DEFENSIVE_CLASS_CANON_READ.md

systems/DEFENSIVE_CLASS_CANON_READ.md

defensive_class — THE CANON READ, 72 of 72

Canon parent: T1_Combat_System_Spec (at _source/01_Tier_1_Foundation/) §Block L445-448 ·

§Parry L452-462 · L645 (the column-minting sentence) · L685.

PROPOSED ONLY. THIS DOCUMENT WRITES NOTHING. No registry cell, no fidelity baseline, no canon

file, no harness file was touched by this lane. The column is applied by the DIRECTOR, with the

csv module, against registries/T0_Weapon_Registry [DRAFT v0.1]/Sheet1.csv, with the scoped

docs/fidelity_baseline.json refresh (--emit-baseline) in the same commit, per the Vector-C

landing path already declared at docs/registry_extensions.json defensive-properties.

CANON SUBORDINATION — this document is PROPOSAL-TIER: it serves canon and never outranks it.

Canon here is _source/01_Tier_1_Foundation/T1_Combat_System_Spec [ACTIVE v1.0].md,

docs/spine/CH_03.md, _source/02_Tier_2_Region_Pages/flores_island.md, and the T0 registry rows

themselves. If this document disagrees with any of them, CANON WINS and this document is the defect.

Lane A of the combat-schema wave. Work order: harness/counter_class_coverage_scorecard.md L12 —

*"A1 parry universe UNREACHABLE: T0_Weapon_Registry carries no defensive_class column … Every

parry_only rail row will FAIL as unjudgeable rather than pass."* The schema brief is

docs/proposals/boss_rail_exemplar/BOSS_RAIL_EXEMPLAR_BE_0002.md §7. This document is the

per-weapon canon read that brief's own closing paragraph demands, and it corrects the brief in

two places (§2.2 and §6).

---

0. DERIVATION

DERIVED FROM

• _source/01_Tier_1_Foundation/T1_Combat_System_Spec [ACTIVE v1.0].md — read in full.
• § Block, Parry, Dodge Mechanics L425-472 — the defensive surface. L443-450 § Block

Mechanic (*"Shield-based active defense with Stamina cost"*), L447 the Shield row, L448 the Great

Shield row. L452-462 § Parry Mechanic, L454 *"Weapon-based defensive timing window. Successful

parry generates Stamina or Vril per weapon class. Canonical parry-capable weapons per

pre-restructure source:"*, L456 Rapier, L457 Long Sword, L458 Cutlass, L459 Madu, L460 Sword (1H),

L462 the Madu uniqueness line. L472 § Defensive Layer Integration — block, parry and dodge as

one family of defensive successes on one mitigation scale.

• § Nature-Derived Counter Classes L627-694 — L645 *"Parry-only … Answered by a

parry-capable weapon, and the weapon side of that answer is T0_Weapon_Registry defensive_class"*

(the sentence that mints this column); L651 the seven data tokens; L653-660 the elemental

parry; L662-671 the mitigation scale; L685 *"the parry class enters at Ch 3 and later"*;

L692-694 the registry anchors and the enforcing gate.

• § Sorcery Off-Hand Canonical Lock L226 — *"Neither has direct damage capability. Both cannot

block."* § Wand Canonical Lock L232 — *"Cannot block."* The only three CITED NEGATIVES in the

corpus.

• Group taxonomies L144-155 (Brute, 8 groups) and L204-218 (Trickery/Sorcery, 9 groups) —

read per weapon to place every roster row in its family; L153 OFF-HAND (3): Xiphos, Shield,

Great Shield; L216 OFF-HAND (2): Orb, Tomb.

• § T0_Weapon_Registry Schema L183-196 — the 8-column minimum-viable lock and the

pipeline-driven column-expansion principle this column lands under.

• L43 and L725 — per-weapon detail delegated to the Tier 6 Weapons_Detailed docs (see

finding D6: those docs are one-line stubs, so T1 § Parry / § Block plus the spine and the region

pages ARE the whole per-weapon defensive canon that exists today).

• docs/spine/CH_03.md — L144 *"take the shield first and turn each cracking arc on the clean

parry until the count is kept"*; L145 systems: [boss:BE_0003.P1, object:wpn_033_shield];

L97 BE_0003 *"primary PARRY"*; L258 *"chapter_anchored_weapon_add WPN_033 Shield (BRUTE,

OFF-HAND) roots here — the caci whip-and-shield seating the player's first off-hand Shield,

grounding the main boss's parry (whip-and-shield rhythm) verb"*; L28 the chapter prose

(*"the shielded parry"*).

• _source/02_Tier_2_Region_Pages/flores_island.md § Section 6 — L672-673 *"the caci rounds seat

the player's first off-hand Shield and ground the CH_03 main boss's PARRY verb"*; L690-693 the

legendary seat, *"a legendary defensive weapon whose four-ability pool draws on the caci's

read-and-turn discipline (the clean parry, the kept count, the answering pass)"*.

• registries/T0_Weapon_Registry [DRAFT v0.1]/Sheet1.csv — **all 72 rows, all 32 columns, read

cell-by-cell.** The cells this read relied on per row: weapon_id, weapon_name, weapon_class,

weapon_subclass, damage_type, canonical_chapter_acquisition. Measured facts used below:

damage_type = GUARD on exactly 3 rows (WPN_033, WPN_034, WPN_042); canonical_chapter_acquisition

blank on exactly 4 rows (WPN_015, WPN_016, WPN_022, WPN_037); the earliest acquisition in the whole

roster is WPN_033 Shield at CH_03.

• harness/check_counter_class_coverage.py — L91 `PARRY_DEFENSIVE_CLASSES = {"parry", "catch",

"riposte"}; L573-585 the grant resolver (min(canonical_chapter_acquisition)` over parry-family

rows, and the UNREACHABLE branch when the column is absent); L188-208 the A1 verdicts a resolved

grant chapter produces.

• harness/counter_class_coverage_scorecard.md L12 — the work order, quoted above.
• docs/proposals/boss_rail_exemplar/BOSS_RAIL_EXEMPLAR_BE_0002.md §7 (L361-387) — the schema brief,

read in full, including its closing trap (*"Do not populate defensive_class by inference from

weapon_class or damage_type"*), which §2.3 below executes and evidences.

• docs/pipeline_review/MINT_DESIGN_B_ITEM_ALLY_2026-07-27.md L1204-1222 — the earlier, wider

five-member enum proposal (none · block · parry · catch · riposte), its default = none

justification, and the Madu-as-catch seed. Proposal-tier; used as the vocabulary CANDIDATE that

§2.2 then verifies against T1 canon.

• docs/registry_extensions.json defensive-properties (L161, L164) — the declared Vector-C landing

path: five columns on T0_Weapon_Registry/Sheet1, landed by extending the stub_superseded

candidate_header and re-pinning, with NO added_columns key. The director's ritual, named so the

apply has an address.

• docs/REGISTRY_ROW_STRUCT_SPEC.md §2.1 THE NAMING LAW (L63-92) — new canon columns are always

qualified <qualifier>_class. defensive_class complies as named; no rename needed.

• docs/COMBAT_ENCOUNTER_SYSTEM.md L295 — *"parry_only — a parry-capable weapon, read from

T0_Weapon_Registry.defensive_class"*; docs/fk_spec.json L997 — the same answer-home mapping.

• docs/FUN_REBUILD_PLAN.md L88-95 (R2.2) — *"canon gates it to parry-capable weapons the player does

not hold until Ch 3+ … shield/block is Ch-3 Manggarai"* — the identification of the Ch-3 grant with

the Shield, stated outside T1.

• docs/PILLAR10_REGIONAL_COMBAT_VOCAB.md L172 and its guard_yield row — *"the parry-capable roster

in T1_Combat_System_Spec § Parry Mechanic keeps its defensive_class"*, and guard_yield = *"the

window a guard holds and the resource a successful defence returns"*, warranted on flores_island's

seating of the Shield.

• _source/01_Tier_1_Foundation/T1_Build_Pipeline_Contracts [ACTIVE v1.0].md L2752 —

weapon_sfx_id_ref_array *"covering hit, block, parry, ranged release…"*: the audio contract

already assumed a per-weapon defensive property. Corroboration for the vocabulary's shape.

• docs/proposals/combat/WEAPON_PROFILES.json (subordinate, provenance only) — class_notes.GUARD

*"for receiving — which is a damage class in a game with a parry"*, and census.by_damage_class

GUARD 3. It carries no defensive column and authored nothing here.

• Negative searches run, and reported as negative rather than left silent: the Tier-6 per-weapon docs

(T6_Weapons_Brute_Detailed, T6_Weapons_Trickery_Detailed, T6_Weapons_Master_Index) are

one-line title stubs; no region page or spine node names the Xiphos at all; the Madu is named

twice — docs/spine/CH_07.md L284 (*"won by catch-and-redirect"*) and docs/spine/CH_31.md L271

(*"the Madu parry-weapon"*), both corroborating WPN_042 = catch; the only

defensive-verb hits across the spine nodes and the region pages are CH_03's shield-parry, CH_07

L190/L279 (*"the honorable kalari duel's parry-riposte primary"* — a chapter combat MODE, not a

weapon property), and CH_11 L140 plus west_africa.md L1768 (a demanded parry-riposte ANSWER with

no weapon named).

NOT DERIVED (authored judgment, and why it had no canon home)

1. Which of the three parry-family tokens each canon-named parry weapon takes (§3.2). Canon names

five parry-capable weapons and five distinct mechanics; the token set has three members. Rapier,

Cutlass, Madu and Sword (1H) map by canon's own words (*"Parry ability"*, *"Parry Riposte"*,

*"Horn Catch"*, *"Riposte"*). The Long Sword does not: *"simultaneously attack and defend"* has

no token of its own. The reading is argued at §3.2, the alternative is stated, and **the gate treats

all three tokens identically (L91), so nothing downstream turns on it.**

2. The none default on the 62 rows canon is silent about. Canon states no defensive property for

them and states no negative either. none is the reading MINT_DESIGN_B L1210 defends as

reproducing current canon exactly (today's runtime bParryCapable = false on every weapon), and it

is the only value that does not invent a capability. Each such row carries its own silence as its

citation rather than a borrowed one.

---

1. WHAT THE COLUMN IS, AND WHY THE GATE CANNOT SEE PAST IT

T1_Combat_System_Spec L645 mints it in one sentence: *"Parry-only — the attack cannot be dodged

because it is too sweeping, too indirect, or too large. Answered by a parry-capable weapon, and the

weapon side of that answer is T0_Weapon_Registry defensive_class."*

The consumer is harness/check_counter_class_coverage.py L573-585. It resolves the earliest parry

grant as min(canonical_chapter_acquisition) over rows whose `defensive_class ∈ {parry, catch,

riposte}` (L91), and — finding the column absent — declares the parry universe UNREACHABLE and

FAILs every parry_only rail row as unjudgeable (L196-199). *"A search that cannot match must never

report a clean pass."* That is the whole blocker: the answer universe for one of the seven ruled

response classes does not exist as data.

---

2. THE VALUE VOCABULARY, DERIVED

2.1 The five members, each with its canon warrant

Every one of the five is a word canon uses, in a sentence canon writes, about a mechanic canon owns.

None is minted here.

2.2 CORRECTION TO THE §7 BRIEF: the vocabulary is FIVE members, not four

BOSS_RAIL_EXEMPLAR_BE_0002.md §7 (L376-379) proposes *"parry | catch | riposte | none — the

gate's own three-member parry family plus a non-parry value."* Canon disagrees, and canon wins.

• T1_Combat_System_Spec gives BLOCK its own section (L443-450), parallel and equal to § Parry

(L452-462), with its own resource (Stamina), its own verb (*"hold to maintain"*), its own two named

weapons, and its own formula placement (L450, *"Block reduces incoming damage application before

integrity-band and RNG sigma factors apply"*). A vocabulary that cannot say block cannot record a

mechanic canon spends a section on.

• Canon states three cited negatives that only mean something in a vocabulary containing block:

L226 *"Both cannot block"* (Orb, Tomb) and L232 *"Cannot block"* (Wand). A four-member set turns

three positive canon statements into indistinguishable silence.

• L472 names the defensive family as *"block, parry, dodge"* on one mitigation scale. Block is a

first-class member of the surface this column describes.

• The five-member set is also the one already proposed and defended at

docs/pipeline_review/MINT_DESIGN_B_ITEM_ALLY_2026-07-27.md L1210 (`none · block · parry · catch ·

riposte, default none`). The §7 brief's four-member set is the narrower of two live proposals and

the one canon does not support.

Cost of the correction: zero. block is not in the gate's PARRY_DEFENSIVE_CLASSES (L91), so

adding the member changes no A1 verdict. It only stops the column from lying about the Great Shield.

2.3 THE TRAP, EXECUTED AND EVIDENCED

The §7 brief warns (L385-387): *"Do not populate defensive_class by inference from weapon_class or

damage_type. GUARD, BRUTE and OFF-HAND are three different axes and none of them means 'parries'."*

This lane read the canon per weapon. Here is the measured proof that the shortcut would have been

wrong, computed over the live CSV:

damage_type = GUARD agrees on all three rows it covers, but it is NOT independent evidence:

harness/emit_weapon_profiles.py L61 derives GUARD by name-token from "shield" / "madu", so it fires

because of the row's name, not a separate canon read. It is recorded as agreement, not corroboration —

and it recovers only 43 per cent of the defensive roster regardless. The column is a canon read, and

this is the measurement that says so.

2.4 Naming-law compliance

docs/REGISTRY_ROW_STRUCT_SPEC.md §2.1 (L63-92): *"New canon columns are always qualified —

<qualifier>_class … never a bare C++/UE keyword."* defensive_class is already qualified and is the

name canon itself uses at L645. No rename, no alias, no §4 escape hatch needed.

---

3. THE TWO INTERPRETIVE CALLS, RULED WITH THEIR ALTERNATIVES

3.1 WPN_033 Shield — parry, not block

The problem. T1_Combat_System_Spec files the Shield under § Block (L447: *"Stamina-based

active block, hold to maintain"*) and omits it from § Parry's canonical parry-capable list (L454-460).

Read on T1 alone, the Shield is block — and block is not parry-family, so the earliest parry grant

would resolve to CH_07 (WPN_042 Madu), and every parry_only row authored at Ch 3, 4, 5 or 6

would FAIL A1 as unanswerable.

The canon that decides it is downstream of T1 and more specific than T1.

• docs/spine/CH_03.md L144: *"take the shield first and turn each cracking arc on the **clean

parry** until the count is kept."* The shield is in hand and the action canon names is a parry.

• docs/spine/CH_03.md L258: *"chapter_anchored_weapon_add WPN_033 Shield (BRUTE, OFF-HAND) roots

here … grounding the main boss's parry (whip-and-shield rhythm) verb."* The spine states the

causal link between this weapon row and a PARRY verb explicitly, and cites `T0_Weapon_Registry @

WPN_033` while doing it.

• _source/02_Tier_2_Region_Pages/flores_island.md L672-673: *"the caci rounds seat the player's

first off-hand Shield and ground the CH_03 main boss's PARRY verb."*

• flores_island.md L691-692: the legendary Shield line is *"a legendary defensive weapon

whose four-ability pool draws on the caci's read-and-turn discipline (the clean parry, the kept

count, the answering pass)."*

• T1_Combat_System_Spec L685: *"the parry class enters at Ch 3 and later."* WPN_033 Shield at

CH_03 is the earliest weapon acquisition in the entire 72-row registry (measured). It is the only

weapon that can make L685 true.

• docs/FUN_REBUILD_PLAN.md L94 makes the same identification from the other side:

*"shield/block is Ch-3 Manggarai."*

Options, stated.

• (A) Shield = parry. RECOMMENDED and applied. One column, one value, and the resolver returns

parry_only → 3, which is exactly what L685 says. No harness edit. The Shield's block mechanic is

not lost: it is L447's canonical text, and the four companion Vector-C columns

(parry_window_ms, parry_stamina_cost, parry_generate, riposte_followup_ref) are where the

hold-and-Stamina detail lands. block stays a live member of the vocabulary, held by the Great

Shield.

• **(B) Shield = block, and the director widens PARRY_DEFENSIVE_CLASSES (gate L91) to include

block.** Also lands the grant at CH_03, and is arguably the more literal read of T1 § Block. Costs

a harness edit this lane may not make and the director must remember to make; a column shipped

without the matching gate edit silently moves the parry grant from Ch 3 to Ch 7 — and it silently

arms WPN_034 Great Shield (block, CH_27) as a parry_only answer, which no canon supports — a

second reason (A) is the better landing.

• (C) Shield = block, gate unchanged. The honest-looking option and the destructive one: the

parry grant becomes CH_07, parry_only becomes illegal at Ch 3-6, **BE_0003 (the Ch-3 parry

exemplar this entire wave exists to unblock) cannot author a single parry row**, and spine canon at

CH_03 L144 becomes unreachable in data. Rejected.

The strongest objection to (A), answered. *T1 is Tier-1 authority and it puts the Shield under

Block; the spine is downstream and cannot re-file a T1 mechanic.* Answered on three grounds. First,

the two texts do not conflict: L454's list is prefaced *"Canonical parry-capable weapons per

pre-restructure source"* — an inherited enumeration, not an exhaustive closure — and L472 puts

block and parry on one scale rather than in opposition. Second, the spine and the region page are not

re-filing a mechanic; they state a per-weapon fact (*this weapon, in its own chapter, performs a

parry*), which is precisely the per-weapon grain this column records and precisely the grain T1 L43

delegates away from itself. Third, a reading that makes T1's own L685 unsatisfiable by any weapon in

the roster is the weaker reading of T1. It is reversible in one cell.

A conflicting sibling proposal, found by the critic and rejected here.

docs/proposals/combat/A7_WEAPON_EXEMPLAR_WPN_033.md L491 proposes a multi-valued cell,

parry;cover. REJECTED on two grounds: cover is a counter-class token whose canon home is declared

at T1_Combat_System_Spec L647 as T0_Environment_Grammar_Registry tactical_affordance_class, not

this column; and a multi-valued cell fails the gate's set-membership test

(check_counter_class_coverage.py L576 — "parry;cover" is not in PARRY_DEFENSIVE_CLASSES), which

would silently resolve the parry grant to CH_07 — exactly the drift D2 warns about. **The applied

cell is single-valued parry, no delimiter.**

3.2 WPN_011 Long Sword — riposte, not parry

Canon (L457): *"Long Sword Halbschwerter and Meisterhau — half-swording and master cuts

simultaneously attack and defend."* Canon's own definition of riposte on the neighbouring rows is

*"block then counter"* (L458) and *"parry and counter in one motion"* (L460). A defence with the

counter carried in the same action is the riposte register — tighter than Sword (1H)'s *"one motion"*

rather than looser. Alternative reading: parry, on the ground that riposte implies a sequence

(defend, THEN counter) while the Meisterhau is genuinely simultaneous and canon mints no token for

that. Why it does not matter downstream: parry, catch and riposte are one set to the gate

(L91), so the grant chapter, every A1 verdict and every rail row are identical under either reading.

Recorded here so a later reader meets the reasoning rather than re-deriving it.

---

4. THE 72 ROWS

WPN_001 .. WPN_072, in registry order, every cell cited.

---

5. COUNTS

Parry-family (parry + catch + riposte, the set the gate reads): 6 rows. Non-none: 7 rows.

Canon-named defensive weapons in T1_Combat_System_Spec: 5 parry (L456-460) + 2 block (L447-448) = 7.

The read produces exactly the canon roster and not one row more.

What the resolver returns after apply. Parry-family acquisition chapters: CH_03 (Shield), CH_07

(Madu), CH_21 (Sword 1H), CH_27 (Long Sword), CH_28 (Rapier), CH_33 (Cutlass). min = 3 →

grant_chapter["parry_only"] = 3, matching T1_Combat_System_Spec L685 exactly. The UNREACHABLE

branch at gate L581-585 stops firing, and parry_only rows at CH_03 and later become judged

instead of failed.

---

6. THE WPN_033 VERDICT — the §7 brief's claim, tested

The brief states (L376-379): *"One column, defensive_class, over the gate's own three-member parry

family plus a non-parry value — parry | catch | riposte | none — populated across the 72 rows, **with

WPN_033 the first non-none cell.** That single row makes the resolver return parry_only → 3."*

Verdict: the operative half is CONFIRMED; the literal half is REFUTED as stated.

• CONFIRMED — WPN_033 is the row that makes the resolver return 3. Measured over the CSV, WPN_033

Shield at canonical_chapter_acquisition = CH_03 is the earliest acquisition of ANY weapon in the

whole 72-row roster, defensive or not (next: WPN_039 Dagger CH_04, WPN_041 Sickle 1H CH_05, then the

CH_07 cluster). It is therefore necessarily the earliest parry-family row, and min() over the six

parry-family rows resolves to 3 because of it and only because of it. Remove it and the grant jumps

to CH_07 (Madu), stranding parry_only at Ch 3-6.

• REFUTED — WPN_033 is not the first non-none cell in the file. In registry row order the first

non-none cell is WPN_005 Cutlass (riposte), then WPN_008 Rapier (parry), then WPN_011 Long

Sword (riposte) — three rows before WPN_033 is reached. The brief's sentence is true only if

"first" means *earliest by acquisition chapter*, which is the sense the resolver uses and evidently

the sense intended, but it is not what the sentence says. Anyone applying the column by walking rows

top-down and expecting the first hit at WPN_033 will conclude the read is wrong.

• **REFUTED, second and more consequential — the brief's four-member vocabulary forces WPN_033 to a

parry-family value BY ELIMINATION rather than by a canon read.** With no block member available, a

Shield that canon files under § Block can only be recorded as parry / catch / riposte or as

none; the brief's own trap (*"GUARD … does not mean parries"*) is then unenforceable on the one row

where the question is live. Under the five-member vocabulary the Shield's parry is an **argued

canon read** (§3.1: spine CH_03 L144 and L258, flores_island L672-673 and L691-692) rather than a

default, and the Great Shield can be recorded honestly as block instead of being flattened into the

same value.

---

7. FINDINGS FOR THE DIRECTOR

---

8. WHAT THIS LANE DID NOT DO

No canon was changed: not T1_Combat_System_Spec, not the spine, not a region page, not a registry

row, not the CVD, not a numerical, naming, hard-line or §17 cell. **No registry cell was written, no

docs/fidelity_baseline.json was touched, no gate was run, docs/DOC_MAP.md was not edited, and

nothing was committed.** The two interpretive calls (§3.1, §3.2) are ruled on the record with their

alternatives, their recommendations and their strongest objections; §3.1 is reversible by the director

in one cell and §3.2 changes no gate verdict at all. Nothing here is put to Josh.

The apply, for the director. One column defensive_class appended to

registries/T0_Weapon_Registry [DRAFT v0.1]/Sheet1.csv via the csv module, values per §4 (the

WPN_033 cell single-valued parry, no delimiter), landing by the Vector-C path at

docs/registry_extensions.json defensive-properties L161/L164 (extend the stub_superseded

candidate_header and re-pin; NO added_columns key and NO extension_hashes key), with

docs/fidelity_baseline.json refreshed via --emit-baseline in the same commit and

python harness/run_gates.py exiting 0. Expected gate-28 delta: the *"A1 parry universe UNREACHABLE"*

declared-not-armed note disappears, and parry_answer first granted in chapter reads 3 instead of

UNREACHABLE. NOTE (critic F4): a green fidelity gate on T0_Weapon_Registry is VACUOUS — the

frozen reference xlsx carries ONE column against 32 live (docs/DOC_MAP.md L760), so nothing is

compared. The apply's real evidence is the counter_class_coverage scorecard flipping from UNREACHABLE

to parry_answer first granted in chapter 3, plus a pre/post byte-compare of all 32 existing columns

— not the gate's exit code.

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