Add architecture-evaluation and criterion-strength to SpecReview analysis enum

[kreneskyp]

Summary

• Adds architecture-evaluation and criterion-strength to the SpecReview.analysis enum (AnalysisKind in semantic/main.tsp, regenerated schemas/AnalysisKind.json, and the handwritten schemas/spec-review-frontmatter.schema.json), plus the FR-002 spec prose and TC-027's fixture list (tests/test_manifest.py). Logged as FR-002 CR-064 in spec/log.md.
• IR-40 (agent-ix/quire-contract-ir#134): SpecReview SR-539 declared analysis: architecture-evaluation, which this enum didn't admit, so quire validate failed structural validation. This is not a one-document gap — a sweep of spec/reviews/*.md frontmatter across every checkout under ~/dev found 18 documents already declaring this value, across quire-protocol and quire-specification (and their worktrees).
• PLAT-837: the criterion-strength spec-review lens is blocked on exactly this value, per its own acceptance criteria (spec/reviews/criterion-strength.md, analysis: criterion-strength) and its stated blocking text: "a selected analysis absent from the installed SpecReview.analysis schema is an unavailable dependency, not permission to emit an invalid review document."
• SpecReview.json's data_schema.digest in manifest.yaml is unchanged — it only $refs AnalysisKind.json and doesn't inline the enum, confirmed via make manifest-digests (no-op).

Owning repo and how it was determined

SpecReview's frontmatter and semantic schemas are owned by agent-ix/spec-artifacts-process, not by any consumer. Established by reading ~/.ix/filament/registry.json, which pins the installed spec-artifacts-process Filament module to this repo (git-subdir, path spec_artifacts_process) at a specific sha — the installed copy under ~/.ix/filament/modules/spec-artifacts-process is a pinned snapshot of this source, not itself editable durably. This PR edits only the source: spec_artifacts_process/semantic/main.tsp (TypeSpec — the actual source of truth for AnalysisKind/SpecReview.json, regenerated via make schemas) and the separately handwritten spec_artifacts_process/schemas/spec-review-frontmatter.schema.json (excluded from TypeSpec generation by scripts/generate.mjs).

Before/after evidence for IR-40

quire-contract-ir's own origin/main (fe400d8) no longer reproduces IR-40's quoted error today: its SR-539 was re-typed from analysis: architecture-evaluation to analysis: scope-boundary in agent-ix/quire-contract-ir@a5154d3 (landed 2026-09-16), so quire validate on that document now passes. make spec on that repo is still red, but from an unrelated quire coverage --strict gap (16 unbacked matrix rows) — out of scope for this schema PR.

To demonstrate the actual fix (isolated from that repo's unrelated coverage gap, and from a concurrent, unrelated reinstall of the shared ~/.ix/filament module registry that happened mid-session), I re-created IR-40's exact scenario against controlled module copies:

Before — quire validate --module <origin/main spec-artifacts-process, unfixed> against a copy of SR-539 with analysis: architecture-evaluation restored:

.../reviews/2026-09-14-temporal-ecosystem-model-architecture.md: [SpecReview] frontmatter:
  "architecture-evaluation" is not one of ["base","failure-domain","integrity","dependency",
  "evidence","risk-complexity","scope-boundary","gap-analysis","ears-conformance","code-review",
  "spec-correctness"] (at analysis) [frontmatter]
1/1 docs grammar-clean (100%); 0 grammar finding(s): none
1 document(s) failed structural validation

(exact match to IR-40's quoted error)

After — same document, quire validate --module <this branch's spec-artifacts-process>:

1/1 docs grammar-clean (100%); 0 grammar finding(s): none

No document(s) failed structural validation line.

What consumers must do to pick this up

This PR fixes the source only. ~/.ix/filament/modules/spec-artifacts-process (and any repo's installed copy) stays pinned to its old sha until someone runs a module reinstall (quoin module install / quoin plugin install) against a release that includes this commit. quire-contract-ir's red make spec will not turn green from this PR landing alone — even after reinstall, that repo's gate is red for the unrelated coverage reason above, and its own SR-539 document no longer even declares architecture-evaluation (it was retyped to scope-boundary). The 18 documents in quire-protocol / quire-specification that still declare architecture-evaluation will start validating clean once each of those repos reinstalls this module. PLAT-837's Jev lens work (in progress elsewhere, quoin repo) can emit analysis: criterion-strength once its consuming repo reinstalls at a sha including this change.

Additional enum values found in use

A sweep of analysis: frontmatter across every spec/reviews/*.md in ~/dev (repo clones and worktrees) found no third missing value beyond the two named in this ticket. Everything else in use already exists in the enum (base, dependency, failure-domain, ears-conformance, scope-boundary, integrity, risk-complexity, evidence, code-review, gap-analysis, spec-correctness). Broader ungrounded greps also turned up security and adoption-analysis strings, but neither appears as an actual SpecReview frontmatter analysis: value when restricted to spec/reviews/*.md frontmatter — they're incidental matches elsewhere (comments/other fields), not undeclared enum values in use. No ruling needed; nothing added beyond the two named.

Test plan

• make lint (ruff, black, schemas-check — TypeSpec projection matches committed schemas/*.json)
• make test — 296 passed, 1 skipped, 2 xfailed; the same 2 tests fail both before and after this change (test_the_measurement_can_fail, test_tc047_adoption_is_optional), confirmed pre-existing and unrelated (environment/engine-version mismatches unrelated to the analysis enum) via git stash/baseline rerun
• tests/test_manifest.py::test_spec_review_analysis_admits_both_review_families (TC-027) extended and passing with both new values
• Before/after quire validate evidence above

🤖 Generated with Claude Code