COMPANION_MECHANICS_CONTRACT.md

npcs/COMPANION_MECHANICS_CONTRACT.md

FOUNDATION B — The Companion Mechanics Contract

The exact interface a companion must satisfy to be legal in the ratified systems, and — separately and

explicitly — where that interface leaves real design room. This is a read of the ratified ground, not a

proposal: every mechanism below is already ruled or already schema-landed. Casting and per-companion loyalty

assignment remain Josh's (characterization canon, WS7 L158); this document tells the roster designer what a

companion row/persona MUST contain and what MUST be true of it, so the roster brief proposes only inside the

lines the systems already draw.

Citations are [SRC: path # anchor]. Four files are being mutated by a concurrent lane (T1_Integrity tier

doc, T0_Character_Index, RATE_MISMATCH_ANALYSIS, and CH_17/28/39/56/57); I read them for content but cite

them by section/name, never line number. Everything else is line-cited and verified to resolve.

---

0. The one-paragraph contract

A companion is a person-class entity (T0_Character_Index row + a flagship/supporting persona) that

(a) fills all 13 persona spine fields plus the person extension; (b) carries the **6 companion-loyalty

columns on Character_Index with a legal enum tuple; (c) sits on THE BOND** relational ladder (one shared

−100..+100 numeric axis, companion-axis display labels Enemy…Family) whose movement is a **runtime state

function, not an accumulator; (d) obeys the opposition-magnitude departure rule** (departs on sustained

belief-opposition, investment is inertia but never immunity, complete opposition always forces departure);

(e) is attackability Tier-3 (bonded) — betrayable at maximal cost, never god-moded, never a progression

wall; (f) lives aboard the mothership and may crew sub-craft; (g) may teach or peer a trade, contribute

a combat verb/mode as an ally, and open/close opportunities — but never gates progression (the

no-dead-end invariant; every departure ships a non-empty rerouted_paths). A companion is on the **human/

companion axis; the 22 familiar slots are a separate axis** (WS_012, creature-class, binding-stone) and

must never be conflated.

---

1. INTERFACE 1 — the 13-field persona spine (every companion fills all 13)

Source of truth: [SRC: docs/proposals/ENTITY_PERSONA_SYSTEM.md # §2.2 L30-44]. The persona extends the

anchor row; it owns no fact the anchor row does not carry (do-not-invent tether). For a companion, each field

has a companion-specific obligation:

• anchor_ref — FK to the companion's CHAR_0NNN row in T0_Character_Index. This is the identity and the

do-not-invent tether. A companion with no clean anchor row is illegal.

• identity — self-naming per §17.1 (the culture's own term, no stereotyped descriptor), class = person,

one-line essence. Names are naming-canon → flag for Josh per docs/NAMING_METHOD.md.

• significance_lenses — which of the 33 lenses the companion embodies, primary marked

[SRC: docs/proposals/SIGNIFICANCE_LENSES.md]. This is the relevance ticket AND the anti-museum-flavor

hook: a companion is in the cast because they carry an arc-serving lens, not for decoration.

• web_position — the companion's edges (kin, teachers, faction ties, rivalries), each resolving to another

persona/registry row, mirroring Cross_Cultural_Link_Graph. A companion is a hub in the relational web

[SRC: memory regions-are-relational-webs-not-villages].

• background_history — documented real-world grounding (Frame A, no hyperdiffusion) + in-world backstory.

This is the §17.1 dependency surface: where a culture's substrate in the corpus is thin, the companion's

history CANNOT be faked — flag book-processing enrichment as the blocker (care = weight + authenticity).

• own_life — daily rhythm, independent existence, how they age across the two-clock age model

[SRC: memory d06-age-mastery-reclaiming-birthright].

• motivations — goals/drives/fears. For a companion this is load-bearing: it is the raw material of the

belief axis (§2) that drives departure.

• personality — 4-8 terse traits + temperament, in the exact form the runtime persona card consumes

[SRC: docs/RUNTIME_GENERATIVE_LAYER.md # §2.1].

• voice_register — speech register + the six-option dialogue coloring; references T0_Voice_Registry

(only 10 rows today — a thin-anchor dependency).

• relationships_to_player — the hinge into the relationship engine. Declares that this entity holds a

relationship ledger (companions always do), its starting disposition, and which opportunities its bond

gates. This field is where THE BOND attaches.

• care_band — §17.1 tier (standard / elevated / high-care) from T0_Hard_Lines. High-care companions

carry the stricter authenticity bar and the authored-only-at-launch runtime default.

• layer_surfacing — L1 playable presence / L2 investigatable depth / L3 substrate meaning (never

declared; no pre-Ch-76 Grand-Sage or House-pattern leak).

• runtime_binding — binds to runtime Class 4 (partner/companion/familiar-handler)

[SRC: docs/RUNTIME_GENERATIVE_LAYER.md # §A]; strategic-tick cadence; authored vs generated; authored

fallback line bank.

1.1 The person extension — the six companion-specific fields

[SRC: docs/proposals/ENTITY_PERSONA_SYSTEM.md # §2.4.1 L64-69]

• age_stage_presence — which of the protagonist's three age stages the companion appears in, and how they

age across revisits.

• faction_affiliations and awareness_tier — faction ties + (for any House-adjacent figure) the

House-of-Velheim awareness tier, governed by the reveal discipline: no companion may perceive/voice the

family-behind-the-institution pre-~Ch-38, no who-frame, no agent-noun in the safe register

[SRC: memory antagonist-pattern-reveal-timing].

• teaching_role — the trade / ability / worldview the companion teaches and the mastery-tier it gates

(Interface 6).

• six_option_register — how this companion colors the canonical six dialogue options (Truly Good, Truly

Bad, + four biography-shaped: Sympathetic, Detached, Curious, Defiant) in their cultural register

[SRC: _source/01_Tier_1_Foundation/T1_Integrity_Paths_Worldstates_Master [ACTIVE v2.1].md # §6].

• assimilation_stance — the companion's role in the concealment arc: do they extend trust and take the

protagonist in, read her vital force and hold her secret, or turn hostile-recognizer? This is what the

assimilation mechanic costs and buys through a companion's eyes: a companion who WITNESSES the player's

in-culture integrity choices is the site of a bond swing.

• companion_or_familiar_mechanics — bond-progression + shared-combat behavior, deferred to the registry

per-slot mechanics (Interfaces 2 + 5).

The living-faithful protection binds companions hardest: a faithful community member is never a combat

target and never the antagonist; no sacred proper name sits on a mortal companion (the Kxao rule); no revered

real person is recast as a villain companion (the Liqä rule) `[SRC: docs/proposals/ENTITY_PERSONA_SYSTEM.md #

§2.4.1 L71; memory deities-real-portrayable-17-1-nuance]`.

---

2. INTERFACE 2 — the Character_Index loyalty columns (WS7 anchor, LANDED)

The 6 columns are live on T0_Character_Index (append zone after the frozen 12-col reference), owner

companion-loyalty, gate-green [SRC: docs/proposals/systems/WS7_WIRING_BUILD_PLAN.md # L136-146]. A

companion row MUST carry a legal tuple:

• loyalty_model ∈ for_life | invested_sticky | come_and_go | drawn_conditional | not_companion. This is

the companion's structural type. not_companion is every non-companion entity (the default for the 157 of

162 rows that are not cast).

• loyalty_state ∈ unmet | recruitable | traveling | wavering | departed | estranged | bonded_locked —

the runtime membership state.

• loyalty_belief_axis — a controlled-vocab token naming the companion's value pole (e.g.

truth_witness_mercy, freedom_from_bondage_redemption, caci_honor_community, craft_and_lineage),

compared against the player integrity band WS_001/WS_002. The vocabulary itself is naming-canon —

propose a set aligned to the T1_Integrity bands for Josh to ratify; seeds are provisional

[SRC: WS7_WIRING_BUILD_PLAN.md # L159].

• loyalty_departure_threshold ∈ `never | complete_opposition | major_opposition |

any_sustained_misalignment | circumstance_only`.

• loyalty_join_trigger — chapter+condition / integrity-band / soft quest ref (e.g.

invest_Ch03_active_join_Ch13, drawn_when_integrity_light_L1_L2).

• loyalty_notes — prose.

A second column family co-locates on Character_Index and must be kept owner-disjoint (the one-owner-per-

column gate FAILs on a double owner):

• attackability_tier + is_load_bearing — owner attackability, NOT companion-loyalty (Interface 7).
• companion_loyalty_tier ∈ n-a | come-and-go | drawn | invested | for-life — a display alias family; the

authoritative model is loyalty_model above [SRC: WS7_WIRING_BUILD_PLAN.md # L172-176].

Hard constraint on population: per-companion assignments are characterization canon = Josh's. Only 5

of 162 rows are seeded, and those seeds are ILLUSTRATIVE — the roster brief must NOT mass-populate the ~30

companion rows before Josh ratifies the roster-wide mapping [SRC: WS7_WIRING_BUILD_PLAN.md # L158].

Known live-data caveats the roster brief inherits (do not silently "fix"):

• The spine travels with four characters (CHAR_0140/0144/0145/0146) that carry no loyalty_model at all,

while Tunde (CHAR_0127), who never travels, carries a full drawn_conditional row. This roster/seed

mismatch is the real open gap — surfaced by the rate-mismatch re-test (RATE_MISMATCH_ANALYSIS, §6). It is

Josh's call and is not yet a ruling slot (a bare "who are the ~30?" is forbidden by the decision

protocol) — Claude authors the roster brief from the spine's own evidence first, then Josh rules.

• Live seed enums drift from the WS7 plan (Ana major_opposition vs planned circumstance_only; Tunde

any_sustained_misalignment vs planned major_opposition; CHAR_0003 circumstance vs circumstance_only).

All are illustrative rows Josh rules anyway — note the drift, do not treat any as canon.

---

3. INTERFACE 3 — THE BOND ladder (RULED: Scheme C, per-axis labels)

Ruled by Josh 2026-07-16 [SRC: docs/spine/RELATIONSHIP_TIER_BANDS_BRIEF.md # Rulings 2 & 4 L14-19, L52-64].

One shared numeric ladder across every relational axis (bond_state NPC, standing faction, loyalty_score

companion), −100..+100, with per-axis display labels. The companion axis is THE BOND over

loyalty_score:

Rung invariant: only Cooled and Stranger are shared labels, at identical rungs. Family (not Kin) —

"kin" is reserved for the Ch-66 origin-lineage recognition [SRC: RELATIONSHIP_TIER_BANDS_BRIEF.md # L82].

Load-bearing correction (do not build against the accumulator): THE BOND bands a STATE, not a running

total. The ratified runtime moves the bond via a ±-clamped relationship_delta at ~0.03 Hz, and disposition

is "generated fresh each encounter from the current state" `[SRC: docs/RUNTIME_GENERATIVE_LAYER.md # L113-116,

L143; docs/proposals/ENTITY_PERSONA_SYSTEM.md # L137]`. A state bands identically at any event count — so there

is no per-beat accumulator for the roster to design, and the earlier "rate" cardinal was retired

(RATE_MISMATCH_ANALYSIS self-correction). The bond delta and its thresholds are owned by the deterministic

runtime state machine; the roster designer supplies belief axis + join trigger + departure threshold, not

a points budget.

The §17.1 grammar solve (why the labels are safe): every STANDING label is a participle of reception — it

names the player's earned position, never what a people *are* (Sámi: Hunted is safe where Sámi: Hostile is

disqualified). THE BOND labels are a companion's disposition toward the player, same principle

[SRC: RELATIONSHIP_TIER_BANDS_BRIEF.md # L65-70].

---

4. INTERFACE 4 — the departure rule (RULED: design A, opposition-magnitude)

Ruled [SRC: RELATIONSHIP_TIER_BANDS_BRIEF.md # Ruling 3 L36-51]. Departure is not driven by

loyalty_score — it is driven by the opposition between the player's sustained integrity pattern and the

companion's loyalty_belief_axis pole; the score acts only as inertia.

• opposition = a 0..100 magnitude of belief-distance, computed from the integrity band (WS_001/WS_002),

not from observed deeds (integrity is a real property read directly).

• triggers map the enum: never = no trigger; complete_opposition = 85; major_opposition = 60;

any_sustained_misalignment = 35; circumstance_only = plot-driven, not integrity-driven.

• inertia, capped: effective trigger = base + (loyalty_score / 10), bonus capped +15, effective trigger

capped 90. The 90-cap invariant is ratified and may not be tuned away: opposition tops at 100 > any

reachable trigger, so complete opposition ALWAYS forces departure and investment can never buy immunity

("invested is sticky but not immune," stated as a theorem).

• WAVERING fires at effective-trigger −15 — a visible warning window to course-correct before DEPARTED.
• sustained, never instant: the trigger must hold across two chapter nodes before DEPARTED (§5.3

cumulative-pattern rule — a single dark choice never strips a companion).

• The eight magnitudes (85/60/35, ÷10 inertia, +15 cap, 90 cap, −15 WAVERING, two-node sustain) are

ratified as authored STARTING VALUES; playtest may move them without a re-ruling, but the SHAPE and

the 90-cap are fixed.

Two ratified-but-inert numbers the roster brief must know (surfaced to Josh, invariant survives):

1. The +15 inertia cap can never bind: it requires loyalty_score = 150, but the domain is −100..+100, so

the max reachable bonus is 10.0. The 90-cap invariant still holds; one stated number is dead

(RATE_MISMATCH_ANALYSIS §self-correction, verified).

2. The "only for_life is exempt" clause is refuted by the live spine: CH_65 stages "Ntala departs," and

Ntala is rowed for_life/never; Josh's D-23 ruling offers departure on every surviving band. The exempt

companion canonically departs. The 90-cap invariant is untouched — only the exemption clause needs

alignment at the wiring commit. **Implication for casting: for_life does not mean "present every node to

the Epilogue"; it means the model's departure trigger is never, which a scripted plot departure (D-23)

can still override.**

---

5. INTERFACE 5 — DRAWN (the join mirror) and the worldstate ledger

• DRAWN is departure's mirror: a drawn_conditional companion joins when the player's integrity band

aligns with their belief pole at magnitude ≥ 60, sustained the same two-node way

[SRC: RELATIONSHIP_TIER_BANDS_BRIEF.md # L224; memory companion-loyalty-integrity-system].

• Loyalty models resolve their join behavior: for_life (anchor, joins at its trigger and stays), `invested_

sticky (joins on investment, high inertia), come_and_go` (rejoins on corridor returns, leaves by

circumstance), drawn_conditional (DRAWN by aligned integrity, repelled by opposed), not_companion.

The engine drives two worldstate rows (WS7 Increment C, T0_Worldstate_Variables)

[SRC: WS7_WIRING_BUILD_PLAN.md # L147-154]:

• WS_027 companion_roster_state — the live traveling-party ledger: per-member `{character_id,

loyalty_model, loyalty_state, loyalty_score(−100..+100), joined_chapter, last_integrity_signature}`.

Membership + loyalty are seeded from the Character_Index anchors. It **REFERENCES WS_008

relationship_history** for encounter history rather than restating it.

• WS_028 companion_loyalty_transition_events — queued `{character_id, event(DRAWN|WAVERING|DEPARTED|

REBONDED), chapter, integrity_at_event, rerouted_paths[]}`. **Every DEPARTED event MUST carry a non-empty

rerouted_paths[]** — the no-dead-end contract, enforced by harness/check_no_dead_end.py.

Adjacent variables a companion couples to but must not duplicate: WS_008 relationship_history (general

per-NPC ledger), WS_021 faction_standing_per_faction (THE STANDING axis — separate, Interface 8),

WS_012 familiar_bond_state (the familiar axis — separate, Interface 9). WS_027 references, never restates,

WS_008.

---

6. INTERFACE 6 — trades (companion as teacher or peer)

Ten canonical trades, each with one ancient + one modern home base, each 12-tier quest-gated (up to 120 gates)

[SRC: registries/T0_Trade_Registry [DRAFT v0.1]/T0_Trade_Registry DRAFT v0.1.csv]:

A companion's teaching_role may make them:

• a trade teacher — a trusted teacher opens a trade home base as an *opportunity* (kit/progression

class), gating a mastery-tier climb. Real-science disciplines bind the substrate: Alchemy=chemistry,

Engineering=physics/statics, Herbalism=biology; the game rewards learning the real science

[SRC: memory rewards-intelligence-real-science-immersive-sim].

• a trade peer — a battling-partner who shares a trade lane and grows alongside the protagonist.

Non-gating constraint: a trade opportunity a companion opens or closes is by construction

non-progression-critical — the trade climb is completionist/kit, and losing a teacher must reroute to an

equivalent-class alternative (opportunity conservation), never dead-end the trade

[SRC: docs/proposals/ENTITY_PERSONA_SYSTEM.md # §3.4 Mechanism 1-3]. Real trade home-base = an

A_TRADE-layer dwelling with added_modules_only integrity visibility and a cultural_fabric_protected

siting rule (the care firewall — no integrity-scar aesthetics on a living culture's trade home)

[SRC: WS7_WIRING_BUILD_PLAN.md # L216, L246].

---

7. INTERFACE 7 — combat (what an ALLY contributes mechanically)

The fight system is Pillar-8 blended-mode over seven canonical modes: `direct / stealth / puzzle_routed /

environmental / vril_management / familiar / astral [SRC: docs/COMBAT_ENCOUNTER_SYSTEM.md # Part B §2 L143;

_source/.../T1_Combat_System_Spec [ACTIVE v1.0].md § modes]`. Difficulty is world vril-density, never enemy

stat inflation [SRC: COMBAT_ENCOUNTER_SYSTEM.md # Part B §0 L111]. A companion is an ally combatant, and

the contract on what an ally supplies:

• A verb / mode contribution. An ally who fights beside the player contributes to breaking Composure

— the mode-agnostic posture bar with named sub-nodes, each depletable by a specific mode's verb

[SRC: COMBAT_ENCOUNTER_SYSTEM.md # Part B §5 L196]. A companion whose combat identity is stealth or

puzzle_routed can break a node raw DPS cannot. The ally's mode should compose with, not duplicate, the

player's — the design intent is blended coverage.

• Real-science + vril, wielded by allies too. The game runs **real physics AND real magic (vril alters

real physics), and player, enemies, AND allies all wield both** for good and ill

[SRC: memory vril-altered-physics-real-physics-and-magic]. An ally's contribution is expressed in the same

verb grammar and the same vril-altered-physics substrate, not a separate "buddy AI" system.

• Vril asymmetry. An ally's vril capability is tier-bound like the player's — early-chapter perception is

NOT tier-6 Vril Sight; a companion cannot read vril-signatures at Neophyte tier

[SRC: memory early-chapter-perception-not-tier6-vril-sight].

• The familiar paired-operation is a distinct, named affordance. The familiar Composure node is broken

via a bonded paired-operation (canon at Ch 73 Sub-Phase C, TGP-exclusive)

[SRC: COMBAT_ENCOUNTER_SYSTEM.md # Part B §5 L196, §3 L160]. This runs on the familiar axis, not the

human-companion axis (Interface 9) — a human companion contributes a human-mode verb; a familiar contributes

the familiar-mode paired-op.

• Protected-community firewall. Combat is never staged on a protected living community; where combat

reaches a protected space it is antagonist intrusion, and the environmental-boss protocol (Perception →

Vril-management → Restraint → Purgation) is the playable layer `[SRC: COMBAT_ENCOUNTER_SYSTEM.md # Part A §9

L64; memory rated-r-aaa-dont-defang-monsters]`.

Combat detail (movesets, verbs, counters, phase escalation) is authored per-node at the cascade, not

pre-specified here [SRC: docs/proposals/RELATIONSHIP_COMBAT_SYSTEMS_NOTE.md # §5].

---

8. INTERFACE 8 — travel (the mothership; companions live aboard)

The travel model is one transforming vril-flagship mothership + collectible local sub-craft; companions

travel aboard `[SRC: registries/T0_Vehicle_Registry [DRAFT v0.1]/Sheet1.csv; WS7_WIRING_BUILD_PLAN.md #

L12-49]`.

• VH_001 The Mothership-Hub — flagship-mothership, acquired Ch 08 (Swahili coast onset), upgrade

chain raft→dugout→dhow→caravel→research-vessel→submersible-capable, transform configs

cruise;battle;dive;anchored-base. Its crew_station_object names the stations a companion can hold:

helm; lookout; gunner; engineer; cook; menagerie; rigger. **A companion's own_life aboard the mothership

should bind to a crew station** — this is where the "lives aboard" obligation becomes concrete.

• Sub-craft deploy FROM the mothership: VH_002 submersible pod (z-axis-deep, pilot = bonded-water-

familiar/Trade-10), VH_003 the Vimana (peer-flagship, air, Ch 55+; docks as a peer, never carried).

A companion may crew or pilot a sub-craft (e.g. a Trade-10 navigator companion at the helm; a bonded

water-familiar handler piloting the pod).

• Firewalls that bind companion travel writing: the craft is never called a "Vessel" (C1 firewall);

geography stays walked (no flight pre-Ch-55, F-a); the mothership is a single craft (§6.5), so companions

concentrate aboard one hub, not a fleet of homes. The anchored-base config + the Layer-B dwelling-home

are where a companion resides between deployments [SRC: WS7_WIRING_BUILD_PLAN.md # L215 DW_0002].

Chapter assignments for craft are Josh-canon: only Ch 08 (dhow onset) and Ch 55 (Vimana) are cited; a

companion's boarding chapter must reference an existing acquisition, never mint one.

---

9. INTERFACE 9 — attackability (companion = Tier-3 bonded) + the no-dead-end invariant

Five attackability tiers, enum hostile | bystander | bonded | protected-collective | shielded

`[SRC: docs/proposals/systems/attackability-and-consequence-model.md # L15, L65; WS7_WIRING_BUILD_PLAN.md #

L166-171]. A companion is Tier-3 (bonded`):

• Attackable, bond-mediated, maximal weight. Turning on a companion is a real dark beat — they leave, die,

or become an enemy; questlines close; a permanent mark lands. Not god-moded, not invisible-walled —

protected by CONSEQUENCE and BOND, not a can't-target flag `[SRC: attackability-and-consequence-model.md #

L15, L33]`.

• is_load_bearing (bool) — true only on the narrow roster of diegetically-resilient main-story figures

(spine canon = Josh's). Ntala is the seeded bonded/is_load_bearing=true example

[SRC: WS7_WIRING_BUILD_PLAN.md # L169].

The non-gating invariant governs every companion opportunity `[SRC: docs/proposals/ENTITY_PERSONA_SYSTEM.md

§3.4-3.5]`:

• progression-critical immune set — anything whose loss would stop progression/achievement/100% is

progression_critical and CANNOT be an opportunity; companion-gated affordances are by definition the

complement (kit/route/relationship/lore only).

• three playstyle through-lines — participate / subvert-from-inside / honorable-open; each node completes

on all three, each opening/closing different companion opportunities.

• opportunity conservation — any opportunity a companion's departure closes has a declared equivalent-class

alternative on the divergent path (a lost broker reroutes the network).

• the no-dead-end gate walks the spine graph and hard-fails any state that empties the forward set;

every DEPARTED ships rerouted_paths[]. Losing a companion reshapes routes; it never blocks the game.

---

10. INTERFACE 10 — the axis firewalls (never conflate)

Three axes touch a companion; keep them structurally separate:

• THE BOND (companion) — loyalty_score, WS_027, labels Enemy…Family, driven by integrity-opposition.

This is the companion axis.

• THE STANDING (faction/reputation) — WS_021 faction_standing_per_faction, labels Hunted…Exalted, a

per-faction −100..+100 accumulator, distinct from integrity and from THE BOND

[SRC: _source/.../T1_Integrity_Paths_Worldstates_Master [ACTIVE v2.1].md # §8.7]. A companion has a BOND;

their faction has a STANDING; high BOND does not imply high STANDING with their group.

• THE FAMILIAR AXIS — 22 familiar slots (21 base + 1 path-variable childhood familiar reunited **Ch

58), creature-class, bound by binding-stone, tracked by WS_012 familiar_bond_state**

[SRC: COMBAT_ENCOUNTER_SYSTEM.md # Part A §8 L61; memory layer-model-v2]. Familiars are T0_Creature_Roster

/ T0_Familiar_Registry entities (the Familiar_Registry is currently an empty stub, 0 rows — a hard

anchor dependency). A familiar is NOT a companion: it does not carry a loyalty_model, does not sit on

THE BOND, and does not depart on integrity-opposition. WS_012 is heavily written (99 writes / 38 nodes);

WS_008.bond_state (the NPC axis) is genuinely unwritten — do not confuse a bond_state grep with the

familiar axis.

---

11. THE DEGREES OF FREEDOM — what is PINNED vs where design room is real

PINNED (design WITH these; changing them re-litigates a ruling)

• The BOND ladder — nine bands, the exact ranges, the companion labels Enemy…Family, per-axis labels over

one shared numeric axis (Ruling 2 + 4). A per-axis threshold override is the *declined* variant and needs

a NEW Josh ruling; only a per-faction override is tuning `[SRC: RELATIONSHIP_TIER_BANDS_BRIEF.md #

Authority Note L20-32; RATE_MISMATCH_ANALYSIS §8 smuggling-vector]`.

• The departure rule SHAPE — opposition-magnitude, integrity-driven, inertia-as-non-immunity, the

90-cap invariant, WAVERING as a computed band, two-node sustain (Ruling 3).

• The five loyalty models, the 7 loyalty states, the departure-threshold enum, the **5

attackability tiers, companion = Tier-3 bonded**.

• The 13 persona fields + 6 person-extension fields; a companion fills all.
• The non-gating invariant and its four mechanisms + the no-dead-end gate; rerouted_paths mandatory.
• The axis firewalls — BOND vs STANDING vs 22-slot familiar axis.
• The travel model — single transforming mothership + sub-craft; companions live aboard; walked-geography

and no-flight-pre-Ch-55 firewalls.

• The 10 trades and their home-base chapters; the 7 combat modes; difficulty = vril-density.
• Reveal discipline — no who-frame/agent-noun pre-Ch-38; Cassius named Ch 76+; no thesis declaration.
• §17.1 care — living-faithful never a combat target/villain; no sacred name on a mortal; care = weight +

authenticity, thin substrate is a book-processing dependency, never faked.

• The ratified spine facts — Ntala (for_life) and Ana (come_and_go) are the two spine-proven

travellers, rowed; the Ch-1 best-friend choice seeds Ntala's gender/skills/combat-focus; Ch-58 childhood-

familiar reunion; Ch-65 path-lock; endgame Realm Road 66-77.

GENUINE DESIGN ROOM (the roster brief's actual job — propose here, labelled, for Josh)

1. Roster size (~30). Explicitly Josh's OPEN QUESTION; the "~30" is an unratified scope estimate. The

brief's job is the grounded answer from the spine's staged cast, not a guess.

2. Which canonical figures are companions, and which loyalty_model each carries — including the four

travelling characters (CHAR_0140/0144/0145/0146) that currently carry no model, and whether Tunde's

drawn_conditional fits a one-node presence or he needs more beats (RATE_MISMATCH_ANALYSIS §6). Cardinal —

present as a brief, do not populate.

3. The loyalty_belief_axis controlled vocabulary — the enumerated set of value-poles, aligned to the

T1_Integrity bands. Naming-canon → propose for ratification.

4. Each companion's loyalty_join_trigger (chapter + condition) and loyalty_departure_threshold

assignment — grounded in the companion's motivations/belief pole, within the ratified enums.

5. Each companion's culture, persona, trade lane, combat mode/verb, crew station, and web edges — the

creative-within-vision bulk; apply-and-show-the-path where grounded, but every NAME flags for Josh.

6. The calibration TARGET (design intent, not a number) — e.g. "Ntala → Family at ~Ch-65 path-lock; a

merely-good run misses it." Cheap, genuinely-Josh, guides later authoring when a consumer exists. The

per-beat RATE is retired — do NOT resurrect it as a question.

7. age_stage_presence and come-and-go windows — WHEN each companion is available and rejoins/leaves,

grounded in the arc structure (ancient 2-38 region-per-chapter, modern 39-65, endgame 66-77).

8. Where a culture's substrate is thin — flag the specific book-processing enrichment dependency rather

than author unsupported cultural depth.

---

12. The two burned failure modes this contract is built to prevent

• Undisclosed filters behind measured numbers. Every count here is stated with its source; the retired

per-beat "rate" and the "bond-moving beat" filter are named as retired, not silently reused.

• Seeds/proposals read as canon. The 5 seeded loyalty rows, the "~30", the belief-axis tokens, and every

name are labelled provisional/illustrative; only Josh's four D-REL-BANDS rulings, the WS7 landed schema, and

the cited T1/CVD hard lines are treated as fixed.

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