=== RESEARCH ===
{'issue': 290, 'kind': 'docs', 'defect_status': 'confirmed', 'category': 'docs-sweep', 'target_files': ['docs/src/ia_fevd.md', 'docs/src/ia_irf.md', 'docs/src/ia_hd.md', 'docs/src/innovation_accounting.md', 'docs/src/id_testing.md', 'docs/src/id_nongaussian.md', 'docs/src/id_heteroskedastic.md', 'docs/src/nongaussian.md', 'docs/src/structural_identification.md', 'docs/src/nowcast.md'], 'summary': 'Multi-item MED sweep across the identification / innovation-accounting / nowcast doc groups: fix the mislabeled FEVD-CI recipe, add hub/sibling back-links, de-duplicate the six-method table, apply @setup reductions, add two missing bib entries and correct the diagnostic-test count, reconcile the nowcast approach count, and (deferred to T195) the missing Complete Example.', 'current_state': 'All 7 sub-defects still present at HEAD (release/v0.6.6 @ cb4f048).\nITEM 1 (ia_fevd mislabel): ia_fevd.md:30 "Recipe 2: FEVD with bootstrap confidence intervals" wraps `irf(model, 20; ci_type=:bootstrap, reps=50)` (lines 32-35) — an IRF, not FEVD. ia_fevd.md:5 also claims "Frequentist FEVD: VMA-based decomposition with optional bootstrap confidence intervals". SOURCE PROOF: struct FEVD at src/var/types.jl:198-203 has fields decomposition/proportions/variables/shocks only — NO CI fields; bootstrap CIs exist only on LPFEVD (src/lp/types.jl:285) not frequentist FEVD. Claim is false.\nITEM 2 (back-links): IA group hub innovation_accounting.md already links all 3 children (lines 5-7, 99-101) but children under-link: ia_irf.md only has lp_page (line 290), ia_fevd.md only lp_page (line 216), ia_hd.md has ZERO `@ref` cross-refs. StatID group overview nongaussian.md links all 3 children (lines 8, 89-90, 140-142) but children under-link: id_nongaussian.md:10 links siblings only (id_heteroskedastic_page, id_testing_page), id_heteroskedastic.md:292 links sibling id_testing_page only, id_testing.md has ZERO `@ref`. Model pattern per nowcast_dfm.md:5 (links hub nowcast_page + one sibling).\nITEM 3 (dup table): innovation_accounting.md:78-88 carries a six-method identification table (Cholesky/Sign/Narrative/Long-run/Arias/Uhlig) duplicating material owned by structural_identification.md (per-method sections 84-311 + chooser table 317-325); neither page cross-links the other for this table. docrule.md:379 rule 10 = cross-reference, do not duplicate.\nITEM 4 (setup reductions): ia_irf.md:11/13, ia_fevd.md:15/17, ia_hd.md:15/17 all use `Y = Y[end-99:end, :]` and `n_draws=500`. docrule targets: docrule.md:170 `Y[end-59:end,:]` (60 obs), docrule.md:171 `n_draws`→100.\nITEM 5 (bib + count): Székely et al. 2007 cited id_testing.md:167 but absent from id_testing.md References (281-294) — note the full entry already exists verbatim in id_nongaussian.md References ("Székely, Gábor J., Maria L. Rizzo, and Nail K. Bakirov. 2007..."). Lanne & Luoto 2021 cited id_nongaussian.md:342 but absent from id_nongaussian.md References (395-430, only Lanne-Meitz-Saikkonen 2017 and Lanne-Lütkepohl 2010 present). nongaussian.md:6 says "5 diagnostic tests"; SOURCE PROOF id_testing.md has 6 top-level test sections (## at lines 56 Normality, 119 Shock Gaussianity, 140 LR test, 162 Independence, 183 Identification Strength, 203 Overidentification) — the bullet omits Overidentification.\nITEM 6 (Complete Example): structural_identification.md ## headers are 24,68,84,114,163,188,217,271,315,329,343 — no `## Complete Example` (docrule.md:59 requires it before Common Pitfalls). Confirmed missing; hub-vs-method decision owned by T195 (#294).\nITEM 7 (nowcast count/refs): nowcast.md:3 "three complementary approaches" followed by 4 bullets (lines 5-8: DFM, BVAR, Bridge, News). ECB Working Paper refs nowcast.md:266 (Banbura 2023 No.2815), :267 (Cimadomo 2022 No.2696), :269 (Linzenich & Meunier 2024 No.3004) carry neither DOI nor URL.', 'proposed_change': 'ITEM 1 — ia_fevd.md: (a) line 5: drop "with optional bootstrap confidence intervals" from the Frequentist FEVD bullet (frequentist FEVD has no CI; keep just "VMA-based decomposition"). (b) Recipe 2 (lines 30-36): either retitle to "**Recipe 2: IRF with bootstrap confidence intervals**" and add a one-line note that frequentist FEVD is a point estimate (CIs available only via Bayesian FEVD or LP-FEVD), OR replace the recipe body with a genuine no-CI FEVD variant. Do NOT advertise a bootstrap-FEVD-CI capability. (Leave LP-FEVD Recipe 4 and its ci_lower/ci_upper table 225-226 untouched — those are real per src/lp/types.jl:285.)\nITEM 2 — add the missing child->hub + child->sibling links, mirroring nowcast_dfm.md:5. IA group: ia_irf.md add near top "For an overview and method comparison see [Innovation Accounting](@ref innovation_accounting_page); for variance decomposition see [Variance Decomposition](@ref ia_fevd_page)."; ia_fevd.md add hub link + sibling (ia_irf_page or ia_hd_page); ia_hd.md add hub link + sibling. StatID group: id_testing.md add "For an overview see [Statistical Identification](@ref nongaussian_page)" + a sibling (id_nongaussian_page); id_nongaussian.md:10 append overview link to nongaussian_page; id_heteroskedastic.md add overview link to nongaussian_page (its line 292 sibling link can stay). Anchors all exist (ia_irf_page, ia_fevd_page, ia_hd_page, innovation_accounting_page, nongaussian_page, id_nongaussian_page, id_heteroskedastic_page, id_testing_page — verified on the H1 lines of each file).\nITEM 3 — innovation_accounting.md: replace the six-method table (lines 80-88) with a single sentence cross-referencing [Structural Identification](@ref structural_identification_page) for the method catalog; keep the surrounding point/set-ID prose. Add a reciprocal link from structural_identification.md (e.g. in its intro or chooser section) back to [Innovation Accounting](@ref innovation_accounting_page). COLLISION: innovation_accounting.md:91 "(18 additional methods)" is owned by #284 (T185, 18->13) — do NOT edit that number here; sequence T185 first.\nITEM 4 — change `Y[end-99:end, :]`→`Y[end-59:end, :]` and `n_draws=500`→`n_draws=100` in the three @setup blocks: ia_irf.md:11,13; ia_fevd.md:15,17; ia_hd.md:15,17. Per docrule.md:170-171. Also review any prose that says "with 100 observations"/"fewer than 100" (e.g. ia_fevd.md:295, ia_hd.md:274) for consistency but no hard change required.\nITEM 5 — (a) id_testing.md: add Székely 2007 to its References (copy the exact entry from id_nongaussian.md: "Székely, Gábor J., Maria L. Rizzo, and Nail K. Bakirov. 2007. \\"Measuring and Testing Dependence by Correlation of Distances.\\" *Annals of Statistics* 35 (6): 2769--2794. [DOI](https://doi.org/10.1214/009053607000000505)"); optionally normalize the in-text spelling id_testing.md:167 "Szekely"->"Székely". (b) id_nongaussian.md: add Lanne & Luoto 2021 to References — reference: Lanne, Markku, and Jani Luoto. 2021. "GMM Estimation of Non-Gaussian Structural Vector Autoregression." *Journal of Business & Economic Statistics* 39 (1): 69--81. DOI 10.1080/07350015.2019.1629940. NOTE: this DOI has no in-repo source-of-truth; the implementer / #299 (T200) must verify the DOI before committing. (c) nongaussian.md:6: change "5 diagnostic tests" to "6 diagnostic tests" and append "overidentification" to the enumerated list (normality suite, shock Gaussianity, LR test, independence, identification strength, overidentification) — matches the 6 `##` test sections in id_testing.md (56/119/140/162/183/203).\nITEM 6 — structural_identification.md: DEFER to #294 (T195). If T195 keeps it a Method page, add `## Complete Example` before `## Common Pitfalls` (line 329) per docrule.md:59; if T195 designates it a hub, apply the hub skeleton instead. Do not implement standalone — this item is a coordination note, not an independent edit.\nITEM 7 — nowcast.md:3: change "three complementary approaches" to "four complementary approaches" (4 bullets confirmed at lines 5-8). ECB WP refs (nowcast.md:266,267,269): add DOI/URL locators; coordinate with #299 (T200) which owns citation-style unification and may relocate these into bibliography.md.\nmake.jl: NO changes — all 10 pages already registered (docs/make.jl:71-90); no page relocation in this issue.', 'verify': 'Re-verify each edited page (the ia_* pages MUST be re-verified since @setup outputs regenerate): julia --project=docs docs/verify_examples.jl docs/src/ia_fevd.md ; same for ia_irf.md, ia_hd.md, innovation_accounting.md, id_testing.md, id_nongaussian.md, id_heteroskedastic.md, nongaussian.md, structural_identification.md, nowcast.md. Pages with only prose/reference/cross-link edits (id_testing.md, nongaussian.md, nowcast.md, structural_identification.md) have no @example changes but running verify confirms @ref anchors resolve. Full-site anchor check optional: julia --project=docs docs/make.jl (do not run in planning).', 'dependencies': [284], 'collisions': ['docs/src/innovation_accounting.md — #284 (T185 edits line 91 "18 additional methods"->13, same six-method-table region as ITEM 3)', 'docs/src/structural_identification.md — #284 (T185 edits :12,:325 "18 methods"->13) and #294 (T195 owns hub-vs-method decision = ITEM 6)', "docs/src/ia_irf.md — #293 (T194 LOW: :127 'quarters'->months narration)", "docs/src/ia_hd.md — #293 (T194 LOW: :264 'quarters', lambda shadowing)", 'docs/src/ia_fevd.md — #293 (T194 LOW: h1_shares/h20_shares never displayed, Plagborg-Møller orphan ref)', "docs/src/nongaussian.md — #293 (T194 LOW: :128 'Supported symbols' list) and #294 (T195 hub retrofit of overview page)", 'docs/src/innovation_accounting.md — #294 (T195 hub retrofit of Overview page)', 'docs/src/nowcast.md — #299 (T200 owns ECB WP DOI/URL = ITEM 7b) and #294 (T195 hub retrofit)', 'docs/src/id_testing.md — #299 (T200 bibliography extraction = ITEM 5a Székely entry)', 'docs/src/id_nongaussian.md — #299 (T200 bibliography = ITEM 5b Lanne & Luoto entry) and #293 (T194 Keweloh page-range fix)'], 'effort': 'L', 'risk': 'med'}

=== VERIFY ===
{'issue': 290, 'verdict': 'CORRECTED', 'assessment': "Independently read all 10 target docs plus src/var/types.jl, src/lp/types.jl, docs/docrule.md, docs/make.jl at HEAD (release/v0.6.6 @ cb4f048). All 7 sub-defects are genuinely present and every cited file:line matches exactly. Source-of-truth proofs hold: FEVD struct (types.jl:198-203) has no CI fields; LPFEVD (types.jl:285-290) does; docrule.md:170-171 setup targets are Y[end-59:end,:] and n_draws=100; docrule.md:379 rule 10 = cross-reference not duplicate; docrule.md:59 requires Complete Example before Common Pitfalls. Anchors ia_irf_page/ia_fevd_page/ia_hd_page/innovation_accounting_page/nongaussian_page/id_nongaussian_page/id_heteroskedastic_page/id_testing_page/structural_identification_page all verified on H1 lines. Székely full entry confirmed at id_nongaussian.md:421-422 (copy source), Lanne&Luoto confirmed present only in-text (id_nongaussian.md:342) and absent from References. id_testing.md has exactly 6 ## test sections vs nongaussian.md:6 saying 5. The proposed EDITS are all correct. Marking CORRECTED (not CONFIRMED) only because the spec's ITEM 5b advisory metadata is factually wrong: it tells the implementer the Lanne&Luoto DOI has no in-repo source-of-truth and to defer DOI verification to #299, but nongaussian.md:193 already carries the identical entry (same DOI 10.1080/07350015.2019.1629940, JBES 39(1):69-81) — verifiable in-repo, no external dependency needed.", 'corrections': "ITEM 5b: The spec's note that the Lanne & Luoto (2021) DOI 'has no in-repo source-of-truth' and must be verified via #299/T200 is WRONG. nongaussian.md:193 already contains the exact entry with the same DOI (10.1080/07350015.2019.1629940), volume/issue/pages 39(1):69-81. Copy the id_nongaussian-style entry directly; no external DOI verification or #299 blocking is required — it is already cross-consistent in-repo.\n\nITEM 4 (risk, not a spec error): Cutting the ia_* @setup sample from Y[end-99:end,:] (100 obs) to Y[end-59:end,:] (60 obs) regenerates all @example outputs. ia_fevd.md's LP-FEVD Complete Example (structural_lp horizon 20, lags 4, bias_correct bootstrap) leaves only ~36 effective obs at h=20 and fits a bivariate VAR with HQIC per cell — the implementer MUST run docs/verify_examples.jl on ia_irf.md/ia_fevd.md/ia_hd.md after the cut and be ready to keep 100 obs if verify fails. docrule sanctions 60 obs, but the LP-FEVD path is the marginal case.\n\nITEM 1 (preference): Retitling Recipe 2 to 'IRF with bootstrap confidence intervals' makes it a near-duplicate of ia_irf.md Recipe 2 on a FEVD page. Prefer replacing the body with a genuine FEVD variant (e.g. a non-Cholesky-method fevd, or a note that frequentist FEVD is a point estimate with CIs only via Bayesian/LP-FEVD) rather than transplanting an IRF recipe. Also drop 'with optional bootstrap confidence intervals' from ia_fevd.md:5 as specified.\n\nITEM 3 sequencing: replace only the table rows (innovation_accounting.md:80-88); do NOT touch line 91 '(18 additional methods)' — #284 (T185) rewrites it to 13. Sequence T185 first (dependency already listed).", 'collision_issues': [284, 293, 294, 299], 'discriminating_power': 'n/a (docs-only sweep; no test assertions in scope). The only executable surface is @setup/@example regeneration on the three ia_* pages after ITEM 4, which the plan already routes through docs/verify_examples.jl.', 'final_effort': 'L'}