pipelines/FORKB_ANALYSIS_ADVERSARY_2026-07-27.md
Deep dive on Josh's 2026-07-27 direction: *"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."*
Axis: attack it from both calibrated poles — AGAINST (canon breaks, exploit surface, design risk)
and FOR (the devil's-advocate against over-conservatism, per the care doctrine's own method).
Read-only pass; every claim below carries its citation or is marked as inference.
---
Before attacking anything, four discoveries that most of the argument turns on. Two of them
invalidate objections I expected to make, and one of them is the single most load-bearing fact in
this brief.
Josh's own pre-Claude GDD authored a full revival ladder in the Ether/Crown school:
energy to breathe life into a fallen target. Enhanced Vril Revival fully restores the target's
vitality upon revival."*
[SRC: docs/original_predclaude/extracted/GDD_Humanity_Game_Complete.md L15675-15700]
life into the player's entire fallen team simultaneously."* [SRC: ibid. L15725]
— 60+ — prereq Vril Healing 8; Vril Mass Revival — Ether/Crown — Healing/AoE Revival — tier 4 —
85+ — prereq Vril Revival Mastery 10`
[SRC: docs/original_predclaude/extracted/Ability_Tree.md L1469-1472]
no-cheap-death rule — revival costs scale per the spec"* [SRC: COMBAT_PROGRAM_ADDENDUM.md §4]
And revival is an in-world fact with an enemy-side instance. The Scythe's mastery ability:
*"Death's Scythe — opponents defeated with this weapon cannot be revived by antagonist family
means."* [SRC: docs/original_predclaude/extracted/Weapons_Trickery.md L472]
That single line does three things at once: it proves the **House of Velheim has revival
capability, it proves revival-denial is already a designed counter-lever**, and it proves the
fiction has already priced revival as a normal contested capability rather than a sacred exception.
Consequence for the whole debate. The direction does not introduce resurrection to Humanity.
Resurrection is already here, flowing in three directions (player→ally, player→team,
antagonist→antagonist). The only vector with no canon is ally→player. So the honest framing is
not "should we add resurrection" — it is "the one hole in an already-authored symmetric system is
the one that matters most to the player, and Josh just asked for it filled." That reframing kills
my strongest expected AGAINST argument before I make it, and I record that honestly.
RATIFIED canon: *"Defeat model RATIFIED as canon: consciousness withdrawal to the last attuned
recharge source + defeat writes a path-variable journal observation."*
[SRC: docs/BATCH1_RULINGS.md L50-51]. Expanded: *"on defeat, consciousness withdraws to the last
attuned recharge source (recharge-as-checkpoint); … defeat AND victory both advance something — the
Hades loop without an XP-on-death economy"* [SRC: docs/COMBAT_ENCOUNTER_SYSTEM.md L419]. The
dossier reinforces that this is deliberately a *substrate* event, not a UI event
[SRC: docs/DECISION_DOSSIER.md L230, L266].
This is the sharpest canon friction in the entire direction and I develop it in §1.2.
Positive-controlled: grep -rE "InCombat|CombatState|bIsFighting|bEngaged" over
C:/dev/Humanity/Humanity/Source/ returns zero combat-state predicates (the only hits are
Aggro/DeAggro radius fields on HumanityEnemyCharacter and a leash test). Control: Composure
returns 12 hits in HumanityBossCharacter.cpp, so the search apparatus works. There is an aggro
state machine (IsAggroed(), AggroRadius, DeAggroRadius, leash/RETURN) but **nothing that
answers "is the player in combat."** The ruled constraint is therefore a constraint on a concept
the project has never had to define.
AHumanityCharacter::HandlePlayerDeath() → consolation resolve → camera fade (0.55s) → timer →
RespawnAtNearestSite() → nearest AHumanitySiteMarker → SetHealth(MaxHealth) → fade in →
consolation line surfaces for a few seconds. And in that function, verbatim:
*"GRADUATED-PERSISTENCE DOCTRINE (W2 note): the player returns, but BOSS state PERSISTS across the
respawn — the boss is not reset (its Composure damage / defeat state / withdrawal all stand)."*
[SRC: Source/Humanity/Private/Core/HumanityCharacter.cpp L349-352]
No companion runtime exists at all. Word-bounded grep for \bcompanion|\bAlly\b over Source/
returns only worldstate tag definitions (WS_companion_roster_state,
WS_companion_loyalty_transition_events) and two data-row columns. There is no ally pawn, no ally
AI, no party. Control: Boss matches across five headers. The direction's runtime is 100% unbuilt.
---
kicked or bolted
The register ruled it: *"named/persona'd NPCs stay permanently dead (the community-ripple
doctrine)"* [SRC: docs/DESIGN_GAP_REGISTER.md entry 13]. The BG3 lens confirmed the ruling and
found the schema can't even represent it — loyalty_state is a 7-value enum with no death value,
and WS_028's event enum is DRAWN|WAVERING|DEPARTED|REBONDED with no KILLED
[SRC: docs/pipeline_review/LENS_BG3_2026-07-27.md BG3-2].
The attack. The instant the player watches Ana kneel over their body and breathe them back, the
player will try it on the villager who died in the raid, on the companion who fell at Ch 55, on the
healer's child. And the player will be *right to try*, because canon has already told them it works:
Vril Mass Revival targets *"the player's entire fallen team"* (§0.1). A game that shows you
resurrection and then answers "no, not that one" with a greyed-out button has told the player the
rule is arbitrary — which is exactly the AAA-slop tell the re-architecture brief exists to prevent.
But the door was already open before this direction. Mass Revival on a fallen team already
implies the question. The direction does not create the pressure; it makes the pressure *visible*,
which is a different and much more manageable problem — because a visible pressure gets an in-fiction
answer, and an invisible one gets a patch note.
The answer canon already supplies, in its own words: FALLEN ≠ DEAD. The GDD says *"a fallen
target,"* not a dead one. The ruled defeat model says consciousness *withdraws* — it does not say
the person ends. The line is therefore drawable and it is drawable on an event the schema already
fires: the community-ripple / persona-death event. Once a persona's death event fires and
ripples through their community, that person is gone; before it fires, they are *fallen* and the
window is open. This is enforceable (the ripple event is a real write), diegetic (the pattern
disperses), and costs no new metaphysics.
And canon already stocks the narrative lock for the exceptions: Death's Scythe's *"cannot be
revived by antagonist family means"* proves revival-denial is a legitimate authored property. Any
story death that must stay dead can carry the property in-fiction, not in a UI apology.
VERDICT: real risk, real answer, and the answer is already latent in canon. But it is a
FORMALIZATION Josh's direction now *requires* — shipping the revive without drawing this line is
the actual break.
This is the hardest canon collision and it is not resolvable by careful wording alone.
The ruled defeat model is not "you fall over." It is **consciousness withdraws to the last attuned
recharge source** [SRC: BATCH1_RULINGS L50]. The withdrawal is not a checkpoint metaphor — the
DECISION_DOSSIER explicitly defends it as a Layer-3 substrate event, the mechanic that *keeps the
metaphysics real* [SRC: DECISION_DOSSIER L230].
If your consciousness has already withdrawn to the well, then:
incoherent as written: "revive at last checkpoint" is what *already happened* by the fiction's own
account, and "wait for your healer" asks a consciousness that is at the well to wait beside a body
it has left.
Three repairs exist, and each costs something:
1. Withdrawal becomes GRADUATED. Consciousness does not snap to the well; it *begins* to
withdraw and completes over a window. The wait is the window. Taking the checkpoint is choosing
to let go. This is the most elegant repair — it preserves both the ruled model and the direction,
and it makes the wait diegetically *meaningful* rather than a menu. Cost: it amends ratified
canon (BATCH1 item 4). That is a Josh call, not a loop call.
2. The revive acts on the BODY and the consciousness returns along the tether. The vril tether
between the withdrawn consciousness and the still-coherent body is what the healer works on. This
fits the vril substrate well. Cost: it invents a tether concept; check it against
T1_Vril_and_Magic_System_Master before anyone writes it down.
3. Reframe the checkpoint option as "let go." Keep the ruled model intact and change what the
two options MEAN: you are already withdrawing; the choice is whether to complete the withdrawal
or hold on. Cost: none to canon — this is the reading that requires the least amendment, and
it happens to be the most emotionally loaded framing of the two options.
VERDICT: a genuine contradiction with the RATIFIED defeat model, not a nitpick. It must be
resolved before a single string is written, because every player-facing word on this surface depends
on which repair is chosen. Repair 3 (or 1) is my recommendation; both are Josh-facing because
BATCH1 item 4 is ratified canon.
This is the finding I would fight hardest for.
The direction creates a liminal, player-inhabited, repeatable, low-stimulus consciousness state
— the player is dead, waiting, watching, with nothing to do but perceive. That is, structurally, the
single most inviting surface in the entire game for metaphysical content. And the hard lines are
absolute here:
not in marketing copy, not in tutorial text, not in UI."*
in-game content before Ch 76-77 across dialogue, journal entries, inscription content, NPC
exposition, environmental storytelling, art direction, music."*
vibrational imprints."* [SRC: registries/T0_Hard_Lines]
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 the Grand Sage's canonical
communication vocabulary, deployed on a surface the player will visit dozens of times before Ch 76.
The reveal is *the player's own collective consciousness reflected back*; a death state where the
player floats free of the body and is drawn back by others is a thematic rhyme with the reveal so
close it functions as a hint.
Worse: the surface is repeated. HL_0044's protection assumes hints are rare enough to be
individually policed. A death screen played 200 times is a drip.
VERDICT: a MAJOR-class hazard that the existing gates would not catch —
check_reveal_discipline.py reads canon artifacts, not runtime presentation. If this direction
ships, the death/wait surface needs its own named entry in the reveal-discipline baseline and its
strings need the same critic treatment the natural-voice wave just gave the sidecars.
Josh ruled Fork A on 2026-07-27, same day: continuity by default, cut earned — *"The lens is one
continuous presence through gameplay, scene entry, scene body, and scene exit; a HARD CUT is a
rationed instrument reserved to the beat classes §3.2 lists"*
[SRC: docs/proposals/PRESENTATION_DOCTRINE.md §0.A, RULED].
A modal choice prompt at HP-0 is the most anti-continuity object a game can place: it stops the
world, breaks the lens, and hands the player a menu. And it does so in the loop the player sees more
than any other. The doctrine's own stated failure mode — *"a list is a rule with a loophole, and
lists grow"* — begins the moment "the death prompt" is added as a cut class.
Mitigation exists and it is good: make the choice non-modal and diegetic. The world does not
stop; you are down, the fight continues around you, and the input is a *hold* (hold to let go and
return at the well) versus *doing nothing* (hold on, and see if someone reaches you). No menu, no
cut, no taxonomy on the surface — which also satisfies the no-mode-taxonomy rule and the
natural-voice doctrine at the same time. That is a genuinely better presentation than the incumbent
0.55s fade, and it is the version of this direction I would actually ship.
VERDICT: not a break IF the choice is built as a diegetic hold rather than a menu. A modal
death menu breaks a ruling Josh made the same day. Flag it now, before someone builds the obvious
thing.
Standing memory rule: *early-chapter perception is not tier-6 Vril Sight; early chapters perceive
via inscription/reverence/lore tells.* A death state where the player's awareness persists outside
their body and observes the fight is, on its face, astral perception — canonically tier-6
Adeptus Minor, Ch 15+, the one place substrate perception begins
[SRC: docs/ESOTERIC_GROUNDING.md §217; the D-06 age+mastery model].
The Josh Gate targets the AAAAA-polished Ch 2-5 slice — the exact window where the protagonist
has *no* such capability, and where a companion trained in Vril Revival cannot exist either (the
ladder is Vril Healing 8 → Vril Revival at tier 3 / 60+; nobody at Ch 2-5 is near it).
Consequences, stacked:
play, per the Josh Gate.
post-capability), which is a real content cost, not a toggle.
nothing, learns the option is fake, and never checks it again — a taught-then-broken affordance,
which is the worst outcome of all.
VERDICT: real, and it is a SEQUENCING finding, not a kill. The honest shape is that the choice
surface *unlocks diegetically* the first time a companion learns the capability — which is a lovely
beat in its own right (the first time someone can come for you is a story moment), and it is exactly
the "diegetic-cause rule proven at R4.3" the foundry already follows.
directions cost something ruled
Three ruled facts collide here:
1. Incoming healing is already integrity-banded. HL_0076: *"L1 +10% incoming healing; L2 +5%
incoming healing; L3 +5% all damage received; L4 +5% damage dealt; L5 +10% damage dealt + +10%
damage vulnerability."* Good paths receive more healing; evil paths receive none and take more.
A revive is incoming healing at its apex — it inherits this modifier by construction.
2. Familiar combat behavior is already hard integrity-gated. T3_Familiars_Named §1.4, live as
seed rows AP_0001/AP_0002 in T0_Ally_Behavior_Policy: on L1/L2 *"the familiar steps in
voluntarily at 50% or less player health … the player cannot override this behavior on these
bands — the familiar acts on its own judgment."* On L3 the ally can be *directed* to attack only
past an earned trust threshold. Gate 29 tooth B4 makes an aggressive disposition on an L1/L2
row structurally illegal.
3. But §8.11.8 firewalls the opposite way: *"Behavior quality is a build property, never an
integrity property … no amount of investment moves the belief-distance scalar."* And Gate 29
tooth B5(i) enforces it: no policy condition_expr may read a loyalty ws key.
So: does an L4/L5 protagonist's companion come for them? The two canon lanes point opposite
directions, and the direction forces the question.
should bite hardest, and the game's most emotionally loaded mechanic becomes integrity-neutral
wallpaper.
loop in the game, which pressures the ruled "three equally-weighted endings" architecture and the
difficulty doctrine's judgement-free posture.
The resolution I would argue for: it is neither a loyalty read nor a behavior-quality read — it
is a healing-magnitude read, and HL_0076 already rules it. The companion *comes* regardless
(loyalty decides whether they are still with you at all — that is the loyalty axis doing its job
upstream), and how well the revive lands rides the band modifier that already exists. That routes
around the §8.11.8 firewall cleanly (no condition_expr reads loyalty; the magnitude reads the band
exactly as all healing does), and it lets the Evil path feel *different* without being *penalized* —
the companions you have left are the companions who come.
**VERDICT: a real fork the direction forces, with an elegant existing answer. Loop-rulable under
the care doctrine, but worth putting in front of Josh because it touches path-parity.**
Josh's phrasing is probabilistic twice over: *"wait to see if … may be able to revive."*
That is a chance event, and chance events in Humanity are not free:
HL_0061 — *"All RNG-based drops and events are constrained to plus-or-minus 1 sigma of the bell
curve. Anti-streak floor at minus-1 sigma. Mandatory drop ceiling at plus-1 sigma. Resets on drop.
The constraint is foundational across drops, spawn rates, and RNG events."* And the σ doc scopes it
to *"every RNG-bounded event in canonical scope,"* explicitly excluding only deterministic gating
[SRC: T1_RNG_Sigma_Constraint_System §3-4].
Two consequences, both real:
enough failures, the next one *must* succeed. That is arguably wonderful (it is the anti-frustration
guarantee already in canon) but it is a mechanical fact the design has to know.
"see if" is uncertainty about conditions, not about a roll — you don't know if your healer
survived, is in range, has vril. That is the better design and it is also the reading that
makes the wait *dramatic* rather than slot-machine.
VERDICT: the "see if" must be resolved as CONDITIONAL uncertainty, not a dice roll — otherwise
the game's most emotional beat becomes a σ-bounded loot roll, which is the exact register mismatch
the rune-imprint lane exists to contain.
The ruled death economy pays on death: *"a Withdrawal-echo payout — the realm remembers your effort.
You bank the tell-reads earned this attempt … a substrate-lore fragment (L2/L3 Grimoire, so death
teaches)"* [SRC: COMBAT_ENCOUNTER_SYSTEM.md §7 L220]. It is BUILT: PendingDeathConsolation is
resolved at death and surfaced on respawn for ConsolationDisplaySeconds
[SRC: HumanityCharacter.cpp L270-286, L362].
If the companion revives you, there is no respawn — and the consolation surface, the tell-read
banking moment, and the journal observation write all lose their trigger. Either they fire on a
revive too (in which case "death teaches" fires on a death you did not actually pay for, cheapening
it), or they don't (in which case the *better* outcome pays you *less*, which teaches players to
prefer dying properly — an inverted incentive).
VERDICT: a real economy hole with no obvious free answer. My read: the journal observation
should fire on BOTH (you learned the tell either way — the knowledge economy is honest), and the
consolation *line* should be replaced on the revive path by the companion's own words, which is a
strictly better beat. But it is a design decision the direction forces and nobody has made.
---
Boss state persists across player death by design: *"the boss is not reset (its Composure damage
/ defeat state / withdrawal all stand)"* [SRC: HumanityCharacter.cpp L349]. Today the only thing
limiting attrition-farming a boss is the walk back from the well — a cost that exists purely by
accident of the respawn location.
A companion revive at the body, in the arena deletes that cost. Every boss in the game becomes
winnable by pure attrition: chip, die, get up, chip, die, get up. That is not a corner case; it is
the dominant strategy for any player who finds a fight hard, which is every player at least once.
Note this is a collision between two ruled doctrines, not a mere bug: graduated persistence says
the boss remembers; the direction says you get up where you fell. One of them has to bend.
Candidate resolutions:
advance*, not the walk. (Weak — the fantasy Josh described is your healer reaching you.)
measured in *encounters*, a per-rest charge). The `no-cheap-death rule — revival costs scale per
the spec` in addendum §4 is already the hook.
finish. This is the cleanest, and it converts the attrition exploit into the thing it should be:
the party has to actually win the fight for the revive to matter.
If waiting is free and the checkpoint is always available as a fallback, **there is no reason ever
to pick the checkpoint first.** Wait; if nobody comes, take the checkpoint. The "choice" collapses
into a strictly-dominant sequence, which means it is not a choice — it is a delay with a menu on it.
A choice is only real if waiting costs something the checkpoint doesn't. Candidate costs, all
canon-native:
the enrage-replacement, and while you are down *the realm recedes*: tells tighten, wells slow,
answers prune, and your party gets worse the longer you lie there
[SRC: COMBAT_ENCOUNTER_SYSTEM.md §4.7]. The anti-stall device is already built into canon and
costs nothing to adopt.
applied as a *dial*, not a switch).
you were waiting for.
The rule "only out of combat" is doing enormous protective work, and it rests on a predicate that
does not exist (§0.3). Whatever definition gets built becomes the exploit surface:
no enemy IsAggroed() / leash RETURN): players will kite enemies past the leashradius, die deliberately outside it, and revive at the door of the arena. The leash is a fixed
radius on HumanityEnemyCharacter; it is trivially gameable.
of sight, wait out the timer, and revive mid-fight in all but name.
HumanityEncounterVolume says the encounter is resolved): the mostrobust, and the only one that actually means what Josh means — the fight is *over*. It also
implies the party had to survive/finish it.
"in combat" is unambiguous. This is the highest-value, lowest-ambiguity subset and should probably
be where the rule is *strictest*.
The finding: the out-of-combat rule is only as honest as its predicate, and the predicate choice
IS a design decision with exploit consequences. It must not be delegated to whoever builds it.
If the reviver can also be downed — and per §1.6/§0.1 the whole team can be *fallen* — then a party
wipe produces a state where the player is waiting for someone who cannot come, possibly forever, in
a world where enemies are still present. The state needs a bounded resolution or it is a soft-lock,
and a soft-lock is a direct hit on the no-dead-end floor (check_no_dead_end.py is a live gate,
and §8.11.8's discipline is explicit: *"Behavior quality never gates progression … never makes it
unwinnable"*).
Related: can companions die permanently? BG3-2 found the schema has no dead state and
recommended adding one. If a revive-capable companion can be permanently killed, the player can
permanently delete their own revive access — a silent, invisible difficulty cliff with no
player-facing explanation. That is a no-dead-end-adjacent defect even if progression technically
survives.
The vril site is ruled as save + recharge + loadout sanctum [SRC: addendum §9; T1_Combat §14].
It is also the checkpoint. Any revive economy denominated in "per rest at a well" is therefore
denominated in a resource the player can farm by walking back to the well — which means the revive
cost has to be measured in something the sanctum does *not* refill, or the sanctum refills the
exploit.
---
Addendum §10, verbatim: *"still allow builds to be mathematically overpowered — but not in the way
that all the AAA games fail, where that means you only have 1 or 2 viable options … META-COLLAPSE is
the banned failure mode."* And the teeth are real: check_build_space.py (gate 26) exists to
measure viability breadth, and *"a pool whose options are never picked in the sweep is a balance
DEFECT."*
A revive that meaningfully reduces the cost of failure is the strongest possible effect in an action
RPG, because it is multiplicative with every other thing you own. If one companion has it and
others don't, the build-space sweep will find exactly one correct party. The **§12 INT tradeoff
inverts too**: Josh ruled INT must *compete* with combat stats — but if INT drives whether your
reviver reaches you in time, INT stops being a tradeoff and becomes the tax everyone pays.
This is the single strongest AGAINST argument on design grounds, and the counter is structural: the
capability must be broadly reachable (a taught proficiency any support-spec ally can learn, per
Josh's own "trained or taught" phrasing) rather than a companion's identity. Josh's wording
already leans the right way — "who may have been trained or taught" is a capability, not a
character. Build it as a capability and the meta-collapse risk drops sharply; build it as
"Ana is the healer" and it collapses on day one.
Death is the highest-frequency interruption in an action RPG. Adding a decision to it means the
player makes this decision hundreds of times. Decisions that are made hundreds of times and are
*usually the same answer* are not decisions — they are taxes. Every second of the death loop is paid
by the most frustrated version of the player.
The presentation doctrine names this hazard precisely: *"the classic AAA mistake — punishing the
player twice, once with the death and once with the un-skippable ceremony"*
[SRC: PRESENTATION_DOCTRINE.md §0.B adversarial pass].
The direction as stated has the player *waiting to see*. Waiting is the one thing an action game
must never ask for. But — and this is where the AGAINST case turns into the FOR case — **the wait is
only dead time if there is nothing to watch.** If the wait is "your team is still fighting for you
and you are watching them do it," it is not dead time; it is the most tense twenty seconds in the
game. Which leads directly to:
This is the biggest engineering finding on my axis and it follows inescapably from the
out-of-combat rule.
Almost all deaths happen in combat. If the revive is out-of-combat-only, then either:
ending without the player** — i.e., the party must be able to win, drive off, or survive an
encounter solo.
Reading (b) is certainly what Josh means, and it is a major requirement:
win_reward, themastery credit, the tell-reads? If no, then bosses — the fights where death matters most — are
exactly where the mechanic never fires, and it becomes a trash-mob feature.
realm — which is either brilliant tension or an unwinnable trap depending on tuning.
This requirement is invisible in the direction's phrasing and it is the largest single cost.
Stated plainly: the ruled axis is rated-M, HP 0 = death. A revive that fires often makes death
into a stagger. But I have to hold this against the calibration rule: the game *already* has
zero-cost death (fade, respawn, full health, boss keeps its damage). The current loop is a
softer death than the one Josh is proposing, because the current one is automatic and
consequence-free at the personal level. A revive that costs a companion's resource, runs the
Withdrawal clock, and only works if your team survived is harder, not softer, than what ships
today. The "death loses weight" objection is therefore only valid against the *free* version of this
mechanic — and the free version is not what should be built anyway.
---
Per the care doctrine's method: the loop's known bias is timid, so the pro-case gets argued at full
strength.
§0.1 is decisive. Josh authored Vril Healing → Vril Revival → Vril Mass Revival years ago, with
tier gates, prerequisites, and an enemy-side counter-lever (Death's Scythe). Ally→player is the one
missing arrow in a system that already points three other ways. **Refusing it is not conservatism
about canon; it is a hole in canon.** Any objection framed as "this adds resurrection to the game"
is factually wrong, and I made sure to check before writing the AGAINST case.
Josh ruled INT-as-AI-quality on 2026-07-26: pattern detection, proactivity, awareness — *"the ally's
BRAIN is a stat the player invests in."* The honest problem with that axis is legibility: better
AI is felt as ambient competence, and ambient competence is the hardest thing in games for a player
to attribute to a stat they bought. The comparator program flagged exactly this — INT-as-ally-AI is
one of the *"day-old rulings with zero teacher"* on the lens roster
[SRC: commit a238a87, COMPARATOR_LENS_PROGRAM].
"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. It converts an invisible axis into a
story the player tells about a build decision they made. There is no cheaper way to make §12 land.
Read the companion canon as it stands: companions leave over your integrity (§8.11.2), leaving
costs you routes (rerouted_paths), the gear comes back so the loss is purely relational (§8.11.6),
and on the Evil path your anchor companion becomes an opposition encounter (§8.11.3). Every
mechanical expression of the loyalty system is a loss. The system's entire mechanical vocabulary
is punishment and consolation-for-punishment.
This direction gives it its first gain — a moment where the relationship *saves you*. That is
not a nice-to-have; it is the missing half of a system that currently teaches players only that
relationships can be taken away. And it lands precisely where the canon says the weight belongs:
*"the moral weight of a departure sits on the relationship."* The inverse should also sit there.
Josh ruled spec purposes on 2026-07-26 — *"vril-HEALING focus vs vril-DAMAGE focus; regen/s, HoT,
personal vs AoE."* Right now vril-healing spec is a set of ordinal shapes with no dramatic ceiling:
it makes numbers go up slightly less dramatically than damage does. Every healing spec in every RPG
lives or dies on whether it has an apex moment. Resurrection is that moment, and it is already
where the authored ladder points (Vril Healing 8 → Revival → Mass Revival at 85+ is *literally a
climb toward resurrection*). Without this, the healing spec's top tier is "more healing."
Every design instinct wants to protect combat stakes from revive mechanics, and the usual solutions
are ugly (long cast times, limited charges, revive sickness). Josh's constraint is **cleaner than
any of them**: combat stakes are *categorically* untouched because the mechanic cannot fire during
combat. Fights are exactly as lethal as they are today. All the mechanic changes is what happens
*after* — and "what happens after a fight you lost" is precisely the design space the Hades loop
already occupies. He arrived at the correct line himself, and the loop should say so rather than
looking for a way to soften it.
The ruled loop is: *win or die, the run pays out; defeat AND victory both advance something.* The
built payout is a text line on a HUD after a camera fade — mechanically correct, dramatically
inert. The direction replaces a fade-and-a-line with a *scene*: you are down, the fight continues,
someone reaches you or doesn't, and either way the world moved. That is the same economy, expressed
as drama instead of as a notification. It is a strict upgrade on the Hades comparison the doc
already invokes.
The presentation doctrine says it plainly: *"Option 0 additionally decides the question by inertia,
which is how a greybox constant becomes shipped canon without anyone ruling it."* The timid answer
here — "death should stay simple, keep the fade" — is not the safe answer. It is the answer that
ships a 0.55-second placeholder as the death presentation of a 30M-word game, and it forfeits the
one place where the party system, the INT axis, the healing spec, and the loyalty canon could all
land at once.
The comparator roster names 24 uncovered aspect×lens cells and explicitly flags parry/chain-pip,
INT-as-ally-AI, and AAAAA build breadth as rulings with no teacher. Companion-revives-*you*, gated
out-of-combat, driven by a taught capability, inflected by integrity, in a game where death advances
the world — has no teacher because nobody has built it. Dark Souls has no companions. BG3
revivify is a scroll in a turn-based pause. Hades has no party. Mass Effect downs you and ends the
mission. This is a signature mechanic, and signature mechanics are the entire thesis of the AAA
re-architecture brief.
---
Both attacks converge on the same object. The version of this direction that survives:
1. The choice is a diegetic HOLD, never a menu — no cut, no modal, no mode taxonomy on the
surface, natural voice only (§1.4).
2. The two options are "let go" vs "hold on," which repairs the consciousness-withdrawal
contradiction with zero canon amendment (§1.2 repair 3).
3. The wait is not empty — it is watching your team fight, which turns the direction's biggest
risk (dead time) into its biggest asset, at the cost of ally-AI-that-can-close-an-encounter
(§3.3/§3.4).
4. The Withdrawal clock runs while you're down — canon's existing anti-stall device, adopted
free, kills wait-scumming and adds real tension (§2.2).
5. The revive is a taught CAPABILITY, not a character — kills the meta-collapse risk and matches
Josh's own "trained or taught" phrasing (§3.1).
6. "See if" is conditional uncertainty, not a σ-bounded roll (§1.7).
7. FALLEN ≠ DEAD, drawn on the community-ripple event — bolts the named-dead door with a line
canon already implies and a gate can already check (§1.1).
8. The revive returns you where the design says, not where physics says — resolve the attrition
collision with graduated persistence explicitly (§2.1).
9. The surface unlocks diegetically when the capability first exists — no fake affordance in
Ch 2-5, and the first unlock is a story beat (§1.5).
10. The death surface gets its own reveal-discipline entry and critic treatment (§1.3).
---
1. The consciousness-withdrawal reconciliation. The RATIFIED defeat model says consciousness
withdraws to the last attuned recharge source. The direction requires it to linger. Three
repairs, each with a different cost (§1.2): reframe the options as *let go / hold on* (no
amendment), make withdrawal *graduated* (amends BATCH1 item 4), or introduce a *tether* (new
vril concept). This is the gating decision — every player-facing word depends on it.
2. FALLEN vs DEAD as a canon line. Formalizing "revival acts before the community-ripple event
fires, never after" is a rule about death, personas, and the named-dead doctrine. The loop can
author the mechanism; the line itself is Josh's.
3. The Evil-path answer. Does a companion come for an L4/L5 protagonist? My recommendation is
*yes, with HL_0076's band modifier deciding how well it lands* — but this touches path parity and
the equally-weighted-endings architecture, so it goes to Josh.
4. Whether the party can WIN a fight without the player — and if so, whether they can finish a
BOSS, and who gets the credit. This is a scope-and-stakes decision worth more engineering than
everything else in the direction combined (§3.4).
5. The attrition collision: graduated persistence (boss remembers) vs revive-at-body. One of two
ruled doctrines has to bend; which one is Josh's call (§2.1).
§2.3.
doctrine — §1.4.
the T0_Ally_Behavior_Policy rows, the T0_Bonus_Pool_Registry entries, the role-axis tags —
§3.1.
spec` hook — §2.1/§2.5.
check_no_dead_end.py predicate — §2.4.companion's own words on the revive path — §1.8.
The direction is not an addition to canon — it is the missing symmetric case of a resurrection
ladder Josh authored himself, and refusing it is the timid error, not the safe one. It survives every
attack I could mount except one: it collides head-on with the ratified consciousness-withdrawal
defeat model, and it silently requires companions who can finish a fight without you. Rule the first,
cost the second, and the rest is execution.