colregs 0.3.1 → 0.3.3

package/docs/requirements.md

@@ -53,10 +53,10 @@ Neither consumer lives in this repo.
 | **paragraph path** | The citation unit: `27(a)(i)`, `25(d)(ii)`. Not the rule number. |
 | **fact record** | A set of facts about one vessel at one moment; the input. |
 | **entry** | One applicability record: predicate → lights/refs → modality → citation. |
-| **modality** | `shall` / `may` / `shall-if-practicable`. |
+| **modality** | `modality:shall` / `modality:may` / `modality:shall-if-practicable`. |
 | **jurisdiction** | A body of rules: `intl`, `us/inland`, `ca/inland`, … |
 | **delta** | A jurisdiction's departures from the international text. |
-| **identifier** | Any name the data is addressed by: entry id, paragraph path, light id, fact key, fact value, relation name. |
+| **identifier** | Any name the data is addressed by: entry id, paragraph path, light id, fact key, fact value, relation name, modality/role/encounter/category value, jurisdiction value. |
 
 ---
 
@@ -68,19 +68,20 @@ Neither consumer lives in this repo.
   applicability entry and every rule-text record, expressed as
   `<country-or-body>/<waters>` with `intl` as the reserved base value.
   Examples: `intl`, `us/inland`, `us/great-lakes`, `us/western-rivers`,
-  `ca/inland`, `de/binnen`, `eu/cevni`.
+  `ca/inland`, `de/binnen`, `eu/cevni`. The jurisdiction value is itself an identifier, unprefixed and immutable under REQ-MODEL-10 (ADR 0017).
 - **REQ-SCOPE-3** — A jurisdiction MUST be expressible as a *delta*: entries
   absent from a jurisdiction's data inherit from `intl`. A jurisdiction MUST
   NOT require restating the whole body of rules. Inheritance is "unless
-  suppressed", not unconditional — see Q-11: verified Inland structure
-  (Rule 28 "[Reserved]") means silence-means-inherit would apply
-  international law where the national body deliberately has none, so no
-  non-`intl` jurisdiction lands before an explicit suppression mechanism
-  exists. That bar is about suppression, not licensing: it is independent of
-  REQ-PROV-2 and ADR 0010, and it binds whether a jurisdiction's text ships or
-  is withheld. A delta that only *adds* entries is exempt: it suppresses nothing,
-  so the Q-11 hazard cannot arise from it (ADR 0008, `rule:30a:mooring_buoy`/`rule:30b:mooring_buoy`).
-  A delta that suppresses or replaces an `intl` entry still waits.
+  suppressed", not unconditional: verified Inland structure (Rule 28
+  "[Reserved]") means silence-means-inherit would apply international law
+  where the national body deliberately has none, so a jurisdiction MUST
+  tombstone each inherited entry it deliberately lacks in `suppressions[]`
+  (ADR 0018; Q-11). The delta is an RFC 7396 merge patch over `intl` by
+  entry id — own entries present, tombstones `null`, silence inherits — and
+  a test proves the stored tables equal to that patch applied. This binds
+  whether a jurisdiction's text ships or is withheld (REQ-PROV-2, ADR 0010
+  are separate). A delta that only *adds* entries has no tombstones (ADR
+  0008); one that replaces an `intl` entry tombstones it and adds its own.
 - **REQ-SCOPE-4** — Adding a jurisdiction MUST be additive. It MUST NOT require
   a schema change or edits to existing `intl` entries.
 - **REQ-SCOPE-5** — Geography that gates a rule (Great Lakes, Western Rivers,
@@ -164,14 +165,14 @@ Four layers, each independently addressable.
 - **REQ-MODEL-6** — Entries MUST compose. Multiple entries applying to one fact
   record is the normal case, not an error (Rule 28 is "in addition to" Rule 23).
 - **REQ-MODEL-7** — Five relations MUST be supported:
-  - `rel:includes` — import another entry's **lights only**, never its predicate;
-  - `rel:in_lieu_of` — legal alternatives for the same fact record;
+  - `rel:includes` — import another entry's lights, modality and scalar gates, never its axes (ADR 0019);
+  - `rel:in_lieu_of` — legal alternatives for the same fact record; overlapping targets are alternatives to each other (ADR 0019);
   - `rel:excludes` — mutual exclusion, including across rules; symmetric
     (A excludes B implies B excludes A), and no participating entry MAY be
-    forceful (`shall`/`shall-if-practicable`) — a directed "this one
-    prevails" is `rel:overrides` (ADR 0007), not `rel:excludes`;
-  - `rel:exempts` — one entry lifting another's obligation;
-  - `rel:conditional_includes` — import or alternatives, gated on a predicate.
+    forceful (`modality:shall`/`modality:shall-if-practicable`) — a directed "this one
+    prevails" is `rel:overrides` (ADR 0007), not `rel:excludes`; a constraint on one display, never a removal (ADR 0019);
+  - `rel:exempts` — one entry lifting another's obligation; reaches an entry in force, never an import (ADR 0019);
+  - `rel:conditional_includes` — import or alternatives, gated on a predicate; a `one_of` yields exactly one per display, or none under a `may` carrier (ADR 0019).
   The five are not interchangeable; README.md holds the working semantics.
 - **REQ-MODEL-13** — Where a condition goes is not a style choice. A
   condition on whether a paragraph *applies to this vessel at all* MUST be in
@@ -297,27 +298,27 @@ towards, not because the shape is settled.
   two-subject entries and defaulted for the rest; no record in `rules.json`
   carries one, so a paragraph with no entry is uncategorised)** —
   Every rule paragraph record MUST carry exactly one `category` from the
-  closed set `scope`, `definition`, `standard`, `display`,
-  `classification`, `precedence`, `conduct`, `care`, `meta`. The field
-  defaults to `display`, so existing entries are correct unedited. CI MUST
+  closed set `category:scope`, `category:definition`, `category:standard`, `category:display`,
+  `category:classification`, `category:precedence`, `category:conduct`, `category:care`, `category:meta`. The field
+  defaults to `category:display`, so existing entries are correct unedited. CI MUST
   fail on a value outside the set. Where a paragraph plays a second role,
   that role MUST be expressed as a relation (REQ-CAT-3), never as a second
   category.
-- **REQ-CAT-2** — `care` and `meta` paragraphs (Rules 2(a) and 2(b)) MUST NOT
+- **REQ-CAT-2** — `category:care` and `category:meta` paragraphs (Rules 2(a) and 2(b)) MUST NOT
   be applicability entries. They MUST be recorded in a registry sibling to
   `known_omissions`, stating that the package represents them and evaluates
-  neither. CI MUST fail on a `care` or `meta` paragraph that appears as an
+  neither. CI MUST fail on a `category:care` or `category:meta` paragraph that appears as an
   entry.
-- **REQ-CAT-3** **(unimplemented in part — `shall-not` has no entry yet)** —
-  The modality vocabulary MUST admit `shall-not` and `shall-not-impede`
-  alongside `shall`, `may`, `shall-if-practicable`, `conditional` and
-  `exempt`, and the relation vocabulary MUST admit `rel:overrides` as a
+- **REQ-CAT-3** **(unimplemented in part — `modality:shall-not` has no entry yet)** —
+  The modality vocabulary MUST admit `modality:shall-not` and `modality:shall-not-impede`
+  alongside `modality:shall`, `modality:may`, `modality:shall-if-practicable`, `modality:conditional` and
+  `modality:exempt`, and the relation vocabulary MUST admit `rel:overrides` as a
   sixth verb beside REQ-MODEL-7's five. Both remain closed sets; CI MUST
   fail on a value outside them, and on a cycle in `rel:overrides`.
 - **REQ-CAT-4** — A two-subject rule MUST read a **situation record**:
   two per-vessel fact records, a kinematic state per vessel, relative
   geometry, and history. The per-vessel fact record MUST NOT change to accommodate it, and kinematic
-  state MUST be a distinct fact class — a consumer that reads only `display`
+  state MUST be a distinct fact class — a consumer that reads only `category:display`
   entries MUST NOT be required to supply one. Adding the situation record
   MUST leave every existing fixture valid unedited.
 - **REQ-CAT-5** — A situation MUST NOT be expressible in the current
@@ -328,10 +329,10 @@ towards, not because the shape is settled.
 - **REQ-CAT-6** — The situation record MUST be declared in
   `data/facts.json` under `situation`, and MUST address each vessel's facts
   through the subject namespace of `docs/identifiers.md`:
-  `<subject>:<class>:<key>`, subject from `own`/`other`/`pair`, class from
-  `fact`/`kin`/`geo`/`hist`/`env`. A key with no subject segment MUST mean `own:`,
+  `<subject>:<class>:<key>`, subject from `self`/`other`/`pair`, class from
+  `fact`/`kin`/`geo`/`hist`/`env`. A key with no subject segment MUST mean `self:`,
   so that every predicate and fixture published today is a valid situation
-  predicate unedited. `own:fact:*` and `other:fact:*` MUST resolve to the
+  predicate unedited. `self:fact:*` and `other:fact:*` MUST resolve to the
   per-vessel fact record key for key, with no key renamed or copied.
   `pair` MUST carry only classes whose facts are symmetric between the two
   vessels; `env` — where the encounter is happening — is `pair`-only for that
@@ -350,27 +351,27 @@ towards, not because the shape is settled.
   `expect` MUST be either a bare entry id — the published one-subject form,
   asserting nothing about modality — or `{entry, modality}` naming the
   modality that entry is expected to carry, which is what Q-5 needs. A case
-  whose `status` is `illustrative` MUST assert no entries and MUST NOT join
-  the fixture replay; it fixes the shape and the namespace before the
-  entries exist. CI MUST fail on an undeclared fact, an unresolvable key, an
-  unknown entry id, an unknown modality, and on an `illustrative` case that
-  names an entry.
+  MAY state `roles`, the pooled two-frame answer for both subjects (ADR
+  0016). An `illustrative` case MUST assert no entries and MUST NOT join the
+  replay. CI MUST fail on an undeclared fact, an unresolvable key, an unknown
+  entry id or modality, a stated role the pooled read does not yield, and on
+  an `illustrative` case that names an entry.
 
 - **REQ-CAT-8** — A two-subject entry MUST state an `effect` and MUST NOT
-  state `lights`. For a `precedence` entry the effect MUST be a role per
-  subject, `{own, other}`, from the closed set `give-way`, `stand-on`,
-  `shall-not-impede`, `keep-clear`, `none`; for a `scope` entry it MUST name
+  state `lights`. For a `category:precedence` entry the effect MUST be a role per
+  subject, `{self, other}`, from the closed set `role:give-way`, `role:stand-on`,
+  `role:shall-not-impede`, `role:keep-clear`, `role:none`; for a `category:scope` entry it MUST name
   the part, the section and the rules that section governs; for a
-  `classification` entry it MUST carry exactly one key, either `encounter`
-  from the closed set `head-on`, `crossing`, `overtaking`, `none`, or
-  `risk_of_collision`, whose only value is `true`. `stand-on` MUST
-  appear only as the counterpart of `give-way`, and the counterpart of
-  `shall-not-impede` MUST be `none` — Rule 8(f)(iii) is why. CI MUST fail on a
+  `category:classification` entry it MUST carry exactly one key, either `encounter`
+  from the closed set `encounter:head-on`, `encounter:crossing`, `encounter:overtaking`, `encounter:none`, or
+  `risk_of_collision`, whose only value is `true`. `role:stand-on` MUST
+  appear only as the counterpart of `role:give-way`, and the counterpart of
+  `role:shall-not-impede` MUST be `role:none` — Rule 8(f)(iii) is why. CI MUST fail on a
   role or an encounter outside its set, on an effect whose shape does not match
   its category, and on a `rel:overrides` that resolves to no entry, points at an
   entry of a different category, or closes a cycle. `✎` pencil, with the rest of
   §4.1.
-- **REQ-CAT-9** — The `classification` entries for Rules 13, 14 and 15 MUST
+- **REQ-CAT-9** — The `category:classification` entries for Rules 13, 14 and 15 MUST
   **partition** the pair's relative geometry: for any situation those entries
   all reach, exactly one encounter type MUST apply — never two, never none.
   The crossing sector MUST NOT be enumerated; it MUST be derived, as the
@@ -447,7 +448,7 @@ See ADR 0003.
   from the other, and no property beyond the language of the text — not
   source, audience, nor legal applicability — MUST be inferred from a tag.
 - **REQ-LANG-2** — Identifiers — entry ids, fact values, light ids,
-  paragraph paths, relation names — MUST be language-neutral and MUST NOT
+  paragraph paths, relation names, modality/role/encounter/category values — MUST be language-neutral and MUST NOT
   be localized. Translations attach to identifiers; they never replace them.
   Identifiers are schema keywords, not display strings: each vocabulary
   distinguishes machine identifier, display label (catalog), and definition,
@@ -474,24 +475,21 @@ See ADR 0003.
   resolve to a known paragraph path, and on a corpus filename that disagrees
   with the file's internal metadata. Silence MUST NOT imply coverage (the
   language mirror of REQ-SCOPE-6).
-- **REQ-LANG-6** **(unimplemented — no display catalogs exist)** — Display
-  strings for the identifier vocabularies (light names, fact-value labels,
-  modality labels, image captions) MUST be addressable via stable string keys
-  with per-language catalogs, separate from legal corpora. Catalog entries are
-  static strings: no interpolation, plural or gender grammar — message
-  composition belongs to the consumer's i18n system, and this package MUST NOT
-  grow one. Each catalog MUST carry lightweight provenance: contributors,
-  reviewers, review date, licence. Maintainer notes inside structural files
-  are working documentation, not display strings, not part of the localization
-  surface, and stay untranslated.
-- **REQ-LANG-7** **(unimplemented in part — no fallback policy is encoded, but
-  the mixed-corpus statement is missing from the package documentation)** — The
-  package MUST NOT encode a language fallback policy, and MUST NOT silently
-  substitute one corpus for another. Text is only addressable inside a corpus,
-  so every textual unit a consumer retrieves is attributable to its corpus;
-  package documentation MUST state that a mixed-corpus rendering is never a
-  single authoritative edition. Choice and fallback beyond that are the
-  consumer's (the language mirror of REQ-CONS-3).
+- **REQ-LANG-6** — Display strings for the closed vocabularies the package
+  emits (light, modality, role, encounter, jurisdiction, fact values, and
+  image captions **(unimplemented — no catalog yet)**) MUST be addressable
+  via stable string keys with per-language catalogs, separate from legal
+  corpora; not being an identifier does not exclude a vocabulary. Catalog
+  entries are static strings — no interpolation, plural or gender grammar —
+  and MUST carry provenance (contributors, reviewers, review date, licence);
+  definitions stay untranslated in structural files, not catalogued.
+- **REQ-LANG-7** — The package MUST NOT encode a language fallback policy,
+  and MUST NOT silently substitute one corpus for another. Text is only
+  addressable inside a corpus, so every textual unit a consumer retrieves
+  is attributable to its corpus; package documentation MUST state that a
+  mixed-corpus rendering is never a single authoritative edition. Choice
+  and fallback beyond that are the consumer's (the language mirror of
+  REQ-CONS-3).
 - **REQ-LANG-8** — A `community`-tier corpus MUST record who produced and
   who reviewed it. Machine translation without named human review MUST NOT
   be accepted.
@@ -507,6 +505,9 @@ See ADR 0003.
   reflects. The two MAY differ — a corpus transcribed from an older
   consolidation is legitimate — but the difference MUST be machine-visible,
   never silent.
+- **REQ-LANG-11** — Exactly one language MUST be declared reference-complete;
+  CI MUST fail if its catalog omits an emitted value. Other catalogs MAY be
+  partial, gated only by the existing key-resolution check.
 
 ---
 
@@ -805,7 +806,7 @@ Tracked here until resolved; each becomes an ADR.
   this package: a making-way indicator, and `design.maxSpeed`.
 - **Q-5** — REQ-VERIFY-5 asks for boundary fixtures on every numeric gate.
   Three gates (`rule:23a_ii`, `rule:26b_ii`, `rule:30c`'s `fact:length_m` thresholds) live only in
-  `modality_by`, not in the entry's `when` — they flip `shall` to `may`, not
+  `modality_by`, not in the entry's `when` — they flip `modality:shall` to `modality:may`, not
   which entries apply. The fixture format only asserts applying entry ids, not
   expected modality, so there is no way to fixture these three without
   extending the schema to carry expected modality per entry. Not done
@@ -930,14 +931,13 @@ Tracked here until resolved; each becomes an ADR.
   Rule 28 "[Reserved]" and has no counterpart for `23(d)(ii)`/`(iii)`
   ([verification](verification/2026-08-30-q6-q8.md), Claim 1), so a
   `us/inland` delta that is merely silent there would inherit the `intl`
-  entries and assert international obligations on inland waters. A delta
-  needs explicit suppression records (tombstones) alongside overrides —
-  "this path/entry deliberately does not exist here", distinguishable from
-  "not yet transcribed". Decide the mechanism in the second-jurisdiction
-  bundle (GATE-1 re-take, GATE-2, Q-10); until then no non-`intl`
-  jurisdiction lands. This is the live blocker on CEVNI, and it is *not* the
-  licence one: ADR 0010 removed the licence block by permitting structure with
-  the text withheld, and left this one standing.
+  entries and assert international obligations on inland waters.
+  **Resolved 2026-09-16, ADR 0018.** Tombstones: `suppressions[]` in
+  `data/applicability.json`, one record per (jurisdiction, `intl` entry),
+  and the whole delta is an RFC 7396 merge patch by entry id, proven by
+  test; the two verified Inland absences are the first tombstones. The hold
+  on non-`intl` jurisdictions is lifted (CEVNI: structure unblocked on both
+  counts, this and ADR 0010). GATE-1, Q-10, Q-8 stay open on their triggers.
   Also from the same verification pass, tracked on the global board rather
   than here: four transcription defects in `data/rules.json` itself
   (`21(a)`, `21(b)`, `23(b)`, `29(b)`) — a data fix, not a design
@@ -956,7 +956,7 @@ listed here, one line each, because the ADR is what makes them live. Most are
 - **Q-14** — Which category does each paragraph take? The proposal's table is
   a first cut; settled paragraph by paragraph as Rules 1–19 are transcribed.
   PR #24 settles the first sixteen and departs from the table once: 13(a) is
-  `precedence`, not `classification` (`Q-37`).
+  `category:precedence`, not `category:classification` (`Q-37`).
 - **Q-15** — Which verification tool discharges each category (Alloy, Z3, TLA+,
   STL, Rocq)? Settled by building one proof per category, not by argument.
   *(engine)*
@@ -1004,38 +1004,38 @@ listed here, one line each, because the ADR is what makes them live. Most are
 - **Q-26** — How large is the relative-frame state space? The 10⁴–10⁶ figure
   is a back-of-envelope guess; settled by a worksheet. *(engine)*
 - **Q-27** — What are the field names for `category`, `subjects`, `when`,
-  `effect` and the widened `modality`? Settled when the first non-`display`
+  `effect` and the widened `modality`? Settled when the first non-`category:display`
   entry lands, and cheap to change until then.
   Unchanged by PR #22, which names none of those five. The names it
   does fix — `situation`, the four classes, the three subjects — are
   REQ-CAT-6's and are equally pencil.
 - **Q-28** — What namespace distinguishes the two subjects of a two-subject
-  entry (`own:` / `other:` is the working proposal)? No such segment exists in
+  entry (`self:` / `other:` is the working proposal)? No such segment exists in
   `docs/identifiers.md`; settled before Rule 18 lands, and it is an
   identifier decision under REQ-MODEL-10.
   **Decided in pencil 2026-09-04** (PR #22): `<subject>:<class>:<key>`
-  with subject `own`/`other`/`pair` and class `fact`/`kin`/`geo`/`hist`, and a
-  bare key meaning `own:`. Written up in `docs/identifiers.md` §"Two
+  with subject `self`/`other`/`pair` and class `fact`/`kin`/`geo`/`hist`, and a
+  bare key meaning `self:`. Written up in `docs/identifiers.md` §"Two
   subjects", required by REQ-CAT-6, and exercised by
   `fixtures/situation-fixtures.json`. Three things the working proposal did
   not have: a third subject, because range and in-sight belong to the
   encounter and not to either vessel; a class segment, so kinematics and
   history are new classes rather than new fact keys; and the bare-key default,
   which is what makes the whole thing additive under REQ-MODEL-10 — no
-  existing identifier is renamed or repointed, and `own`/`other`/`pair` become
+  existing identifier is renamed or repointed, and `self`/`other`/`pair` become
   reserved at the head of the identifier space, which is the only cost.
   Settled for good by Rule 18 being written against it.
 - **Q-29** — What are the file names and schemas for the invariants file and
   the region grid, and how does a level-3 invariant carry `jurisdiction`?
   Settled when the first invariant is written down.
-- **Q-30** — What does the `care`/`meta` registry look like as a file — its
+- **Q-30** — What does the `category:care`/`category:meta` registry look like as a file — its
   name, its schema, and its relationship to `known_omissions`? Settled by
   REQ-CAT-2's implementation, which is the next data change after this ADR.
 
   **Decided in pencil, PR #23.** The registry is `represented_paragraphs`,
   a sibling array to `known_omissions` in `data/applicability.json`. Each
   record: `id` (paragraph-derived, e.g. `2a`), `jurisdiction`, `cite`,
-  `category` (`care` or `meta`), and a one-sentence `note`; no `when`, no
+  `category` (`category:care` or `category:meta`), and a one-sentence `note`; no `when`, no
   `lights` — a registry record states that the paragraph is represented
   and evaluated by nothing here, never a predicate. Holds the `2(a)` and
   `2(b)` records this PR adds. `✎` under `docs/conventions.md`: the file
@@ -1049,8 +1049,8 @@ here rather than in a commit message. All pencil.
 
 **Decided in pencil 2026-09-04 (PR #24)**, answering the data half of `Q-27`:
 the field names are `category`, `subjects` and `effect`, and `effect` is
-`{own, other}` roles for a `precedence` entry and `{part, section,
-applies_rules}` for a `scope` one. Written up in `docs/identifiers.md`
+`{self, other}` roles for a `category:precedence` entry and `{part, section,
+applies_rules}` for a `category:scope` one. Written up in `docs/identifiers.md`
 §"Effects", required by `REQ-CAT-8`, and exercised by
 `fixtures/situation-fixtures.json`. `Q-28`'s namespace is settled for good by
 the same PR — Rule 18 is written against it and it held, with one addition:
@@ -1062,13 +1062,13 @@ named that gap before the class existed.
 - **Q-31** — `modality` is a single closed value, and 18(c), 18(d)(i) and 9(a)
   each carry two deontic qualifications at once: a practicability caveat
   ("so far as possible", "if the circumstances of the case admit") *and* the
-  duty itself. 18(c) is expressible because `shall-if-practicable` already
+  duty itself. 18(c) is expressible because `modality:shall-if-practicable` already
   exists; 18(d)(i) is not, and its caveat is dropped with a `gap` note on the
   entry. Settled by deciding whether practicability is a second field beside
   `modality` rather than a value inside it — which is the same question
   `modality_by` answers for the light rules, and should probably be answered
   the same way. **Narrowed, not settled, 2026-09-04 (PR #25):** 18(d)(i) is now
-  `shall-if-practicable` with `effect.own: shall-not-impede`, so the two
+  `modality:shall-if-practicable` with `effect.self: role:shall-not-impede`, so the two
   qualifications sit in two fields and neither is dropped. That works only
   because this duty happens to be a *role*, which `effect` already carries;
   9(a)'s "if the circumstances of the case admit" has no second field to move
@@ -1086,7 +1086,7 @@ named that gap before the class existed.
   `rule18_class:fishing` by 3(d), and the 27(c) tow that severely restricts the
   pair — the one the old predicates could not catch at all — to
   `rule18_class:ram`, via a new boolean `fact:tow_restricts_deviation`. Every
-  Rule 18, 9 and 10 entry reads the class; no `precedence` entry reads
+  Rule 18, 9 and 10 entry reads the class; no `category:precedence` entry reads
   `fact:activity` any more, and a test enforces that. `fact:activity` itself is
   unchanged and no entry that reads it was touched. What is *not* settled: the
   eighth rank Rule 18 distinguishes is the seaplane of 3(e), and it has no
@@ -1116,7 +1116,7 @@ named that gap before the class existed.
   alternative was declined: it would have produced a predicate no reader could
   check against the rule text, to avoid a language feature the rules themselves
   use in plain words.
-- **Q-34** — A `shall-not-impede` entry names a duty toward `other`, but the
+- **Q-34** — A `modality:shall-not-impede` entry names a duty toward `other`, but the
   paragraphs identify the protected vessel by a property this package does not
   carry: "a vessel which can safely navigate only within a narrow channel"
   (9(b), 9(d)), "any vessel following a traffic lane" (10(i), 10(j)), and a
@@ -1142,37 +1142,37 @@ named that gap before the class existed.
   predicate reads only facts, so entry `rule:8f_iii` reads the risk-of-collision half
   alone and applies to every pair with risk of collision. Settled by deciding
   whether a norm may read another norm's effect, which is the same question a
-  `conduct` monitor will ask about role and phase.
+  `category:conduct` monitor will ask about role and phase.
 - **Q-36** — Rule 18(a)(iv) and 18(d)(i) are both in force between a sailing
   vessel and a vessel constrained by her draught: the CBD vessel gives way, and
   the sailing vessel must not impede her. Neither overrides the other and the
-  model records both, so one subject holds `stand-on` and `shall-not-impede`
+  model records both, so one subject holds `role:stand-on` and `role:shall-not-impede`
   at once. The test suite pins it as a finding rather than asserting it away.
   Settled by reading 8(f)(ii) against the cases, not by picking a role.
-- **Q-37** — 13(a) is `classification` in ADR 0005 §1 and in the proposal's
+- **Q-37** — 13(a) is `category:classification` in ADR 0005 §1 and in the proposal's
   first-cut table, but it is the one paragraph of Rule 13 that assigns a role
-  and a `classification` entry has nowhere to put one. Entry `rule:13a` is
-  `precedence` here, with 13(b)'s sector test left for the `classification`
-  entry that would set the `hist:was_overtaking` latch. Settled with the rest
+  and a `category:classification` entry has nowhere to put one. Entry `rule:13a` is
+  `category:precedence` here, with 13(b)'s sector test left for the `category:classification`
+  entry that would set the `hist:was_overtaking` overtaking-history fact. Settled with the rest
   of `Q-14`, paragraph by paragraph.
 
   **Closed in pencil 2026-09-04 (PR #26).** 13(b) is now written, as *two*
-  `classification` entries rather than one — one per subject — because the
+  `category:classification` entries rather than one — one per subject — because the
   encounter type belongs to the pair and reading aspect alone classified the
-  overtaken vessel's side of the same encounter as a crossing. 13(a) stays `precedence`, and its predicate now carries both
+  overtaken vessel's side of the same encounter as a crossing. 13(a) stays `category:precedence`, and its predicate now carries both
   halves of "any vessel overtaking any other": the 13(b) sector and the 13(d)
-  latch, as an `any_of`, which closes the gap the entry recorded. 13(d) is a
-  third `classification` entry, reading history and no geometry at all. The
+  overtaking history, as an `any_of`, which closes the gap the entry recorded. 13(d) is a
+  third `category:classification` entry, reading history and no geometry at all. The
   split confirms the thing ADR 0005 §1 was really defending — one category per
   paragraph. **Reversed in part, 2026-09-16:** ADR 0015 as rewritten makes
   13(b) one symmetric entry, `rule:13b`, reading the same sector object on
-  either subject under `any_of`. 13(a) staying `precedence` is untouched.
+  either subject under `any_of`. 13(a) staying `category:precedence` is untouched.
 - **Q-38** — 9(d), 18(d)(ii), 18(e), 18(f)(ii), 1(a)–(e) and 20(b)–(c) are
   recorded in `known_omissions` rather than modelled: 9(d) needs the channel's
   axis, 18(e) needs a fact for being a seaplane, 20(b)–(c) need time of day,
   and 1(a)–(e) are addressed to an authority rather than to a vessel. Settled
   one at a time as the facts they need land; none blocks the rest of Part B.
-- **Q-39** — Rule 4's `scope` entry has an empty `when`, because "any condition
+- **Q-39** — Rule 4's `category:scope` entry has an empty `when`, because "any condition
   of visibility" is the absence of a condition. That makes it unfalsifiable —
   `REQ-VERIFY-3`'s "excluded by at least one fixture" half cannot be satisfied
   and the test exempts it explicitly. Settled by deciding whether an ungated
@@ -1181,17 +1181,17 @@ named that gap before the class existed.
 
 ### From the classification norms (PR #26)
 
-Rules 7(d), 12, 13(b)–(d), 14 and 15 are the first `classification` entries and
-the second family of `precedence` ones. All pencil. What the model expressed is
+Rules 7(d), 12, 13(b)–(d), 14 and 15 are the first `category:classification` entries and
+the second family of `category:precedence` ones. All pencil. What the model expressed is
 written up in `docs/identifiers.md` §"Effects"; what it could not is here.
 
-- **Q-40** — **Rule 12 is `precedence`, not `classification`.** ADR 0005 §1 and
-  the proposal's table both file it under `classification`; it produces a role
+- **Q-40** — **Rule 12 is `category:precedence`, not `category:classification`.** ADR 0005 §1 and
+  the proposal's table both file it under `category:classification`; it produces a role
   — "one of them shall keep out of the way of the other" — and a
   classification effect has nowhere to put one, which is `Q-37`'s argument
   applied a second time. Two sailing vessels are still in a head-on, a crossing
   or an overtaking; Rule 12 says which of them gives way, not what kind of
-  meeting it is. 12(b) is a `definition` and is the cite on `kin:wind_side`
+  meeting it is. 12(b) is a `category:definition` and is the cite on `kin:wind_side`
   rather than an entry. The second half of the question is narrower and worse:
   12(a) says "two sailing vessels", which is 3(c), but entries the three Rule 12 entries
   read `rule18_class:sail` instead, so a vessel engaged in fishing under sail
@@ -1242,7 +1242,7 @@ written up in `docs/identifiers.md` §"Effects"; what it could not is here.
   again: on those unordered pairs Rule 15 lays a give-way duty on a vessel
   that may be unable to discharge it, which is Rule 2's region and is
   recorded, not gated. `rule:13a` needs no override against `rule:15a:keep_out_of_the_way` — the
-  entry excludes every overtaking by the `hist:was_overtaking` latch and by
+  entry excludes every overtaking by the `hist:was_overtaking` overtaking history and by
   13(b)'s sector — and a test pins that absence so it stays a reason rather
   than an oversight. The hand-list of six is asserted complete by a derived
   check: every Rule 18 entry that assigns a helm role and does not gate a
@@ -1255,7 +1255,7 @@ written up in `docs/identifiers.md` §"Effects"; what it could not is here.
   which is still how the fixtures are written and is what makes these six
   overrides observable rather than invisible.
 
-  The category half — Rule 12 as `precedence` — is unchanged and stays with
+  The category half — Rule 12 as `category:precedence` — is unchanged and stays with
   `Q-14`.
 - **Q-41** — **13(c) and 14(c) invert the absent-fact rule.** "When a vessel is
   in any doubt as to whether she is overtaking, she shall assume that this is
@@ -1284,7 +1284,7 @@ written up in `docs/identifiers.md` §"Effects"; what it could not is here.
 - **Q-43** — **The partition needs history to be present, and says nothing when
   it is absent.** `rule:14b` and `rule:15a:crossing` are gated on `hist:was_overtaking`
   being `false` on both subjects, which is 13(d)'s requirement and the only way
-  the three encounter types stay disjoint once the latch is set. Because absent
+  the three encounter types stay disjoint once the overtaking history is set. Because absent
   is absent, a situation that omits the fact is classified as *no encounter at
   all* rather than as a head-on or a crossing. That is conservative in the
   right direction and completely silent, which is the wrong way to be
@@ -1316,14 +1316,14 @@ written up in `docs/identifiers.md` §"Effects"; what it could not is here.
   the same question `Q-39` asks about an ungated norm, from the other end.
 
   **Decided in pencil 2026-09-05 (PR #36): the head-on classification cites
-  14(b), and 14(a) is `conduct`.** The question assumed 14(b) produced no
+  14(b), and 14(a) is `category:conduct`.** The question assumed 14(b) produced no
   effect of its own. It produces the only effect the entry has: "such a
   situation shall be deemed to exist when…" is the deeming test, and
-  `{"encounter": "head-on"}` is what a deeming test yields — exactly as
+  `{"encounter": "encounter:head-on"}` is what a deeming test yields — exactly as
   `rule:13b` cites 13(b), the deeming paragraph,
   and not 13(a). 14(a) states the situation in words ("reciprocal or nearly
   reciprocal courses") and imposes the duty ("each shall alter her course to
-  starboard"); the duty is `conduct`, and the words are what 14(b) makes
+  starboard"); the duty is `category:conduct`, and the words are what 14(b) makes
   checkable. So the entry is `rule:14b`, cites 14(b), and reads nothing 14(b) does
   not fix; 14(a) joins `known_omissions` as conduct beside 18(d)(ii), and
   the `14a` id it displaced is gone with the rest of the citation-derived ids
@@ -1333,7 +1333,7 @@ written up in `docs/identifiers.md` §"Effects"; what it could not is here.
   instance of it.
 - **Q-46** — **"Coming up with" is a comparison of two facts.** 13(b) deems a
   vessel to be overtaking when *coming up with* another from more than 22.5°
-  abaft her beam. That is `own:kin:sog_kn` against `other:kin:sog_kn`, and the
+  abaft her beam. That is `self:kin:sog_kn` against `other:kin:sog_kn`, and the
   predicate language compares a fact to a constant and never one fact to
   another — the same wall `rule:12a_ii` hits on "both have the wind on the same side",
   which it gets over only because that comparison has two values to enumerate
@@ -1344,27 +1344,27 @@ written up in `docs/identifiers.md` §"Effects"; what it could not is here.
   in seamanship and is one here. Settled by deciding whether the language gains
   fact-to-fact comparison — a bigger change than `not` and `any_of` were, and
   not one to make for a single paragraph.
-- **Q-47** — **13(d)'s latch never clears.** The paragraph runs "until she is
+- **Q-47** — **13(d)'s overtaking history never clears.** The paragraph runs "until she is
   finally past and clear", nothing in the fact vocabulary carries that, and so
   in this model `hist:was_overtaking` is set by a consumer and cleared by a
   consumer while entry `rule:13d` classifies an overtaking for as long as it stands.
   Not approximated with a range or a bearing: past and clear is a seamanship
   judgement of the same kind as risk of collision, and a threshold for it would
   be a number invented rather than declared. Settled by whatever settles the
-  `conduct` monitors, which are the things that watch a duty end rather than
+  `category:conduct` monitors, which are the things that watch a duty end rather than
   begin.
 
-  **Ruled 2026-09-08:** while one vessel holds the latch, 13(a)'s sector
-  test is suppressed on the other, newly-gaining vessel — once overtaking,
-  always overtaking, per 13(d)'s own text, so a geometry test that would
-  newly name a second give-way vessel must not fire while the first still
-  holds the role by history alone. `rule:13a`'s sector branch now also reads
-  `other:hist:was_overtaking: false`; the closing fixture in
+  **Ruled 2026-09-08:** while one vessel's overtaking history stands, 13(a)'s
+  sector test is suppressed on the other, newly-gaining vessel — once
+  overtaking, always overtaking, per 13(d)'s own text, so a geometry test
+  that would newly name a second give-way vessel must not fire while the
+  first still holds the role by history alone. `rule:13a`'s sector branch now
+  also reads `other:hist:was_overtaking: false`; the closing fixture in
   `fixtures/situation-fixtures.json` is the case this closes — a vessel
-  drops back onto the latch-holder's own stern, and only the latch-holder
+  drops back onto the history-holder's own stern, and only the history-holder
   gives way.
 - **Q-48** — **Nothing checks that a situation is geometrically possible.**
-  `own:geo:rel_bearing_deg`, `other:geo:rel_bearing_deg`, the two
+  `self:geo:rel_bearing_deg`, `other:geo:rel_bearing_deg`, the two
   `kin:heading_deg` and the two `kin:sog_kn` are six facts related by two
   equations, and the situation record enforces neither. A consumer — or a
   fixture — can state a pair of bearings that no two headings produce, and the
@@ -1381,7 +1381,7 @@ written up in `docs/identifiers.md` §"Effects"; what it could not is here.
   **Decided in pencil 2026-09-05: the record gains a consistency check,
   declared in data and enforced in the suite.** `data/facts.json` declares the
   equations under `situation.geometry.consistency` — the two relative bearings
-  are two readings of one line of sight (`own + own heading + 180 ≡ aspect +
+  are two readings of one line of sight (`self + self heading + 180 ≡ aspect +
   other heading`), positions reproduce range and both bearings, and the two
   headings and speeds reproduce CPA, TCPA and bearing rate — with tolerances,
   and `REQ-VERIFY-8` requires the suite to apply them. A quantity a record does
@@ -1392,13 +1392,13 @@ written up in `docs/identifiers.md` §"Effects"; what it could not is here.
   bearing-rate values were placeholders. They were re-derived from the headings
   and speeds. Two cases could not keep their story: the R1 head-on had both
   vessels with the other to starboard, which no speeds make a steady bearing,
-  and is recast as one; the 13(d) latch case is 400 m abeam with the CPA
+  and is recast as one; the 13(d) overtaking-history case is 400 m abeam with the CPA
   already past, and lost `rule:7d_i`. The partition sweep now constructs positions
   and headings for every bearing pair and is consistent at every point. The
   "never both give-way" property is asserted over a sweep of steady-bearing
   geometries — every relative bearing, several speed pairs, both intercept
   solutions — where it is a theorem rather than an observation:
-  `u·sin(own bearing) = −v·sin(aspect)`, so the two bearings lie on opposite
+  `u·sin(self bearing) = −v·sin(aspect)`, so the two bearings lie on opposite
   sides and 15(a) can name only one vessel. The both-starboard counterexample
   is pinned as a record the check rejects. What is *not* settled, and is
   pinned too: the theorem is exact only at a bearing rate of zero, and
@@ -1412,7 +1412,7 @@ written up in `docs/identifiers.md` §"Effects"; what it could not is here.
 
 - **Q-49** — **Must `rule:13d`'s effect cross the Section II/III boundary for Rule 19(d)(i) to be expressible?**
   - *broad* — drop `rule:13d`'s `pair:geo:in_sight` gate so `encounter: overtaking` is visible in Section III; breaks the scope invariant.
-  - *narrow* — `rule:13d` stays gated; a 19(d)(i) entry reads `own:hist:was_overtaking`/`other:hist:was_overtaking` directly, as `rule:14b` and `rule:15a:crossing` already do.
+  - *narrow* — `rule:13d` stays gated; a 19(d)(i) entry reads `self:hist:was_overtaking`/`other:hist:was_overtaking` directly, as `rule:14b` and `rule:15a:crossing` already do.
   - Default: narrow (nothing changes).
   - Recommendation: narrow — the test named below shows the fact resolves out of sight
     (`Q-49: hist:was_overtaking resolves out of sight; rule:13d does not fire`, `test/data.test.mjs`).
@@ -1430,18 +1430,18 @@ nothing is blocked while open.
   give-way and Rule 16 binds her. *Narrow:* Section II's duties attach only with
   risk of collision; 14 and 15 are emphasis. Default **wide**: `rule:13a`, `rule:13b`,
   `rule:13d` and every Rule 18 entry read `in_sight` only. No recommendation.
-- **Q-51** — **What arms 13(d)'s latch?** *A:* 13(b)'s deeming, at the first
+- **Q-51** — **What arms 13(d)'s overtaking history?** *A:* 13(b)'s deeming, at the first
   state the geometry holds. *B:* 13(a)'s duty actually attaching. They differ
   where the geometry holds but a condition on 13(a) does not — `Q-50`'s
-  surface. Pencilled **B** ✎ (Solace, 2026-09-14): the latch arms when the
+  surface. Pencilled **B** ✎ (Solace, 2026-09-14): the overtaking history arms when the
   duty attaches. KIVELI [2025] EWHC 1185 (Admlty) holds the analogous Rule 14
-  latch arms on risk of collision and then persists; no authority was found on
+  history arms on risk of collision and then persists; no authority was found on
   overtaking geometry with no risk of collision. Free to reverse: four entries
   (`rule:13a`, `rule:13b`, `rule:13d`) and their fixtures, and nothing consumes the one
   state this changes the answer for. Settled by: a decision on Rule 13 itself,
   or Cockcroft & Lameijer on whether overtaking status needs risk of
   collision.
-- **Q-52** — **What does 13(d)'s latch forbid?** *Narrow:* reclassification to
+- **Q-52** — **What does 13(d)'s overtaking history forbid?** *Narrow:* reclassification to
   *crossing* only, as the paragraph says, leaving head-on to Rule 14 on the
   geometry of the moment. *Broad:* the encounter stays an overtaking and no
   other Section II classification attaches. Both preserve the duty; they differ
@@ -1459,16 +1459,16 @@ nothing is blocked while open.
   flags any other alteration. No default (Rule 17 has no entry). Recommend
   **exception**: 17(a)(ii) describes an action, not a proviso, and leaves a
   monitor something to check — a drafting argument, not a source.
-- **Q-54** — **Are Rule 17's phases monotone?** *Latching:* one three-valued
-  monotone latch per stand-on vessel per encounter; late compliance does not
-  take the permission away. *Re-evaluating:* each phase is a predicate on the
-  current state and the vessel may fall back — and must she then hold her *new*
-  course and speed? TLC distinguishes them on a four-state trace, so settle
-  before P4.2. No default. Recommend **latching**: one late alteration does not
-  restore reliance — an argument from purpose, not a source.
+- **Q-54** — **Are Rule 17's phases monotone?** *Monotone:* one three-valued
+  marker per stand-on vessel per encounter; late compliance does not take the
+  permission away. *Re-evaluating:* each phase is a predicate on the current
+  state and the vessel may fall back — must she then hold her *new* course
+  and speed? TLC distinguishes them on a four-state trace. **Ruled
+  2026-09-16:** monotone — permission or duty, once arisen, survives belated
+  compliance; a fall-back needs a baseline the Rules lack (`Q-57`). No holding on point (colregs#72); *Ever Smart* [2021] UKSC 6 para 61 has the obligation cease at 17(a)(ii)/17(b) with no revival — consistent with monotone, not a ruling on it.
 - **Q-55** — **What does a visibility transition do to Section II state?**
   Rule 11 and 19(a) switch on the current state; no paragraph says what
-  becomes of a 13(d) latch or a Rule 17 phase. *Persisting:* they belong to the
+  becomes of a 13(d) overtaking history or a Rule 17 phase. *Persisting:* they belong to the
   encounter and survive the fog. *Resetting:* Section II starts afresh on the
   geometry when sight is regained — the reclassification 13(d) forbids, via
   visibility. No default (nothing is temporal). Recommend **persisting**, which