-phase gap so it is

not discovered late.

---

8. THE JOSH-DECISION SHORTLIST (each briefed above, none bare)

1. J-1 — the withdrawal reconciliation. Bless *"let go / hold on"* (zero amendment,

recommended), or rule withdrawal explicitly graduated (amends BATCH1 item 4). **Gating: every

player-facing word depends on it.**

2. J-2 — the DOWNED/DEAD hard line. Ratify: *DOWNED is a symmetric combat state; DEAD is an

asymmetric permanent story event; permanent death is never a combat outcome for a persona'd

character; the boundary is the community-ripple event.* §17-adjacent cardinal territory.

3. J-3 — the healer's child. Reading A (default: barred by D-2 F-STAR) vs Reading B (an

out-of-combat, non-vril, herbal resuscitation that grants her no vril, no weapon, no combat

presence). B is the highest-emotion version of this feature in the game.

4. J-4 — path parity. Separate L4/L5 route, HL_0076 magnitude banding, or both. (The loop

authors the rows either way; the parity mechanism touches equally-weighted endings.)

5. J-5 — can the party win without you, and can they finish a BOSS? If yes, who gets the win

credit / win_reward / mastery credit / tell-reads. The largest scope decision in the direction.

6. The attrition collision (guard 3.3): graduated persistence (the boss remembers) vs

revive-at-body. One ruled doctrine has to bend.

7. Naming. Player-facing naming flags per docs/NAMING_METHOD.md — but note the natural-voice

doctrine bites hard here: **there should probably be no name at all on the player's surface,

just a person's voice and a choice.**

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