Gate 1 — repository/data audit: docs/implementation_audit.md (Vrai Prix data substrate verified from local copies; no inspectable cost-approach component; style audit reused from WP18).
Gate 2 — literature: 104 references verified against Crossref/arXiv (docs/verification/), paper/references.bib generated, docs/literature_matrix.csv, synthesis in paper/sections/literature.tex (gap stated as a positioning claim).
Gate 3 — theory: cost identity with scope statements, five evidence classes, aggregation/reconciliation rules, schedule mapping, depreciation, uncertainty taxonomy, two propositions with proofs (framework.tex, appendix).
Gate 4 — implementation: src/ modules, configs/, Makefile, 10 unit tests passing, raw responses persisted, prompts/schemas/models versioned.
Gate 5 — experiments: 14 of 16 pre-registered experiments executed (docs/experiment_plan.md); outputs in results/ with manifest.json; cluster bootstrap; geographic hold-out; MNAR analysis and simulation; Monte Carlo; recorded-sales test.
Gate 6 — manuscript: paper/main.pdf compiled with latexmk (UQO WP template, No. 20), 15 vector figures, 23 generated table fragments, ≈ 1,180 numeric macros; no undefined references or citations.
Gate 7 — integrity: docs/results_validation.md (independent recomputation of headline numbers), citation metadata verified, numbers traceable to result files.
Data rights: docs/data_rights.md; robots-disallowing host excluded (R10); prices/assessments masked; no image stored or reproduced.
Research log: docs/research_log.md (decisions, evidence, spend ≈ US$98–110).
# Blocked (external access) — stated in the manuscript
Vrai Prix production repository (M3U96a) and cost.db not re-inspected: laptop off the cluster LAN, Tailscale logged out → described from local copies and internal notes only.
spbgit remote wp20_uqo.git created and main pushed on 2026-10-20 via the BHS64 tunnel (M1M32 = 10.67.0.2).
MEFQ Part 3E (cost method schedules) full text not retrieved; OEAQ standards edition not confirmed; Altus guide itself gated (secondary source used).
Human expert annotations: no raters available in-session → protocol written (docs/annotation_protocol.md), Experiments 4 and 16 marked not feasible; substitutes (cross-extractor agreement; review-load analysis) reported.
Opus 5.5 image batch (1,994 requests) still processing at the time of writing; Opus text batch collected. The pipeline is tier-dynamic: make collect aggregate experiments figures tables paper will incorporate it (version 1.1).