=== RESEARCH ===
{'issue': 284, 'kind': 'docs', 'defect_status': 'confirmed', 'category': 'prose-fix', 'target_files': ['docs/src/structural_identification.md', 'docs/src/innovation_accounting.md', 'CLAUDE.md'], 'summary': 'Replace stale "18 [additional] methods" statistical-identification count with the source-true 13 (5 ICA + 4 ML + 4 heteroskedasticity) on two doc pages plus two CLAUDE.md lines; nongaussian.md already correct.', 'current_state': 'Defect is STILL PRESENT (not fixed). Stale "18" occurrences at current tip (release/v0.6.6, cb4f048): docs/src/structural_identification.md:12 "(18 additional methods)"; docs/src/structural_identification.md:325 table cell "18 methods via `compute_Q`"; docs/src/innovation_accounting.md:91 "(18 additional methods)"; CLAUDE.md:23 "Statistical Identification (18 methods)"; CLAUDE.md:254 "18 methods via `compute_Q()`". nongaussian.md is ALREADY correct — states 13 with 5+4+4 breakdown at lines 5, 96, 118, so it needs NO change (issue instruction to standardize it is a no-op). Source truth re-derived from src/core/identification.jl:254-295 (18 compute_Q branches = 4 structural + 1 ML dispatcher :nongaussian_ml + 13 statistical); src/nongaussian/ml.jl:470 confirms identify_nongaussian_ml is the "unified dispatcher" excluded from the 4-ML count. Audit report anchor: docs/plans/2026-07-03-documentation-audit-report.md §4.5 lines 300-303 (HIGH).', 'proposed_change': 'Make these exact edits. (1) docs/src/structural_identification.md:12 — change "(18 additional methods)" to "(13 additional methods: 5 ICA + 4 ML + 4 heteroskedasticity)". (2) docs/src/structural_identification.md:325 — change table cell text "18 methods via `compute_Q`" to "13 methods via `compute_Q` (5 ICA + 4 ML + 4 heteroskedasticity)". (3) docs/src/innovation_accounting.md:91 — change "(18 additional methods)" to "(13 additional methods: 5 ICA + 4 ML + 4 heteroskedasticity)". (4) CLAUDE.md:23 — change "Statistical Identification (18 methods)" to "Statistical Identification (13 methods)". (5) CLAUDE.md:254 — change "18 methods via `compute_Q()`." to "13 methods via `compute_Q()`." (this line is in the Lewis-2025 statistical-identification bullet, so it means the statistical count = 13; the issue body only names line 23 but line 254 carries the same stale statistical count and must move with it to satisfy the "no page/doc still claims 18 for the statistical surface" acceptance criterion). Do NOT touch nongaussian.md (already 13). Do NOT alter any factual statement that legitimately refers to the 4 structural identifications or the 18 total compute_Q branches — none of the five targeted strings do; all five denote the statistical-identification surface. Count is source-true per src/core/identification.jl:254-295 and src/nongaussian/{ica.jl,ml.jl,heteroskedastic.jl}: ICA={fastica,jade,sobi,dcov,hsic}=5, ML={student_t,mixture_normal,pml,skew_normal}=4 (nongaussian_ml is the dispatcher), hetero={markov_switching,garch,smooth_transition,external_volatility}=4.', 'verify': 'julia --project=docs docs/verify_examples.jl docs/src/structural_identification.md && julia --project=docs docs/verify_examples.jl docs/src/innovation_accounting.md (nongaussian.md unchanged but can be run too). All edits are in prose/table text, not @example/@setup code, so a clean build with no output diff is the pass condition. CLAUDE.md is not part of the docs build — verify by grep: `grep -rn "18" CLAUDE.md docs/src/structural_identification.md docs/src/innovation_accounting.md | grep -iE "method|compute_Q"` returns no statistical-identification hits.', 'dependencies': [], 'collisions': ['docs/src/structural_identification.md — #290 (T191 identification/IA/nowcast MED sweep: back-link architecture, duplicated table, missing bib — edits same pages)', 'docs/src/innovation_accounting.md — #290 (T191 same sweep, plus ia_fevd mislabel/setup reductions on this page)', 'docs/src/structural_identification.md — #293 (T194 LOW polish sweep over identification+IA pages)', 'docs/src/innovation_accounting.md — #293 (T194 LOW polish sweep)', 'docs/src/nongaussian.md — #294 (T195 hub/overview docrule skeleton retrofit; issue #284 no longer edits this file so low-risk, but the count text lives here)', 'docs/src/nongaussian.md — #303 (T204 proposes renaming nongaussian.md→id_overview.md)', 'CLAUDE.md — #278 (T179 config drift between docrule.md/CLAUDE.md and actual build)', 'CLAUDE.md — #281 (T182 refresh stale counts, index.md-focused but may touch CLAUDE.md counts)'], 'effort': 'S', 'risk': 'low'}

=== VERIFY ===
{'issue': 284, 'verdict': 'CONFIRMED', 'assessment': 'Defect is real and STILL PRESENT at working-tree tip (release/v0.6.6, cb4f048). All five stale "18" strings independently confirmed at the exact cited lines: structural_identification.md:12 "(18 additional methods)", structural_identification.md:325 table cell "18 methods via `compute_Q`", innovation_accounting.md:91 "(18 additional methods)", CLAUDE.md:23 "Statistical Identification (18 methods)", CLAUDE.md:254 "18 methods via `compute_Q()`.". Source truth re-read at src/core/identification.jl:254-295: 18 total compute_Q branches = 4 structural (cholesky/sign/narrative/long_run) + 5 ICA + 5 ML-branches + 4 hetero; :nongaussian_ml is the unified dispatcher (confirmed src/nongaussian/ml.jl:470), so distinct statistical methods = 5 ICA + 4 ML + 4 hetero = 13. The "18" as a statistical-surface count is genuinely stale. nongaussian.md is ALREADY correct (13 with 5+4+4 breakdown at lines 5, 96, 118; comparison table rows 100-112 enumerate exactly 13) so leaving it untouched is right. CLAUDE.md:254 sits in the statistical-identification-only bullet, so changing it to 13 is justified and matches the already-fixed nongaussian.md:118 phrasing "All 13 methods ... via compute_Q()". docrule.md imposes no numeral-spelling or parenthetical rule that the proposed prose/table edits would violate. Spec is correct and complete as written.', 'corrections': '', 'collision_issues': [278, 281, 290, 293], 'discriminating_power': 'n/a (docs prose/count fix, no test assertions involved)', 'final_effort': 'S'}