PILLAR10_REGIONAL_COMBAT_VOCAB.md

systems/PILLAR10_REGIONAL_COMBAT_VOCAB.md

PILLAR 10 — THE FIVE REGIONAL COMBAT VOCABULARIES

CANON SUBORDINATION — derivation record; serves canon, never outranks it. Authority:

docs/DOC_MAP.md § 0. The CANON of this pass is the twelve populated rows of

T0_Regional_Combat_Modifier (a T0 registry, and therefore canon under

docs/SUBAGENT_CONTRACT.md §1). This file is the evidence trail behind those sixty cells: what

each token means, which canon line warrants it, and which two regions it tells apart. Read the

registry for the values; read this for why.

Ruling that commissioned it. Josh, 2026-08-04, verbatim: *"Why do you think this is difficult?

This is an action adventure fantasy RPG and we've provided thousands of hours of canon and

architecture and vision and tier docs and comparable rpgs"*. That ruling SUPERSEDES the earlier

plan to fill one axis and board the other four, and it supersedes the sitting-26 mint decision

that landed all five columns empty by rule. Boarding was the error. All five are authored here.

What the pass does NOT do. It names CLASSES and never numbers. Every magnitude — how much a

guard yields, how far a hazard reaches, what a movement modifier costs — stays behind the Phase-5M

tuning firewall exactly as the rest of T1_Combat_System_Spec's numerical bounds do. A class with

a multiplier attached would be a tuning decision wearing a canon hat.

---

1. The five axes and their canon warrant

CVD § 12.5 *Regional combat identity (Pillar 10)* names five axes in one sentence and delegates

their detail: *"Stance systems per region; environment hazards per region; weapon-of-origin

bonuses; movement modifiers (jungle vs desert vs steppe vs ice); ability interaction changes per

region. Detail → T1_Combat_System_Spec + Regional Combat Modifier table."* The five columns of

T0_Regional_Combat_Modifier are those five axes verbatim and in that order.

Three further canon surfaces bind the shape of the work.

• T1_Combat_System_Spec § Combat Variety Framework restates the five axes as the per-region

variety lane, and § Nature-Derived Counter Classes gives the standing method: a class is *"a

reading of the attack, recorded as data so the engine and the gates can check it"*, never a free

choice among equally available options. The same method governs a reading of a REGION.

• T1_Build_Pipeline_Contracts § Combat System lists the same five as build contract items —

*"weapon-of-origin bonuses per Gate 10 regional modifiers, stance system per region, environment

hazards per region, movement modifiers across jungle desert steppe ice"*.

• docs/COMBAT_ENCOUNTER_SYSTEM.md § 6.1 states what the axes are FOR: *"The same boss in a

different chapter's arena plays differently because the modifier space differs — our aspect /

room-modifier variety, doubling as Pillar 10 regional combat identity."* The region row is the

region-scale modifier space every arena inside it inherits.

1.1 Genre-standard defaults, and where we deviate

Per the thirty-second-sitting ruling carried in docs/GENRE_BASELINE.md, the industry-standard

architecture is the DEFAULT and only deviations get named. The comparator set is the ruled AAAAA

bar — Witcher 3, Monster Hunter Wilds, GTA 6, FF7 Remake — and MH Wilds is the strongest regional

comparator because it is the one that makes the LOCALE a combat system rather than a backdrop.

• Adopted as standard. A per-locale environmental hazard and state layer that changes how a

fight is fought, not merely how it looks (MH Wilds' locale states and environmental traps). A

small closed set of postures the player reads and commits from (the stance/mode families every

action RPG in the set carries). A traversal-cost class attached to ground type (universal). A

stagger-and-punish economy in which the reward for a correct read is an opening rather than a

flat damage bonus (FF7R's pressure-and-stagger loop; MH Wilds' offset and wound punish).

• Ours by vision, named as a deviation. The five axes are keyed on a REAL CULTURE'S DOCUMENTED

MARTIAL RECORD rather than on a biome preset. stance_system_class is read off the region page's

own Section 6 — the caci contest, the kalari pose curriculum, the Fon standing army — which is

what makes Pillar 10 a knowledge pillar and not a skin. No comparator does this, and it is the

reason the vocabulary could not be lifted from one.

• Ours by vision, named as a deviation. weapon_of_origin_bonus_class is REGIONAL and not

per-weapon. Every weapon sharing an origin region shares its bonus shape, because a

weapon-of-origin bonus is a statement about the place that taught the weapon. It composes with,

and never restates, the weapon's own mechanic — the Madu keeps its dedicated parry catch from

T1_Combat_System_Spec § Parry Mechanic AND carries South India's chain_extension.

1.2 The bar every token had to clear

• Closed. The token set is enumerated here and nowhere else; a value outside it is a defect.
• Orthogonal. No two axes may partition the twelve regions the same way. Verified below.
• Discriminating. Every axis splits the twelve into at least two groups, and no axis pair groups

more than 4 of 12 regions identically (§ 7.1). A token that every region carries measures nothing.

• Extensible to 76. The slice is twelve rows; the arc is seventy-six. Each vocabulary was sized

so the remaining sixty-four regions land inside it rather than forcing a re-mint. Members that no

slice region uses are declared, not dropped — an unused member is the arc's, not a defect. **This

is a stated EXEMPTION to the bar above, not a silent one:** four tokens are carried by no slice

region (distance_kept, sand_drag, ice_slide, none), they are listed in § 8, and the apply's

DECLARED_UNUSED set must match the observed unused set exactly — so a token that quietly dies,

and a stale exemption whose token became used, both fail the self-test. An earlier draft stated

this bar per-TOKEN while the code enforced it per-AXIS; the two now agree.

---

2. stance_system_class — seven tokens, six postures and one declared null

The posture the region's own documented martial record teaches: where the guard comes from, what

reach it holds, and what the decisive beat is. Derived per region from the region page's Section 6

*Weapons Military and Conflict*.

Discriminating pair, worked. pose_to_pose vs held_line: South India teaches a solo

curriculum whose unit of measure is one practitioner's transition between named poses; West Africa

teaches a drilled formation whose unit of measure is the line. Both are documented martial systems

of the same era, and a fight staged in one reads nothing like a fight staged in the other.

Two corrections applied 2026-08-04, both from a fresh-context critic pass. Both cells read

distance_kept in the first authoring, and both were wrong for the same underlying reason: the axis

was read off the most vivid archery sentence on the page rather than off the combat the chapter

actually stages.

• rift_valley: distance_kept → no_stance_system. The old cell inferred a posture from the

withdrawing animal's behaviour — a creature's reaction, not a system the region teaches — and

imported its semantics wholesale from South Africa's bow, a different region. Worse, it filled the

one absence the page explicitly declares to be design. The weapon axis had already solved exactly

this problem one column over with no_origin_bonus; the stance axis simply had no null to reach

for, so the vocabulary gained one rather than the region gaining a posture it does not have.

• south_africa: distance_kept → guard_and_answer. The old cell was taken from the /Xam

HUNTING kit. That kit is real and documented, but § Section 6 fixes this chapter's combat framing

BY RULING — the machinery, the overseer and labor-enforcer, the permitted fauna, the materialized

rock-art echo and a fictional gold-trade broker are the only combat routes, and the region's seven

named living communities are inviolate [SRC: T0_Hard_Lines @ HL_0003]. A keep-away posture is also

directly refuted by the only human encounter the ruling admits, whose documented counters are

parry and stagger-break. The stance cell is now read off that encounter.

Why prepared_release still stands on South Africa's weapon axis. The weapon-of-origin axis is a

CRAFT statement about what the place taught (§ 4), not a statement about who may be fought. Recording

that this region's documented archery is patience-and-status rather than impact honours the material

record per CVD § 17.1; HL_0003 governs the encounter surface, and nothing on that surface routes

through the protected communities. The fence is now stated in the row note rather than left implicit.

Care note. Every token above is read off a documented martial tradition treated as craft, per

CVD § 17.1. None of them is a claim about a people's character, and the classes attach to REGIONS

and encounters, never to a community as a combat target — the rift_valley reading in particular

carries the region page's own statement that the absence of a martial layer there is the chapter's

design and not a hole to be filled, which is now expressed by a declared member instead of a

borrowed posture.

---

3. environment_hazard_class — seven tokens

The region-scale hazard cycle: what the environment does to a fight that no enemy in it does.

Composed from the region page's Section 1 climate and terrain statements and, where the region

carries them, the live T0_Environment_Grammar_Registry exposure_hazard and

tactical_affordance_class values on its own zone-catalog environment_grammar_refs.

Why worked_ground is a member and not an exception. It is the only token whose generator is

human rather than climatic, and it is the one the rest of the arc needs most — the extraction,

facility and installation chapters all read on it. Minting it here, from the one slice region that

already carries a machine-timed boss, is cheaper than discovering it at Ch 46.

---

4. weapon_of_origin_bonus_class — nine tokens

Which axis of a weapon's performance its ORIGIN REGION lifts when the weapon is carried there.

T0_Item_Derivation_Rule IDR_10 states the join in its own reachability_expr:

weapon.cultural_region_ref == region_id.

This axis is regional by construction. It never restates a weapon's own mechanic — the parry-capable

roster in T1_Combat_System_Spec § Parry Mechanic keeps its defensive_class, and the origin bonus

composes on top of it in a separate lane, exactly as Lane A proficiency bonuses compose with Lane B

quest-gate bonuses in § Weapon Proficiency Bonus Lanes.

4.1 How T0_Weapon_Registry.regional_bonus_class is derived, 72 of 72

The column is emitted by harness/apply_pillar10_vocab.py, never hand-typed, and re-running the

script reproduces it byte for byte. Every one of the 72 rows carries a declared value:

• A vocabulary token, where the weapon's cultural_region_ref names a region that HAS a

T0_Regional_Combat_Modifier row. The region's weapon_of_origin_bonus_class is copied across

unchanged; the region is the authority and the weapon inherits.

• PENDING_RCM_ROW, where cultural_region_ref names a real region that has no RCM row yet.

The RCM row set is the slice by rule; the remaining regions land with the region cascade, and

each one resolves its own weapons on arrival. This is a MEASURED state with a named resolver,

not a blank.

• NO_ORIGIN_REGION, where cultural_region_ref is empty — canon attributes these forms to no

single region, so no origin bonus exists to inherit. Also a declared state.

The split is reported as a positive control by harness/emit_item_matrix.py on every run, so the

day a region page lands and the pending count drops, the drop is visible rather than assumed.

Which tokens the live join actually reaches, today. Six of the twelve RCM regions are the origin

of no weapon at all (flores_island, angkor_khmer, sri_lanka, swahili_coast, rift_valley,

west_africa), so four of the nine tokens — guard_yield, formation_reach, close_footing and

no_origin_bonus — are carried by zero rows of the derived weapon column. Only chain_extension

(4 weapons), prepared_release (2), opening_conversion (2), guard_break (1) and control_hold

(1) reach a live weapon. This is worth stating because a token's warrant can name a weapon the join

does not deliver: guard_yield is grounded on flores_island seating the arc's first off-hand Shield,

but WPN_033 Shield carries an empty cultural_region_ref and therefore derives NO_ORIGIN_REGION,

not guard_yield. The region-scale authority claim above is unaffected — the region is still the

authority, and its token is still correct for the weapons it will eventually own — but the axis is

not yet a live weapon effect for most of its vocabulary, and should not be read as one.

---

5. movement_modifier_class — nine tokens, six inherited and three minted

CVD § 12.5 names this axis by example — *"jungle vs desert vs steppe vs ice"* — and

T1_Build_Pipeline_Contracts § Combat System repeats it. A vocabulary already existed for the

region-scale twin of this column: docs/proposals/systems/WS7_WIRING_BUILD_PLAN.md declares

T0_Region_Index.movement_modifier as `jungle_slow | ice_slide | sand_drag |

altitude_stamina_drain | bog_immobilize | none, and four live region rows carry jungle_slow`

today. That vocabulary is inherited whole rather than replaced, and then EXTENDED — six inherited

members plus three minted ones, so the two columns share a dialect on the inherited six and diverge

on the three.

What that costs, stated plainly. terrace_step, deck_pitch and scree_slip are not members of

the T0_Region_Index.movement_modifier enum WS7 declares, so five of the twelve rows (bali,

sumatra_java, swahili_coast, rift_valley, south_africa) carry a value the sibling column cannot hold

today. A repo-wide search for the three tokens returns exactly three files: this record, its apply

script, and its own registry. Nothing else in the corpus knows them yet. So the two columns are

not readable against each other without a translation step on those five rows — widening the

T0_Region_Index.movement_modifier enum to the same nine members is the named work order that closes

it (§ 10). An earlier draft of this section claimed no translation step was needed; that claim was

false for those five rows and is corrected here.

Three members are minted because the slice's own ground demands them and the inherited six cannot

express them. Per docs/GENRE_BASELINE.md, the base locomotion set is closed with the genre

default and *"the item-26 locomotion entry narrows to the Pillar-10 regional traversal verbs, which

are the only genuinely ours"* — these three are exactly that narrowing.

Divergence recorded, not silently applied. T0_Region_Index.movement_modifier reads

jungle_slow on bali, sumatra_java and angkor_khmer, all three at `env_profile_status =

stub. This pass reads those three as terrace_step, terrace_step and bog_immobilize`

respectively, on the evidence above — and notably Bali's own zone catalog contains **no forest zone

at all**. Nothing in T0_Region_Index is edited by this pass: the divergence is a FINDING for the

director, carried in each RCM row's note and repeated here, and the region-profile column is that

lane's to reconcile.

---

6. ability_interaction_class — six tokens

Which ambient substrate the region hands the ability layer, and therefore which physics chains are

live in it. T1_Ability_Tree § 5G *Physics-Driven Ability Interactions* is the governing frame: the

six-property taxonomy is *"mass, structural_integrity, flammability, thermal_state, conductivity,

organic_substrate"*, and abilities read the environment rather than ignoring it — Entangle is

*"Effective in any biome with living vegetation; reduced effectiveness in arid or arctic

environments per atmospheric-substrate availability"*, Flow Redirect deflects *"when ambient water

is present"*, Flame Shape *"shapes ambient fire (campfires, torches, brazier flames)"*. The

recovered Josh-authored matrix at docs/original_predclaude/ELEMENTS_MATRIX_RECOVERED.md § 4 names

the same doctrine as original DNA: *"Earth needs earth present... Water needs water... Fire blocked

underwater/precipitation/cold"*.

The per-region evidence is the live T0_Environment_Grammar_Registry elemental_affinity and

ambient_state_gift micro-grammar, aggregated over each region's own zone-catalog

environment_grammar_refs.

---

7. Orthogonality, checked rather than asserted

No two axes partition the twelve regions the same way. The pairs that come closest, and why they

still separate:

• environment_hazard_class vs ability_interaction_class. Both mark flores_island and

sumatra_java together (volcanic ground, thermal substrate). They separate on the monsoon three:

bali and sri_lanka read monsoon_saturation / conductive_medium, while south_india reads

monsoon_saturation / lithic_mass — the same weather over a dry interior of stone.

• stance_system_class vs weapon_of_origin_bonus_class. Both are read from Section 6. They

separate on sri_lanka (guard_and_answer but formation_reach), on ethiopia (held_line but

guard_break) and on central_africa_congo (footing_denied but control_hold) — posture and

bonus-shape are different questions about the same record.

• environment_hazard_class vs movement_modifier_class. Both are read from Section 1. They

separate on ethiopia and rift_valley, which share altitude_cold and differ on

altitude_stamina_drain vs scree_slip; and on bali, sri_lanka and south_india, which share

monsoon_saturation and split terrace_step / jungle_slow / jungle_slow.

7.1 What the coded control can and cannot see

The apply's orthogonality control compares PARTITIONS, and partition inequality is a boolean that

passes on any single disagreement — two axes could agree on eleven of twelve regions and still read

as independent. That was a real weakness, and the control is now two-part: partitions must differ

AND no axis pair may group more than 4 of 12 regions identically, which is the measured ceiling

on the live assignment (stance/weapon, hazard/ability and weapon/movement all sit at 4).

The bar is held tight rather than slack, so worsening the collapse fires the self-test.

Its honest limit: a structural control cannot see two cells citing the SAME SENTENCE. Three

single-region restatements exist in the current table and are named here rather than left for the

next reader to find:

• south_africa worked_ground (§ 3) and worked_machine (§ 6) read the same clause — the

steam-vent burst, the crush lane, and the boss's whip-calls on machine timings. They still earn two

cells because they answer different questions about it: the hazard axis says the arena's cycle is

machine-made rather than weathered, and the ability axis says the substrate the ability layer is

handed is machinery and its by-products rather than water, stone or vegetation. An arena that

ignored the second would still run the first as set dressing.

• **swahili_coast footing_denied (§ 2), close_footing (§ 4) and deck_pitch (§ 5) all read the

boarding clause.** They separate as posture / bonus-shape / traversal cost: the stance says the

guard shortens, the weapon axis says the origin region lifts short-reach control, and the movement

class says dodge distance itself shrinks. Three consequences of one fact is not the same defect as

one fact recorded three times — but the shared provenance is real and is disclosed.

• **ethiopia altitude_cold (§ 3) and altitude_stamina_drain (§ 5) both read the Ras Dashen

line.** Hazard is what the environment does to the fight (exposure, cold, falls); movement is what

it charges per commitment. The same altitude drives both, which is why they were never going to

separate on evidence.

Detecting restatement is a reading task, so it belongs to the fresh-context critics, not to the

self-test. The self-test's job is to make a collapse impossible to introduce SILENTLY.

---

8. Measured absences, positive-controlled

Every zero below was measured with a search that CAN match, and each names what a non-zero would

have looked like.

• Two region pages carry no environment_grammar_refs on any zone row. angkor_khmer and

south_africa both HAVE the column in their zone-catalog header and both leave every cell of it

empty; a repo-wide EG_ search over the two files returns zero. The parser found the column and

found nothing in it, so this is an absence and not a miss. A non-zero would look like the 13-to-28

refs every other slice page carries. The two fall back differently, and the difference matters:

angkor_khmer IS read from its region row, which is populated (`ambient_elemental_affinity =

water:enhance;earth:enhance, weather_cycle_profile = monsoon;flood_pulse;clear`,

movement_modifier = jungle_slow, env_profile_status = stub). south_africa is read from

Section 1 and Section 6 prose ONLY — its T0_Region_Index row carries no env profile at all,

every one of those five fields being empty, so the region-row half of the fallback does not exist

for it. An earlier draft of this bullet claimed both regions fell back on a region-row profile;

that was true of angkor_khmer and false of south_africa, and is corrected here.

• The empty-region-row condition is far wider than those two, and was not measured before. Five

of the twelve slice rows carry no env profile at all (ethiopia, rift_valley,

central_africa_congo, west_africa, south_africa) and three more carry only

terrain_feature_class_ids (sri_lanka, south_india, swahili_coast). rift_valley is the

sharp case: its row is empty while its PAGE carries 21 EG refs, so its EG evidence comes from the

zone catalog and never from the region row. Only flores_island, bali, sumatra_java and

angkor_khmer carry a full row, and all four are env_profile_status = stub. Routed to the

region-profile lane (§ 10).

• Four vocabulary tokens are declared and unused by the slice — distance_kept on the stance

axis (§ 2) and sand_drag, ice_slide, none on the movement axis (§ 5). Their homes are named

there. A vocabulary whose every member the twelve-row slice uses would be a vocabulary sized to the

slice rather than to the arc. The set is machine-declared and exactness-checked (§ 1.2).

• 48 of 72 weapons resolve to PENDING_RCM_ROW and 14 to NO_ORIGIN_REGION. Only 10 weapons

carry a cultural_region_ref inside the twelve RCM rows. That is the honest state of a

twelve-row modifier table against a whole-arc weapon roster, it is reported every run, and it

shrinks as the region cascade lands rows.

• No magnitude was authored anywhere. A repo search for a number attached to any of the 60

cells returns zero, by construction — the columns hold class tokens only, and the Phase-5M

firewall owns every multiplier.

---

9. Consumers

• registries/T0_Regional_Combat_Modifier [DRAFT v0.1]/Sheet1.csv — the canon rows. 60 of 60 axis

cells populated by this pass.

• registries/T0_Weapon_Registry [DRAFT v0.1]/Sheet1.csv — regional_bonus_class, 72 of 72,

derived from the origin join in § 4.1.

• registries/T0_Item_Derivation_Rule [DRAFT v0.1]/Sheet1.csv — IDR_10, flipped off

DERIVED_BLOCKED. It was the one derivation rule that could not run; it runs now.

• harness/apply_pillar10_vocab.py — the deterministic apply. Idempotent, self-testing, and the

only writer of these cells.

• harness/check_pillar10_vocab.py — gate pillar10_vocab, the INDEPENDENT reader. The apply's

membership test is the writer checking its own output; this gate declares the five vocabularies

separately and enforces them against the live registries, so a hand-edit to any axis cell fails a

gate rather than passing unnoticed. It also re-derives the weapon origin join and cross-checks its

own vocabulary declaration against the apply's, so the two copies cannot drift apart in silence.

• harness/emit_item_matrix.py — reads the populated axes, projects the IDR_10 cell space, and

positive-controls the weapon-resolution split every run.

• Encounter and arena authoring — docs/COMBAT_ENCOUNTER_SYSTEM.md § 6.1's region-scale modifier

space, which every arena inside a region inherits.

10. Open follow-ups for the director

• T1_Combat_System_Spec § Combat Variety Framework is the canon home CVD § 12.5 delegates to, and

it should gain a pointer paragraph naming these five vocabularies and this file. That is a canon

edit on a file three other lanes are live on, so it is raised here rather than taken.

• T0_Region_Index.movement_modifier diverges from this pass on bali, sumatra_java and

angkor_khmer (§ 5). All three of those region rows are env_profile_status = stub; reconciling

them belongs to the region-profile lane.

• WIDEN T0_Region_Index.movement_modifier to the same nine members (§ 5). Until it holds

terrace_step, deck_pitch and scree_slip, five of the twelve RCM rows carry a movement value

the sibling column cannot express, and any lane that joins or diffs the two columns hits five

untranslatable values. This is the named cost of the three mints, not a defect in them.

• EG_0002 on the angkor_khmer region row is a data defect (§ 6). It resolves to *Sieidi Stone

(protected)*, a Sami sacred marker with protected_care_ref = HL_0001;HL_0007, sitting on a Khmer

region row — traceable to the WS7 seed rows, which used EG_0001;EG_0002 as placeholder values. A

protected sacred marker mis-assigned across cultures is worth clearing on care grounds alone, not

merely on data-hygiene grounds. Region-profile lane.

• Five of the twelve region rows carry no env profile at all — ethiopia, rift_valley,

central_africa_congo, west_africa and south_africa, each with all five of

ambient_elemental_affinity, weather_cycle_profile, terrain_feature_class_ids,

movement_modifier and env_profile_status empty (§ 8). Three more (sri_lanka, south_india,

swahili_coast) carry terrain_feature_class_ids and nothing else. So any consumer expecting the

region row to be the env fallback finds nothing there for eight of twelve slice regions; the zone

catalogs on the region pages are the real source. Region-profile lane.

• The 64 non-slice regions each need an RCM row before their weapons resolve off

PENDING_RCM_ROW. The vocabularies are sized for them; the rows follow the region cascade.

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