@inventarch/compliance 3.1.1 → 3.2.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/LANGUAGE.md CHANGED
@@ -22,7 +22,7 @@ Spec source locators do not load documents. Resource capture and disclosure requ
22
22
  | @convention | governance-system | Declares a convention in the shared governance shape. |
23
23
  | @decision | work-system | Represents a choice that is needed or has been made: the question, options and decider, and once made, the choice and rationale. |
24
24
  | @dimension | taxonomy | Defines a supported record dimension used by conditions and lookup. |
25
- | @distribution | workspace-system | Declares root records from which a distributable closure is selected. |
25
+ | @distribution | workspace-system | Declares root records from which a distributable closure is selected, and optionally the host plugin product it describes. |
26
26
  | @execution-binding | agent-composition-system | Connects a native target to a host-installed entry or operation descriptor. |
27
27
  | @harness | agent-composition-system | Composes profiles and execution bindings in a workspace under a host contract. |
28
28
  | @hook | hook-authoring-system | Represents a host event, tool/path selection and guard message. Registration and executing a guard require a host adapter. |
@@ -40,7 +40,7 @@ Spec source locators do not load documents. Resource capture and disclosure requ
40
40
  | @placement | taxonomy | Defines source authority and reach metadata. |
41
41
  | @plan | work-system | Represents an arrangement of milestones toward an intent; it heads the work hierarchy and has no parent. Plans do not nest. |
42
42
  | @playbook | governance-system | Represents a procedure as phase/primitive cells. Cell delivery is generic; the authored method belongs to its author. |
43
- | @predicate | taxonomy | Defines a directed relationship and its inverse spelling. |
43
+ | @predicate | taxonomy | Defines a directed relationship, its inverse spelling and its present phrase. |
44
44
  | @primitive | taxonomy | Defines a cognitive primitive coordinate; a cell is selected using its phase and primitive. |
45
45
  | @principle | governance-system | Declares a governing rationale in the shared governance shape. |
46
46
  | @run | session-system | Represents a governed run identity, declared phase, status and owner. Declaration does not start execution. |
package/README.md CHANGED
@@ -1,6 +1,6 @@
1
1
  # @inventarch/compliance
2
2
 
3
- [![npm](https://img.shields.io/npm/v/@inventarch/compliance?label=npm)](https://www.npmjs.com/package/@inventarch/compliance) [![Public quality](https://github.com/inventarch/ia/actions/workflows/platform-quality.yml/badge.svg?branch=main)](https://github.com/inventarch/ia/actions/workflows/platform-quality.yml) [![License](https://img.shields.io/badge/license-Apache--2.0-green)](LICENSE) [![Node](https://img.shields.io/badge/node-22.x-brightgreen)](package.json)
3
+ [![npm](https://img.shields.io/npm/v/@inventarch/compliance?label=npm)](https://www.npmjs.com/package/@inventarch/compliance) [![Public quality](https://github.com/inventarch/ia/actions/workflows/platform-quality.yml/badge.svg?branch=main)](https://github.com/inventarch/ia/actions/workflows/platform-quality.yml) [![License](https://img.shields.io/badge/license-Apache--2.0-green)](LICENSE) [![Node](https://img.shields.io/badge/node-%3E%3D22.22.0%20%3C23-brightgreen)](package.json)
4
4
 
5
5
  > Checks over compiled IA records and graphs: schema conformance, system admission, fragments, coverage, selectors, variants and contract adoption with supplied evaluators, assembled into one revision report. It also resolves obligations against an optional evaluator catalog and qualifies evidence receipts.
6
6
  >
@@ -68,14 +68,22 @@ console.log(verified.conformance.profile); // package-v1
68
68
  | Revision report | `evaluate(graph, options?)` returns an immutable report keyed to `graph.revision`: `outcome`, `verdicts`, `findings` and `shadows`. `fail` dominates `not-evaluated`, which dominates `pass`. |
69
69
  | Schema checks | `validateSchema(record, registry, pool)` for bootstrap without a graph and `validateGraphSchema` on a graph: sections, fields, value types and narrowings (`matchesType`, `matchesForm`) and relationship cardinality. |
70
70
  | System admission | `validateSystems(folders, registry, pool)`: one `@system` per folder, words from systems the folder does not require, steward coverage, schema enrolment and declared consent. |
71
- | Graph checks | `validateFragments`, `validateCoverage`, `validateSelectors`, `validateVariants`, `validateConsent`, `validateIdentity`, `validateParse`, and `validateCheck` (built-in or catalog evaluator ids) or `validateCheckStructure` (any one named evaluator). |
71
+ | Graph checks | `validateFragments`, `validateCoverage`, `validateSelectors`, `validateVariants`, `validateConsent`, `validateIdentity`, `validateParse`, and `validateCheck` (built-in or catalog evaluator ids) or `validateCheckStructure` (any one named evaluator). `CHECK_IDS` lists the built-in evaluator ids a `@check` may name; `checkRunner(node)` reads `check.implementation`, else `check.runs`. |
72
72
  | Contract adoption | `validateAdoption` calls the clause evaluators you supply, keyed by contract and requirement. |
73
73
  | Evaluator catalog | `createEvaluatorCatalog`, `selectEvaluator` and `bindEvaluators` register a host's trusted, namespaced evaluators for the id a `@check` names. |
74
74
  | Obligations and receipts | `resolveObligations`, `inspectObligations`, `profileOccurrences`, `validateReceipt`, `acceptEvaluatorOutput`, `adaptBuiltinVerdict` and `admitTransition` decide which obligations apply, whether a receipt is valid and current, and whether a transition may be admitted. |
75
- | Fixtures | `runLanguageFixture` and `fixtureCoverage` run conformance fixtures you read yourself. The package ships its conformance fixtures under `fixtures/`. |
76
- | Conformance evidence | `verifyKernel`, `verifyFixtures` and `isConformanceEvidence` produce and recognise COMP-KERNEL and COMP-FIXTURES evidence; `evaluate`'s `conformance` option binds it to a graph. Evidence about another graph, floor or toolchain stays `not-evaluated`, and evidence the package did not produce is refused. Results from it carry `basis: 'verified'`; results from the caller's own `evidence` map carry `basis: 'caller-asserted'`. |
75
+ | Fixtures | `runLanguageFixture` runs one conformance fixture you read yourself; `fixtureCoverage` checks that passing failing cases observe every required code. The package ships its conformance fixtures under `fixtures/`. |
76
+ | Conformance evidence | `verifyKernel`, `verifyFixtures` and `isConformanceEvidence` produce and recognise COMP-KERNEL and COMP-FIXTURES evidence; `evaluate`'s `conformance` option binds it to a graph. Evidence about another graph, floor or toolchain stays `not-evaluated`, and evidence the package did not produce is refused. A bound result that passed or failed carries `basis: 'verified'`; one left `not-evaluated` carries none. Results from the caller's own `evidence` map carry `basis: 'caller-asserted'`. |
77
77
  | Host plugin | `renderClaudePlugin` returns the files of the user-level Claude Code plugin that `ia host claude --user` installs. |
78
78
 
79
+ `evaluate` takes these options:
80
+
81
+ - `sourceDiagnostics` and `folders`: the language diagnostics and the discovered system folders.
82
+ - `adoptionEvaluators`: clause evaluators keyed by contract identity, then requirement id, with an optional `coordinate` for conditional clauses.
83
+ - `catalog`, and `checks: 'structural'` for the structural COMP-CHECK.
84
+ - `admission`: prior non-pass assessments for subjects no longer in the graph.
85
+ - Either `conformance` or `evidence`, not both. `evidence` maps `COMP-KERNEL` or `COMP-FIXTURES` to a `Verdict` built with `verdict(assessment, graph.revision)`.
86
+
79
87
  Finding codes are `COMP_CODES` and `EVIDENCE_CODES` (`IA-COMP-*`); language and graph findings keep their own codes. The `./internal/*` subpaths are published for IA's own toolchain; use the root entry point.
80
88
 
81
89
  ## ⚖️ What it does not do
@@ -89,7 +97,7 @@ Finding codes are `COMP_CODES` and `EVIDENCE_CODES` (`IA-COMP-*`); language and
89
97
 
90
98
  ## 📚 Learn more
91
99
 
92
- - [SPEC.md](SPEC.md): the package contract, refusal codes, the evaluator catalog, obligation and receipt clauses, and native conformance evidence (C31)
100
+ - [SPEC.md](SPEC.md): the package contract, refusal codes, the evaluator catalog, obligation and receipt clauses, native conformance evidence (C31) and the host plugin renderer (C32)
93
101
  - [LANGUAGE.md](LANGUAGE.md): the packaged summary of the words this release ships
94
102
  - [Architecture reference](../../docs/reference/architecture/README.md#which-package-to-use) and the [host plugin specification](../../docs/specs/host-plugin-distribution/README.md)
95
103
  - Neighbours in the dependency direction: [`@inventarch/language`](../language/README.md) and [`@inventarch/graph`](../graph/README.md) before it; [`@inventarch/db`](../db/README.md) after it
package/SPEC.md CHANGED
@@ -1,8 +1,8 @@
1
1
  # Compliance specification
2
2
 
3
- Evaluator catalog extension: the producer side of trusted evaluator catalog and obligations is implemented by C25–C29 below (the catalog, check binding, obligation resolution including the strict-typing-v1 profile mapping, receipts and admission, with their qualification cases in tests/evaluator-catalog.test.ts). It is explicit and optional: calls without a catalog keep the built-in-only behavior, and unknown check.runs ids still refuse. Still open and owned elsewhere: the API transactional controller and managed transition seam, an app host's fixed-to-catalog consumer, the consumer-side cases (late edit/reevaluation, concurrent workers and durable intent, silent agents) and the consumer side of strict typing (a real compiler evaluator and excluded-source detection). No scheduler, execution, persistence or native word is added here.
3
+ Evaluator catalog extension: the producer side of trusted evaluator catalog and obligations is implemented by C25–C29 below (the catalog, check binding, obligation resolution including the strict-typing-v1 profile mapping, receipts and admission, with their qualification cases in tests/evaluator-catalog.test.ts). It is explicit and optional: calls without a catalog keep the built-in-only behavior, and unknown check.runs ids still refuse. Hosts outside this package own the consumer side: transactional admission and the managed transition, choosing between fixed and catalog evaluators, re-evaluation after a late edit, concurrent workers with durable intent, silent agents, and a real strict-typing compiler evaluator with excluded-source detection. No scheduler, execution, persistence or native word is added here.
4
4
 
5
- Status: implemented through stage C and the D/R/E integration clause C21; the projection clauses C22–C24 are retired (below). Delivery evidence below is chronological; statements of later work within early checkpoints describe that checkpoint only.
5
+ Status: C01–C21 and C25–C32 are implemented; the projection clauses C22–C24 are retired (below). Delivery evidence below is chronological; statements of later work within early checkpoints describe that checkpoint only.
6
6
 
7
7
  ## Instance schemas
8
8
 
@@ -14,9 +14,9 @@ C03. Require each `must have` section. Under closed, diagnose every unlisted sec
14
14
 
15
15
  C04. For each schema field, use its section and exact key. The current dialect admits exactly section.key with a single-token key; nested paths and schema spelling of multiword keys are not added here. Must fields require a present field. An absent optional section creates no subordinate field obligation. A missing required section consumes missing-field reports beneath it. Validate every occurrence, so one valid duplicate does not mask a wrong-typed duplicate. A declared key whose type is not `list of T` is exactly one field: `must have` means exactly one for non-list types and `may have` at most one. A second occurrence of such a key, counted across repeated sections and in the virtual head section, is FIELD-DUPLICATE, reported once per schema row at the second occurrence's span, while every occurrence is still type-checked and narrowed (C04a). Only a `list of T` key may repeat. Undeclared keys are not counted (2026-09-25); under open they remain admitted, under closed they are FIELD-UNKNOWN (C04b). Missing required field: FIELD-MISSING; bad present value: FIELD-TYPE. One failed value, including a bad list member, yields one finding at the field.
16
16
 
17
- C04b. Under `closed`, a declared body section admits only the keys the schema declares for it (2026-09-25). Every other ordinary field is FIELD-UNKNOWN (error) at its own span, `closed @schema <name> does not declare '<section>.<key>'`; the children of an undeclared block fold into its one finding, `- item` lines are not keys, and W0 narrowing still applies to the declared keys beside it. An ordinary field is a compiled field the compiler left as a field. Exempt, because the language reads them itself: the head (virtual, open); the floor grammar sections `relationships`, `cognition` and `activation` (edges, cells, selectors); a field the compiler consumed as a semantic product, decided by span, namely a governance variant clause (`record.variants`; `readVariants` takes every `governance` field except `severity` and `when`) or a contract requirement clause (`record.requirements`, the `REQ-*` lines); and a `section.key` path listed in the kernel `DIMENSION_PATHS` (today `governance.severity`), which the graph reads as a coordinate. Records of the two floor dialect words `schema` and `system` are exempt whole: their body is the schema/system dialect (`sections.must have`, `fields.may have`, `edges.may`, `closed`), and `@schema schema` is closed with no declared field. The two dialect words and the three grammar sections are the only name-based exemptions; every other carrier is recognised by what the compiler emitted for it.
17
+ C04b. Under `closed`, a declared body section admits only the keys the schema declares for it (2026-09-25). Every other ordinary field is FIELD-UNKNOWN (error) at its own span, `closed @schema <name> does not declare '<section>.<key>'`; the children of an undeclared block fold into its one finding, `- item` lines are not keys, and the language L19a narrowing still applies to the declared keys beside it. An ordinary field is a compiled field the compiler left as a field. Exempt, because the language reads them itself: the head (virtual, open); the floor grammar sections `relationships`, `cognition` and `activation` (edges, cells, selectors); a field the compiler consumed as a semantic product, decided by span, namely a governance variant clause (`record.variants`; `readVariants` takes every `governance` field except `severity` and `when`) or a contract requirement clause (`record.requirements`, the `REQ-*` lines); and a `section.key` path listed in the kernel `DIMENSION_PATHS` (today `governance.severity`), which the graph reads as a coordinate. Records of the two floor dialect words `schema` and `system` are exempt whole: their body is the schema/system dialect (`sections.must have`, `fields.may have`, `edges.may`, `closed`), and `@schema schema` is closed with no declared field. The two dialect words and the three grammar sections are the only name-based exemptions; every other carrier is recognised by what the compiler emitted for it.
18
18
 
19
- C04a. A field whose schema row carries a W0 narrowing (language L19a) is checked further once its value matches its type, each narrowing yielding at most one finding per field occurrence, list items included: `values` (from `id in [...]`) requires every scalar or quoted text to be a member, compared exactly, else FIELD-VALUE; `target` (from `ref to <word>`) requires every reference's discriminator to equal the word, else FIELD-REF-TARGET, and otherwise, when the caller supplied a reference lookup, warns FIELD-REF-MISSING for a reference no admitted record declares; `form` (from `text form <form>`) requires every text to match the form, else FIELD-FORM. `validateSchema` looks references up in its pool (the subject included) through `resolveTarget` over the candidates sharing the reference's word and name; `validateGraphSchema` looks them up in the admitted graph. An ambiguous target is not missing: identity ties are graph-owned. A narrowed field with no matching value, an absent optional section, or a value that already failed its type raises no narrowing finding.
19
+ C04a. A field whose schema row carries a narrowing (language L19a) is checked further once its value matches its type, each narrowing yielding at most one finding per field occurrence, list items included: `values` (from `id in [...]`) requires every scalar or quoted text to be a member, compared exactly, else FIELD-VALUE; `target` (from `ref to <word>`) requires every reference's discriminator to equal the word, else FIELD-REF-TARGET, and otherwise, when the caller supplied a reference lookup, warns FIELD-REF-MISSING for a reference no admitted record declares; `form` (from `text form <form>`) requires every text to match the form, else FIELD-FORM. `validateSchema` looks references up in its pool (the subject included) through `resolveTarget` over the candidates sharing the reference's word and name; `validateGraphSchema` looks them up in the admitted graph. An ambiguous target is not missing: identity ties are graph-owned. A narrowed field with no matching value, an absent optional section, or a value that already failed its type raises no narrowing finding.
20
20
 
21
21
  C05. Value semantics (`matchesType`) and text forms (`matchesForm`, C04a): `iso-date` admits exactly `YYYY-MM-DD` where the month is 01–12 and the day exists in that month, leap years by the Gregorian rule; no time, zone or single-digit part. Text admits scalar, string or prose (including empty quoted text); id admits scalar/string matching an ASCII letter followed by letters/digits/hyphens; qname admits scalar/string with exactly four nonempty canonical identity slots (lowercase system/kind/facet/name, closed kind); number admits a finite decimal scalar/string with optional sign, fraction and exponent, no empty/hex/Infinity; ref admits only typed sigil references; flag admits none (a bare key) or scalar true/false. Block is not a flag. `list of T` requires a list and validates every item against T; an empty list is valid because the dialect has no nonempty-list constraint, with one exception (2026-10-12): a stated empty `authority.moves` on a record whose word is `mandate` (the word agent-system registers, which the runtime's mandate resolver reads by the same test), when its schema declares that key as agent-system's does, is FIELD-VALUE, an error at that field, each such occurrence on its own, whose message says `authority.moves` must name at least one kernel move when stated. A stated move list means one or more kernel moves and the runtime's mandate reader refuses an empty one, so admitting it let every mandate-gated command refuse with an unreadable profile; the dialect cannot say "one or more", so the check is keyed to the `mandate` word in schema field validation, as COMP-STEWARD keys a system's steward to agent-system's `agent` word (`src/systems.ts`, `validateSystems`), and db admission (`validateGraphSchema`) and the bootstrap `validateSchema` (`native:check`) both apply it. An absent `authority.moves` stays valid and restricts no move, and every other empty list, `authority.excluded-words []` and `authority.covers []` included, stays valid. The refusal fixtures observe it (boundary `IA-COMP-FIELD-VALUE/mandate-moves-empty`). Type recognition is over compiled values, not lexical matching of IA source.
22
22
 
@@ -26,7 +26,7 @@ C07. Findings use language's location/severity/message shape with IA-COMP codes;
26
26
 
27
27
  ## Delivery boundary and evidence
28
28
 
29
- The schema delivery must cover all C01–C07 branches with native language compilation, plus direct boundary cases where graph admission is the caller's job. Schema validation is the early reusable part of COMP-SCHEMA/COMP-KERNEL. Enrollment, consent, fragments, coverage, adoption, fixtures, steward/bootstrap, scenario execution and a full revision report remain later work and must never be reported as passing by this API.
29
+ The schema delivery must cover all C01–C07 branches with native language compilation, plus direct boundary cases where graph admission is the caller's job. Schema validation is the early reusable part of COMP-SCHEMA/COMP-KERNEL. Enrollment, consent, fragments, coverage, adoption, fixtures, steward/bootstrap and the full revision report are other checks (C08–C21), and scenario execution is outside this package; this API never reports any of them as passing.
30
30
 
31
31
  Schema delivery evidence: 62 compliance tests (native parse/registry/compile instances and scalar boundary tests), 1001 language tests, workspace typecheck/build and a built-package success/refusal consumer pass. The initial nested-path proposal was removed after native dialect probes; no grammar expansion was made.
32
32
 
@@ -42,9 +42,9 @@ C11. Each active registration's word must occur on either side of at least one c
42
42
 
43
43
  C12. Native systems own README/SPEC next to their records. The first five and their record closure must compile and pass instance/admission checks without substitute TypeScript providers. Their stewards and methods are authored roles/procedures; admission does not launch agents, execute scenarios or install a hook. Success/refusal mutation tests must use the real floor/system sources. Native sources are authored, not generated from legacy catalogs.
44
44
 
45
- Native system admission evidence: the API is exported and consumed by pnpm native:check. The five-system closure has 153 records and 170 passing structural assessments. Nineteen native mutation/connection tests cover C08-C12, alongside 62 schema tests and 21 kernel tests; the language suite remains 1001 passing tests. Typecheck/build pass. No scenario engine or host execution is claimed.
45
+ Native system admission evidence: the API is exported and consumed by pnpm native:check. At this checkpoint the five-system closure had 153 records and 170 passing structural assessments. Nineteen native mutation/connection tests covered C08-C12, alongside 62 schema tests and 21 kernel tests; the language suite was 1001 passing tests. Typecheck/build pass. No scenario engine or host execution is claimed.
46
46
 
47
- ## Full graph and revision checks (stage C)
47
+ ## Full graph and revision checks
48
48
 
49
49
  C13. The full suite consumes an immutable Graph and explicit source diagnostics/folder descriptions. Its registry is graph.registry, and graph winners alone supply semantic subjects. A report retains verdicts keyed by check/scope with graph.revision and a stable aggregate outcome (fail dominates not-evaluated, which dominates pass). Language and graph findings retain their original codes, spans and severity; no second wrapper finding repeats the same fault. Findings sort by path, line, code and message. Exposed results are immutable snapshots. Source diagnostics are supplied explicitly, so absence of source evaluation evidence is not a parse pass.
50
50
 
@@ -60,19 +60,21 @@ C18. COMP-ADOPTION concerns each admitted active require relationship from an ad
60
60
 
61
61
  C18 repository consumer: [tools/compliance](../../tools/compliance/SPEC.md#foundation-authoring-adoption) supplies the three foundation-authoring callbacks from actually executed native source scenarios. That adapter checks source revision, adopter and view before using observations, retains missing/stale evidence as not-evaluated, and reports observed violations as failures. This is explicit host registration, not a package default. Database/runtime reports without callbacks remain not-evaluated, and declaration-only COMP-COVERAGE never satisfies COMP-ADOPTION.
62
62
 
63
- C19. COMP-KERNEL and COMP-FIXTURES require caller-supplied matching-revision assessments from actual verification. No supplied result or a stale result produces NOT-EVALUATED. A pure language-fixture runner can consume source and expected products/diagnostics/format companions without reading files; tools discover fixtures and record observed results. Every named refusal must have a tested failing case; unavailable fixture coverage stays explicit. General prose scenario execution remains outside this stage.
63
+ C19. COMP-KERNEL and COMP-FIXTURES require caller-supplied matching-revision assessments from actual verification. No supplied result or a stale result produces NOT-EVALUATED. The pure language-fixture runner `runLanguageFixture(fixture)` consumes source and expected products/diagnostics/format companions without reading files, and `fixtureCoverage(required, results)` judges whether passing failing cases observe every required code; tools discover fixtures and record observed results. Every named refusal must have a tested failing case; unavailable fixture coverage stays explicit. General prose scenario execution is outside this package.
64
64
 
65
65
  C19 note: C31 is the package producer of these assessments. `verifyKernel` and `verifyFixtures` run the verification in-process, and `evaluate`'s `conformance` option binds their evidence to the graph. The caller-asserted `evidence` map remains, and results taken from it are labelled `caller-asserted`.
66
66
 
67
67
  C20. Public checks use IA-COMP-FRAGMENT-MISSING, COVERAGE-MISSING, COVERAGE-KIND, SELECTOR-INVALID, CHECK-UNKNOWN, CHECK-CONFLICT, VARIANT-AMBIGUOUS, ADOPTION-FAILED, NOT-EVALUATED and FIXTURE-MISMATCH in addition to C07/C11. NOT-EVALUATED is warning evidence that changes the outcome; warning consent-empty/shadow evidence does not fail an otherwise completed check. Source parse/graph codes remain namespaced by their owners. Every implementation delivery records full-suite, negative/native and built-consumer evidence here.
68
68
 
69
- Stage C graph checks delivered: graph schema/fragment/coverage, selector/check-id/source/consent/identity checks and immutable assessments. Native 153-node closure passes these checks. Thirteen graph-check tests cover removed cases, conditional coverage, missing targets/fragments, inverse cell fragments, late resolution, source-code preservation, malformed selectors/runners and consent-refusal diagnostic consumption. Full suite: 1453 tests/48files; typecheck/build/kernel/dependency pass. Exact variant ambiguity, adoption and assembled revision reports remain next.
69
+ These delivery paragraphs are historical: their counts and their statements of next work describe their own checkpoint.
70
70
 
71
- Stage C variant ambiguity and adoption delivered: exact finite ambiguity with witness coordinates and explicit structural adoption evaluators. Thirteen new tests include 128 bounded families compared against exhaustive graph-query evaluation, partial/full domination, missing axes, fixed dimensions, a nine-axis family, absent/throwing/failed evaluators and conditional/unresolved adoption. 88 compliance tests; 1466 workspace tests/49files, typecheck/build/kernel/dependencies pass. Native declarations alone intentionally leave adoption not-evaluated. Full revision reports and executable fixture evidence remain next.
71
+ Graph checks delivered: graph schema/fragment/coverage, selector/check-id/source/consent/identity checks and immutable assessments. The native 153-node closure passed these checks. Thirteen graph-check tests cover removed cases, conditional coverage, missing targets/fragments, inverse cell fragments, late resolution, source-code preservation, malformed selectors/runners and consent-refusal diagnostic consumption. Full suite: 1453 tests/48files; typecheck/build/kernel/dependency pass. Exact variant ambiguity, adoption and assembled revision reports came next.
72
72
 
73
- Stage C revision reports and acceptance delivered: evaluate produces immutable revision reports, preserving originating codes, graph dimension refusals and shadows. Provisional compile missing-target warnings are replaced by graph resolution for the same assertion. Reports restrict folder semantic checks to graph winners. The pure fixture runner consumes supplied source/companions; tools/compliance reads the real tree, verifies generated kernel bytes and executes 176 language plus 34 boundary fixtures covering every current language/graph/compliance code. The native report has 485 verdicts: 484 pass and one adoption not-evaluated without clause evaluators. No executed prose scenarios are claimed.
73
+ Variant ambiguity and adoption delivered: exact finite ambiguity with witness coordinates and explicit structural adoption evaluators. Thirteen new tests include 128 bounded families compared against exhaustive graph-query evaluation, partial/full domination, missing axes, fixed dimensions, a nine-axis family, absent/throwing/failed evaluators and conditional/unresolved adoption. 88 compliance tests; 1466 workspace tests/49files, typecheck/build/kernel/dependencies pass. Native declarations alone intentionally leave adoption not-evaluated. Full revision reports and executable fixture evidence came next.
74
74
 
75
- Verified 1472 tests/51files (1241 language,92 graph,92 compliance,47 tools), typecheck/build/native/kernel/dependency checks and built-package runtime/declaration consumer. Compliance stage C is delivered; database admission/publication, runtime packets and specialized execution retain their own owners.
75
+ Revision reports and acceptance delivered: evaluate produces immutable revision reports, preserving originating codes, graph dimension refusals and shadows. Provisional compile missing-target warnings are replaced by graph resolution for the same assertion. Reports restrict folder semantic checks to graph winners. The pure fixture runner consumes supplied source/companions; tools/compliance reads the real tree, verifies generated kernel bytes and executes 176 language plus 34 boundary fixtures covering every current language/graph/compliance code. The native report then had 485 verdicts: 484 pass and one adoption not-evaluated without clause evaluators. No executed prose scenarios are claimed.
76
+
77
+ Verified 1472 tests/51files (1241 language,92 graph,92 compliance,47 tools), typecheck/build/native/kernel/dependency checks and built-package runtime/declaration consumer. C13–C20 are delivered; database admission/publication, runtime packets and specialized execution retain their own owners.
76
78
 
77
79
  C21 (database admission integration). The report assembler may receive prior failed/unavailable admission assessments for subjects subsequently removed from its graph. Retain those observations, coalescing a repeated check/scope at the worst outcome and deduplicating findings. Do not retain earlier candidate-only passes as final evidence. This lets db publish an admitted subset while the compliance report still names refused source subjects. Coverage and adoption are evaluated on the final graph.
78
80
 
@@ -141,3 +143,7 @@ C31 (native conformance evidence). The package produces COMP-KERNEL and COMP-FIX
141
143
  - The inventory gate is `tools/compliance/inventory.ts --check|--write`, and `compliance:check` runs it. Any fixture, companion, corpus or boundary change fails until the inventory is regenerated.
142
144
 
143
145
  C31 delivered: tests/report.test.ts pins the refusals, the binding rules (including three getter attacks on the inputs and one stale object at a time), the labels, the per-check summary and the profile's tamper cases; tools/compliance/check.test.ts pins the inventory gate. Each of 28 spot-checked mutants, each removing one of these rules, was red. No test can fail a check that the evidence's inventory digest is the running one, because branding admits only the producing module instance, so the binding has no such rule. `kernel:check`, `public-language:check` and `vocabulary:check` pass on a fresh copy with no built output. Measured on the built package (Node 22.22.2, Windows), one process: verifyKernel about 30 ms cold and verifyFixtures about 440 ms cold, with about 70 MB heap. Both pass under Node's permission model with read access to the packages only and no write grant.
146
+
147
+ ## Host plugin
148
+
149
+ C32 (host plugin renderer). `renderClaudePlugin(input, platform?)` is the pure renderer that [host plugin distribution §6.1](../../docs/specs/host-plugin-distribution/README.md) specifies: the same input and platform always give the same `{path, text}` files. `platform` is the machine the plugin is installed on and defaults to the running one. It reads and writes nothing; `ia host claude --user --apply` writes the files. For a `platform` other than `win32`, an `input.node` that is not an absolute path, or that the hook command cannot quote, throws an `Error`. `CLAUDE_PLUGIN`, `CLAUDE_MARKETPLACE` and `PLUGIN_MARKER` name the plugin, its marketplace and the marker on generated files. `tests/host-plugin.test.ts` and `tests/host-plugin-node.test.ts` pin the rendering.
@@ -8,7 +8,7 @@ export interface ClaudePluginInput {
8
8
  readonly channel: 'checkout' | 'npm' | 'unknown';
9
9
  /** Absolute path of the CLI entry that rendered this plugin; the hook runs it first (§8 “Session hook”). */
10
10
  readonly entry: string;
11
- /** The update instruction (§11 “Updates and evidence”), which the hook repeats when `ia` cannot be found. */
11
+ /** The update instruction (§5 “Install channel”), which the hook repeats when `ia` cannot be found. */
12
12
  readonly install: string;
13
13
  /**
14
14
  * The absolute path of the Node executable that runs the hook on macOS and Linux: the Node running `ia host claude
@@ -1 +1 @@
1
- {"version":3,"file":"host-plugin.d.ts","sourceRoot":"","sources":["../src/host-plugin.ts"],"names":[],"mappings":"AAMA,eAAO,MAAM,aAAa,OAAO,CAAC;AAClC,eAAO,MAAM,kBAAkB,eAAe,CAAC;AAC/C,eAAO,MAAM,aAAa,uDAAuD,CAAC;AAClF,MAAM,WAAW,iBAAiB;IAChC,iFAAiF;IACjF,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,OAAO,EAAE,UAAU,GAAG,KAAK,GAAG,SAAS,CAAC;IACjD,4GAA4G;IAC5G,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,6GAA6G;IAC7G,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB;;;;OAIG;IACH,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;CACvB;AACD,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;CACvB;AAkMD,8GAA8G;AAC9G,wBAAgB,kBAAkB,CAChC,KAAK,EAAE,iBAAiB,EACxB,QAAQ,GAAE,MAAM,CAAC,QAA2B,GAC3C,SAAS,UAAU,EAAE,CAkDvB"}
1
+ {"version":3,"file":"host-plugin.d.ts","sourceRoot":"","sources":["../src/host-plugin.ts"],"names":[],"mappings":"AAMA,eAAO,MAAM,aAAa,OAAO,CAAC;AAClC,eAAO,MAAM,kBAAkB,eAAe,CAAC;AAC/C,eAAO,MAAM,aAAa,uDAAuD,CAAC;AAClF,MAAM,WAAW,iBAAiB;IAChC,iFAAiF;IACjF,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,OAAO,EAAE,UAAU,GAAG,KAAK,GAAG,SAAS,CAAC;IACjD,4GAA4G;IAC5G,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,uGAAuG;IACvG,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB;;;;OAIG;IACH,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;CACvB;AACD,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;CACvB;AAkMD,8GAA8G;AAC9G,wBAAgB,kBAAkB,CAChC,KAAK,EAAE,iBAAiB,EACxB,QAAQ,GAAE,MAAM,CAAC,QAA2B,GAC3C,SAAS,UAAU,EAAE,CAkDvB"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@inventarch/compliance",
3
- "version": "3.1.1",
3
+ "version": "3.2.1",
4
4
  "type": "module",
5
5
  "main": "./dist/index.js",
6
6
  "types": "./dist/index.d.ts",
@@ -96,8 +96,8 @@
96
96
  "LANGUAGE.md"
97
97
  ],
98
98
  "dependencies": {
99
- "@inventarch/language": "3.1.1",
100
- "@inventarch/graph": "3.1.1"
99
+ "@inventarch/language": "3.2.1",
100
+ "@inventarch/graph": "3.2.1"
101
101
  },
102
102
  "devDependencies": {
103
103
  "@types/node": "^22.19.0",