spb/wp29_uqo

1import csv, os2ROOT="/Users/simon-pierreboucher/Desktop/Recherche-UQO/methodes/autonomous-research-claude-code/research"3man={r["wp"]:r for r in csv.DictReader(open(os.path.join(ROOT,"corpus_manifest.csv"),newline="",encoding="utf-8"))}4F=["wp","file","document_type","source_url","retrieved_utc","sha256","research_domain","research_type","explicit_question","explicit_estimand_or_hypotheses","stated_assumptions","falsification_or_null_result_conditions","literature_or_novelty_requirement","data_access_controls","temporal_information_controls","baseline_requirement","n_phases_declared","dependency_or_gate_structure","stopping_rules","recovery_or_resume_instructions","artifact_requirements","provenance_requirements","review_or_self_audit_stage","disclosure_or_AI_use_statement","budget_or_resource_limits","integrity_prohibitions","contradictory_or_non_operational_requirements","statements_added_during_implementation","analyst_notes"]5rows=[]6def R(wp,**k):7    m=man[wp]; d={"wp":wp,"file":os.path.basename(m["local_file"]),"source_url":m["url"],"retrieved_utc":m["retrieved_utc"],"sha256":m["sha256"]}8    d.update(k); rows.append(d)910R("WP15",document_type="claude_md",11research_domain="finance / market microstructure (intraday shock propagation, U.S. equities, OHLCV data)",12research_type="empirical (with ML extension and synthetic-data tests)",13explicit_question='"How have the speed, breadth, persistence, and asymmetry of intraday shock propagation across U.S. equities evolved over the past three decades, and under which market conditions does an initially idiosyncratic shock become sectoral or market-wide?"',14explicit_estimand_or_hypotheses='yes — central hypothesis stated and outcome measures formally defined: "Modern markets may absorb ordinary shocks more efficiently while transmitting extreme shocks more rapidly and discontinuously." ; "Define the Intraday Shock Amplification Ratio: ISAR_{i,t}(h) = …" ; "Optimization must not change estimands or silently alter precision."',15stated_assumptions='partial — assumptions must be stated and threats listed, and data assumptions are flagged as unverified: "Explain: … identification assumptions; threats." ; "The natural key is initially assumed to be: (symbol, timestamp) But this assumption must be verified."',16falsification_or_null_result_conditions='yes — "The project must attempt to disprove its own central claim." ; "If a falsification fails, do not hide it. Reassess the claim." ; "A rigorous negative assessment is preferable to a polished but invalid paper."',17literature_or_novelty_requirement='yes — "Do not overclaim uniqueness until a documented literature search confirms it." ; "never cite a work based only on another paper\'s bibliography." ; literature map with "verification date" column required',18data_access_controls='yes — "stop if the schema is unusable or if legal/access restrictions prevent research use." ; "authenticated access only if required" ; "No user-specific absolute path may appear in committed code." ; ".env.example" in tree',19temporal_information_controls='yes — "For a shock at time t, network weights W_{ij,t} must be estimated only from data dated strictly before t." ; "Any liquidity-based inclusion criterion must use lagged data to avoid look-ahead bias." ; "Never randomly split minute observations across train and test." ; "Future information may be used only as a falsification test, never in the baseline estimator."',20baseline_requirement='yes — "Compare with: unconditional historical mean; linear and logistic regressions; penalized regression; random forest; …" ; "one transparent baseline regression" in first-phase deliverables ; "The primary nonlinear result must also be represented with a transparent non-ML specification."',21n_phases_declared='7 (Stages A–G in "3. Execution protocol"; plus "30. Minimum deliverables for the first phase" and "33. Immediate actions" 1–16)',22dependency_or_gate_structure='explicit — "Only after Stage A passes:" ; "Do not run `make release` until the pilot audit and full-data checks pass." ; executable release gate: "The release script must fail if: a table references a missing result; a figure is missing source data; LaTeX does not compile; … manuscript claims disagree with the results registry; data revision is not pinned; tests fail."',23stopping_rules='yes — dedicated section: "# 31. Stop conditions Stop and write a diagnostic report rather than proceeding if: timestamps cannot be interpreted reliably; OHLC values are extensively inconsistent; … data licensing prevents the intended use; …" ; also "stop if the schema is unusable" (Stage A) and "Escalate only issues that materially alter the research question, data legality, identification, or interpretation."',24recovery_or_resume_instructions='yes — "use resumable transfers" ; "permit partial reruns and node-level parallelism" ; "Every distributed task must: be idempotent; write to temporary paths; atomically rename on success; save checkpoints; … support retry; validate output before success." ; "never silently overwrite an existing file"',25artifact_requirements='paper (main.tex + sections, anonymous and named PDFs); Internet Appendix; tables (standalone .tex AND machine-readable .csv/.parquet); figures (vector PDF + PNG, source data in outputs/figures/data/); results registry (outputs/estimates/results_registry.json) feeding LaTeX macros; code (src package, 22 numbered scripts, CLI, Makefile with ~19 targets); tests (8 files); data manifests (raw_files.parquet/csv with sha256, revision, row counts); docs (DATA_DICTIONARY, DATA_AUDIT, EMPIRICAL_DESIGN, LITERATURE_MAP, REPRODUCIBILITY, COMPUTE_LOG, DECISION_LOG, LIMITATIONS); STATUS.md per stage; README, LICENSE, CITATION.cff, pyproject/uv.lock/environment.yml/.env.example; release bundle (cover letter, source_archive.zip, replication_manifest.json, submission_checklist.md); 4 notebooks',26provenance_requirements='yes — "pin the exact dataset revision or commit hash" ; manifest columns include "sha256; status; download_started_at; … code_commit" ; "Each processed file must record: parent inputs; transformation version; filters; row counts before and after; creation timestamp; code commit." ; "Record: Git commit; Python version; package versions; OS; hardware; dataset revision; …"',27review_or_self_audit_stage='yes — "Stage C — Raw-data audit" with docs/DATA_AUDIT.md ; "28.3 Release validation" ; decision log "whether pre-specified or data-driven" ; no adversarial referee stage, but "Report blockers and proposed solutions before launching the full download."',28disclosure_or_AI_use_statement='no — only a style instruction: "The paper must read like a finance paper, not AI-generated prose." (not a disclosure requirement)',29budget_or_resource_limits='partial — resource estimation required but no cap: "Before a full run: 1. benchmark one month; 2. benchmark one year; 3. estimate CPU hours; 4. estimate memory; 5. estimate temporary storage; 6. estimate final storage; 7. record in docs/COMPUTE_LOG.md." ; "disk-space checks" ; "Avoid loading the full dataset into RAM." ; no monetary or token budget',30integrity_prohibitions='yes — "Never fabricate data, results, citations, institutional details, variable definitions, statistical significance, sample sizes, or conclusions. Never write a result in the paper before it has been generated and verified from a saved machine-readable output." ; "Never manually type numerical results into LaTeX." ; "Never winsorize away a possible split without logging it."',31contradictory_or_non_operational_requirements='Non-operational success criteria that no script can check: "the paper documents a new and economically important fact"; "the writing is disciplined"; "prepared to the standard expected in a top finance journal"; "must read like a finance paper, not AI-generated prose". Judgement-dependent trade-off with no threshold: "Do not define isolation so strictly that only trivial events remain. Report the trade-off." External dependency: "main manuscript no longer than the current JF limit" requires checking an external source at run time. Dataset facts are presented as unverified claims ("Claimed coverage: 1992–2026") and must be audited, which is consistent. No internal contradiction found between autonomy ("Do not ask for confirmation for routine implementation choices") and the mandated pause ("Report blockers and proposed solutions before launching the full download") since the latter is an explicit escalation point.',32statements_added_during_implementation='none identified (no dated additions or reconstruction notes; the WP15 PROMPT_INITIAL copy is byte-identical)',33analyst_notes='Longest and most engineering-oriented mandate (7 685 words, 151 headings, 12 code blocks). The only mandate with a dedicated stop-condition section and an executable release-validation gate ("must fail if"). No UQO series framing (targets Journal of Finance), no AI-use disclosure, no page-count target other than JF limit. Cluster-compute section (27) presupposes distributed execution.')3435R("WP18",document_type="claude_md",36research_domain="real estate / housing macro-finance (OLG housing economy, leverage, intergenerational inequality)",37research_type="theoretical (with numerical general-equilibrium simulation)",38explicit_question='"Under what conditions do inherited wealth, mortgage leverage, nominal debt, housing supply and down-payment constraints generate persistent owner-renter wealth inequality among households with similar labor income? What changes under unexpected inflation?"',39explicit_estimand_or_hypotheses='partial — candidate contributions to be treated as hypotheses, not formal H statements: "Candidate contributions: (i) an endogenous ownership-entry wealth threshold jointly shaped by bequests, collateral and price feedback; (ii) conditions for persistent ownership/wealth gaps and path dependence; … Treat each as a hypothesis, not a preordained result." ; ten required experiments listed',40stated_assumptions='partial — assumptions must accompany every proposition but are not stated in the mandate: "For every proposition provide exact assumptions, statement, proof or clearly labeled proof gap." ; one substantive non-assumption imposed: "Do not assume inflation necessarily raises real house prices or enriches every owner."',41falsification_or_null_result_conditions='partial — failed conjectures and contrary outcomes must be recorded, no disproof condition defined: "Save proofs in `paper/appendix_proofs.tex` and failed conjectures in `research/failed_hypotheses.md`." ; "Investigate contrary outcomes." ; "conditions for persistence/multiple equilibria (only if established)"',42literature_or_novelty_requirement='yes — "Produce `research/novelty_matrix.csv` comparing each claimed contribution with the closest prior papers. Redesign weak contributions. Do not assert novelty before verifying it." ; "target of 50–80 **verified** references if genuinely relevant and accessible (never pad)"',43data_access_controls='partial — generic honesty about access only: "Log unavailable files rather than claiming to have read them." ; "ask for essential permissions or missing credentials rather than pretending access" ; calibration "from traceable Canada/Québec sources when justified, or use clearly marked illustrative parameters"',44temporal_information_controls='no — theoretical paper; only "Record units, observation dates, transformations and sources." (no look-ahead rule)',45baseline_requirement='partial — benchmark model and baseline scenario, no comparison methods: "First solve a two-/three-period benchmark with known behavior; unit-test budget identities" ; experiment "1. Baseline wealth and ownership by age, income and inheritance quantile."',46n_phases_declared='5 (Phase 0–4) plus "Final reproducibility and QA"',47dependency_or_gate_structure='explicit — "**Mandatory order:** inspect previous UQO WPs and review literature → specify mathematical model → derive propositions → implement and validate numerical model → … → audit results → **only then write and compile the LaTeX paper**." ; "**Gate:** do not start manuscript drafting until the baseline model is validated, claimed GE equilibria clear markets, principal simulations and exhibits reproduce, bibliography is verified, and theoretical claims have proofs or appropriate qualifications."',48stopping_rules='partial — fallback only, no stop condition: "If blocked, document the issue honestly and deliver a scoped partial result."',49recovery_or_resume_instructions='partial — progress log after every phase, no resume/checkpoint instruction: "Maintain `research/progress.md` after every phase with completed work, evidence, blockers and next steps."',50artifact_requirements='paper/main.tex + main.pdf; references.bib; appendix_proofs.tex; appendix_computation.tex; 10–14 figures (vector PDF + PNG in results/figures/); 6–9 auto-generated LaTeX tables (results/tables/); raw CSV/Parquet (results/data/); logs; code modules src/{model,household_solver,equilibrium,distributions,simulation,calibration,robustness,figures,tables}.py; tests/; configs YAML; Makefile (setup/test/simulate/figures/tables/paper/all); requirements.txt; research/{progress,wp_style_audit,novelty_matrix,literature_matrix,literature_review,model_specification,failed_hypotheses,results_audit}; README',51provenance_requirements='partial — traceability without hashes/manifests: "Every paper number must map to a result file." ; "Record units, observation dates, transformations and sources." ; "pinned dependencies, fixed seeds, tests, structured experiment logs"',52review_or_self_audit_stage='yes — "## Phase 4 — Integrity audit, then write the manuscript Before writing the paper, complete `research/results_audit.md`: experiment, hypotheses, parameters, algorithm, convergence, results, limitations, robustness, and claims actually supported."',53disclosure_or_AI_use_statement='no — "Do not invent coauthors or acknowledgments." concerns authorship metadata, not AI-use disclosure',54budget_or_resource_limits='no',55integrity_prohibitions='yes — "Never invent results, citations, proofs, or data." ; "If general equilibrium does not solve, **never label partial equilibrium as GE**." ; "No fabricated calibration targets or empirical validation." ; "Never manipulate assumptions to force a preferred conclusion." ; "Do not call simulation findings theorems."',56contradictory_or_non_operational_requirements='Quantity targets coexist with anti-padding clauses and are resolved only by a priority statement: "30–45-page English main text", "10–14 meaningful publication-quality figures and 6–9 substantive tables, adjusted to genuine findings rather than padded", versus "**Priority: scientific correctness, originality and reproducibility over speed, page count or positive findings.**" Non-verifiable from the document: "Rerun from a clean environment" (no environment specification), "Check PDFs visually". The gate condition "bibliography is verified" has no operational verification procedure. The French preamble re-maps the directory name used in the plan ("le nom `wp18_uqo/` du plan ci-dessous désigne ce dossier"), i.e. the plan\'s paths are not literal.',57statements_added_during_implementation='Prefatory note (French) placed before the mandate: "(Spécification de l\'auteur, 2026-10-20. Conventions de forme : ~/Desktop/Recherche-UQO/HARMONISATION.md ; ce dossier s\'appelle `immobilier/housing-leverage-intergenerational-inequality` et correspond au dépôt spbgit `wp18_uqo.git` — le nom `wp18_uqo/` du plan ci-dessous désigne ce dossier.)" — a repository-mapping note dated with the specification, not marked as post-implementation.',58analyst_notes='Language: English body with a French prefatory paragraph (automated: mixed). Role framing "Act as an autonomous team of housing economists, …". Requires reading WP1–WP17 for style conventions (research/wp_style_audit.md).')5960R("WP19",document_type="claude_md",61research_domain="real estate / housing market microstructure (search-and-matching, liquidity premium)",62research_type="theoretical (continuous-time search model with numerical simulation)",63explicit_question='"How do market tightness, heterogeneous seller reservation values, financing frictions, time-on-market costs, and bargaining power jointly determine an endogenous housing liquidity discount, transaction prices, selling hazards, and price dispersion?"',64explicit_estimand_or_hypotheses='yes — "Candidate propositions to PROVE: higher carrying cost lowers reservation price; decline in arrival/qualification widens the discount and increases TOM (with caveats); liquidity shocks move prices and volume without a fundamental shock; dispersion from heterogeneous urgency; selection bias in transaction indices." ; defined object "precise liquidity premium/discount definition with sign conventions (D = (p* − p)/p*)" ; "Each: preregistered hypothesis, exact config, output CSV, entry point, diagnostics, interpretation."',65stated_assumptions='partial — required, not stated: "explicit timing, assumptions, parameter restrictions, identification limits, boundary cases" ; "Full assumptions, statements, proofs or counterexamples, intuition; label numerical patterns as conjectures."',66falsification_or_null_result_conditions='partial — counterexamples and failures must be reported; no disproof condition: "proofs or counterexamples" ; "record failed experiments and regions without equilibrium" ; "Report failures."',67literature_or_novelty_requirement='yes — "Novelty gate: if the mechanism is already established, adjust the model and contribution before proceeding." ; "Authors to investigate only with independent bibliographic verification"',68data_access_controls='partial — "No proprietary data, secrets, or unlicensed material." ; "If an endpoint is unavailable, log the failure and proceed" ; "do not claim to have read inaccessible material"',69temporal_information_controls='no — not applicable (theoretical, synthetic experiments)',70baseline_requirement='partial — baseline scenario and frictionless benchmark, not comparison methods: "1 baseline; 2 urgency gradient; …" ; "frictionless benchmark p*" ; "11 Nash vs posted prices"',71n_phases_declared='8 (Phase 0–7); also an 8-step "Mandatory execution sequence: (0) … (7)"',72dependency_or_gate_structure='explicit — "Mandatory execution sequence: (0) inspect prior UQO papers … (7) only then draft and compile the full LaTeX manuscript." ; "Novelty gate" ; "## Phase 6 — Independent audit before prose drafting" ; "## Research integrity and publication quality gates" ; "Introduction and abstract written last."',73stopping_rules='no — "Report failures." only; no condition under which work stops',74recovery_or_resume_instructions='no',75artifact_requirements='full repository; paper/main.pdf, main.tex, references.bib; scripts (src/{model,solver,simulation,calibration,diagnostics,plotting,tables}.py); tests (pytest); config/baseline.yaml; 10–14 figures (PNG + PDF); 6–9 tables; research/{wp_style_audit,literature_review,literature_matrix.csv,novelty_matrix.csv,model_specification,proofs,validation,results_audit,limitations}; README "with commands and output inventory"; "final report (contributions, proved propositions, strongest results with figure/table references, literature coverage, unresolved items, reproducibility status, file locations)"',76provenance_requirements='yes — "Fix random seeds; save environment and configuration hashes." ; "verify every number against result files" ; "`research/results_audit.md` maps every claim to proposition/proof, simulation file/figure/table, or citation"',77review_or_self_audit_stage='yes — "## Phase 6 — Independent audit before prose drafting `research/results_audit.md` maps every claim … classify as proved, numerically illustrated, empirically documented, or speculative"',78disclosure_or_AI_use_statement='no',79budget_or_resource_limits='no',80integrity_prohibitions='yes — "Never invent a proof, result, reference, DOI, dataset, or empirical validation." ; "Never fabricate outputs, citations, proofs, parameters or significance. No empirical language for synthetic experiments." ; "No \\"publishable / first-of-its-kind\\" claims."',81contradictory_or_non_operational_requirements='"preregistered hypothesis" is required for each of the 12 experiments but no registration mechanism, timestamp or location is specified, so pre-registration cannot be verified from the document. "Independent audit" (Phase 6) is to be performed by the same agent that produced the results — independence is not operational. Reference target "Collect and verify 50–80 genuinely relevant references, not a quota" and length "30–45-page main paper" are targets with built-in escape clauses. "compile with latexmk; resolve warnings; inspect the rendered PDF" presumes a TeX toolchain not checked for.',82statements_added_during_implementation='none identified',83analyst_notes='Most compact fully phased mandate (1 092 words, 8 phases). Explicitly scoped against WP18 ("Key distinction from WP18"). Role framing "You are an autonomous interdisciplinary research team".')8485R("WP20",document_type="claude_md",86research_domain="real estate appraisal / multimodal AI (cost approach from listing images and text; Vrai Prix workflow)",87research_type="mixed (empirical evaluation + design-science pipeline, with a clearly labeled synthetic fallback)",88explicit_question='"RQ1: How accurately can multimodal AI extract appraisal-relevant building attributes from listing photographs and descriptions relative to verified property records or expert annotations?" (RQ1 of five)',89explicit_estimand_or_hypotheses='partial — five research questions and a formal valuation identity, no hypotheses: "\\widehat V_i^{cost}=\\widehat L_i + \\widehat{RCN}_i - \\widehat D_i + \\widehat S_i" ; metrics "attribute extraction precision/recall/F1 for categorical features, MAE/RMSE for continuous features, calibration and abstention metrics, and cost-approach accuracy versus independent credible cost or appraisal benchmarks"',90stated_assumptions='partial — "Prove only results supported by stated assumptions." ; "Separate observable visual features, inferred proxy features, verified administrative fields, text-claimed features, and latent/unobservable features."',91falsification_or_null_result_conditions='partial — "Favor transparent uncertainty and falsifiable evidence over marketing language." ; "If evidence for novelty is weak, revise the contribution statement accordingly." ; "never describe synthetic findings as real-world validation" ; no disproof condition for RQ1–RQ5',92literature_or_novelty_requirement='yes — "aiming for 60–100 genuinely verified scholarly references if sufficient relevant work exists; quality and relevance take priority over quota." ; "Establish novelty only after examining existing literature." ; "Gate 2 — Literature: verified bibliography, mapped novelty and gap."',93data_access_controls='yes — "Respect website terms, robots policies where applicable, copyright, privacy, and licensing; do not assume scraping or publication permission." ; "Keep secrets in environment variables, never in tracked files." ; "Never bypass access controls or website restrictions." ; "Avoid any use of private API keys in paper, logs, screenshots, or Git history." ; docs/data_rights.md',94temporal_information_controls='yes — "Prevent train/test leakage by property, duplicate listing, and time." ; "Where feasible, use temporal and geographic holdouts." ; experiment "10. Geographic and temporal out-of-sample performance."',95baseline_requirement='yes — "Compare at least: - B0: traditional cost approach using verified structured fields and standard cost schedules. - B1: listing-text-only extracted attributes + cost model. - B2: image-only … - B3: multimodal image+text … - B4: multimodal + verified administrative fields + uncertainty-aware reconciliation." ; "Do not make comparisons unfair by giving one method privileged information without identifying it."',96n_phases_declared='6 (six-step "Mandatory sequence: (1) … (6) ONLY THEN write"); organised in sections A–J with 7 numbered quality gates',97dependency_or_gate_structure='explicit — "Gate 1 — Repository audit … Gate 2 — Literature … Gate 3 — Theory … Gate 4 — Implementation … Gate 5 — Experiments … Gate 6 — Manuscript: only then draft and compile … Gate 7 — Integrity" ; "Hard gate: Do not draft Results or Conclusions until numerical outputs have been executed, independently checked, and saved."',98stopping_rules='partial — blocker handling, no stop: "When missing permissions, source data, or access makes a claim impossible to validate, clearly identify the blocker, proceed with the portions that are feasible, and provide the exact minimal information needed to finish."',99recovery_or_resume_instructions='partial — "caching, retry limits" ; "Maintain a TODO list and research log." ; no resume instruction',100artifact_requirements='paper/main.pdf + main.tex + sections/ + references.bib + figures/ + tables/; ~12–16 vector PDF figures; 8–12 LaTeX tables generated from saved data; src/{ingestion,extraction,feature_schema,cost_model,evaluation,simulations,visualization}; configs/; data/{README,raw,processed,synthetic}; results/manifest.json; docs/{implementation_audit,literature_matrix.csv,experiment_plan,results_validation,data_rights,limitations,research_log}; tests/; requirements.txt; Makefile; data dictionary + data-flow diagram; annotated extraction schema; reproducibility instructions; limitations report; "final checklist distinguishing completed, blocked, and proposed items"',101provenance_requirements='yes — "Store prompts, model versions, temperatures, and response schemas." ; "generate a machine-readable results/manifest.json" ; "Maintain docs/implementation_audit.md with citations to local paths or source lines" ; "Every substantive factual claim must have a traceable source." ; "Avoid hardcoded table results; generate them from actual outputs."',102review_or_self_audit_stage='yes — "Gate 7 — Integrity: audit every numerical claim against saved outputs and every citation against source metadata." ; "Write docs/results_validation.md documenting each check."',103disclosure_or_AI_use_statement='no',104budget_or_resource_limits='no — "retry limits" only; experiment 16 mentions "annotation time/cost" as an outcome, not a limit',105integrity_prohibitions='yes — "Never invent datasets, numerical findings, literature references, citations, software functionality, API results, human labels, or Vrai Prix implementation details. Never fabricate significance." ; "Do NOT infer foundation integrity, hidden mold, structural soundness, exact year built, precise area, code compliance, permit status, or legal tenure from images alone."',106contradictory_or_non_operational_requirements='Human-dependent steps inside an autonomous mandate: "Propose a rigorous annotation protocol with independent expert raters" and experiment "4. Human expert agreement and AI-human disagreement" cannot be executed by the agent (mitigated by "subject to data availability"). "independently checked" in the hard gate has no independent checker. Verification of 60–100 references "against primary records" presumes network access not confirmed. Figure/table count targets ("approximately 12–16 figures and 8–12 tables") versus "never pad with meaningless exhibits". The French preamble re-maps the directory used in the plan ("le nom `wp20_uqo/` du plan ci-dessous désigne ce dossier").',107statements_added_during_implementation='Prefatory note (French): "(Spécification de l\'auteur, 2026-10-20. Conventions de forme : ~/Desktop/Recherche-UQO/HARMONISATION.md ; ce dossier s\'appelle `immobilier/multimodal-cost-approach-valuation` et correspond au dépôt spbgit `wp20_uqo.git` — le nom `wp20_uqo/` du plan ci-dessous désigne ce dossier. Journal : `docs/research_log.md`.)" — repository mapping and log location, dated with the specification.',108analyst_notes='Language mixed (French preamble). Unusually explicit ethics/rights section (E) and a closing principle: "Produce the strongest defensible scientific contribution possible, not the most optimistic portrayal of AI or Vrai Prix." Keyword family `budget` is inflated by "cost approach" (42 hits of "cost").')109110R("WP21",document_type="claude_md",111research_domain="real estate economics / applied econometrics (systematic review and meta-analysis of hedonic pricing models)",112research_type="empirical (meta-analysis with automated full-text extraction pipeline)",113explicit_question='"RQ1 characteristics that consistently affect prices across countries and periods; RQ2 size and uncertainty of marginal contributions; RQ3 variation with geography, market conditions, property types, socioeconomics" (RQ1–RQ3 of eight, written as fragments)',114explicit_estimand_or_hypotheses='partial — effect-size families and pooled quantities defined, no hypotheses: "Families: percentage premiums, elasticities, semi-elasticities, standardised effects, partial correlations, monetary marginal prices." ; "8.1 random-effects (REML) with pooled estimate, SE, 95% CI, 95% PI, τ², I², k studies, n estimates"',115stated_assumptions='partial — methodological constraints stated as rules rather than assumptions: "no monetary-to-percentage conversion without a defensible reference price" ; "Pool only within comparable groups; document and propagate uncertainty." ; "protocol before synthesis"',116falsification_or_null_result_conditions='no — robustness and publication-selection diagnostics only ("Leave-one-study-out, leave-one-country-out, …"; "funnel, small-study effects, PET-PEESE where appropriate"); no condition under which a claim is abandoned, except "never relax criteria to reach targets"',117literature_or_novelty_requirement='yes — the paper is a literature synthesis; "Compare with competing reviews before any \\"largest\\" claim." ; "Explain precisely how the paper extends Sirmans et al. (2005, 2006)"',118data_access_controls='partial — "read in full when legally accessible" ; "Google Scholar (compliant)" ; no secrets/credential rules',119temporal_information_controls='no — a search cutoff date is set ("Reproducible search 1970–2026-10-20") but no information-leakage rule applies',120baseline_requirement='partial — comparison to a prior synthesis rather than baseline methods: "comparison with Sirmans et al." (required table) ; "RQ5 traditional vs spatial vs machine-learning estimates"',121n_phases_declared='7 (execution order A–G, inline)',122dependency_or_gate_structure='implicit — ordering only, no gate conditions: "15. EXECUTION ORDER. A research design → B literature acquisition → C data construction → D empirical analysis → E scientific interpretation → F manuscript → G final audit. No results written before analysis"',123stopping_rules='no — "document obstacles and continue with feasible work."',124recovery_or_resume_instructions='no',125artifact_requirements='LaTeX paper (main.tex, sections, tables, figures, appendices, references.bib, reproducibility statement); structured research database data/{raw,metadata,extracted,validated,final}; scripts/{discovery,screening,extraction,harmonization,validation,econometrics,visualization}; protocol/{preregistration,eligibility,search_strategy}.md; reports/{search_audit,extraction_audit,methodological_decisions,limitations}.md; PRISMA flow diagram; 13 required tables and 12 required figures; tests; requirements.txt; README',126provenance_requirements='yes — "full records of queries, dates, sources, screening decisions, retrieval failures" ; "Store raw and harmonised values; never infer missing SEs except transparently from reported statistics; preserve every conversion formula; every number traceable to source table." ; "coefficient-level information (… page and table, extraction confidence)"',127review_or_self_audit_stage='yes — "G final audit" ; "Extraction validation: random sample of tables independently verified; report precision, recall, numerical accuracy, error rates; unverified or low-confidence extractions excluded from the primary synthesis." ; "flag ambiguous studies for human review"',128disclosure_or_AI_use_statement='no',129budget_or_resource_limits='no',130integrity_prohibitions='yes — "Never invent studies, citations, coefficients, standard errors, observations, statistical results, or claims of comprehensiveness." ; "never relax criteria to reach targets" ; "Never synthetic results." ; "no promotional language"',131contradictory_or_non_operational_requirements='Aspirational targets ("≥5,000 unique records screened, ≥1,000 full texts, several hundred eligible studies") are explicitly non-binding ("never relax criteria to reach targets"), so compliance cannot be judged. "protocol/preregistration.md" is required but no mechanism makes it precede the synthesis verifiably. "random sample of tables independently verified" has no independent verifier for a single agent; "flag ambiguous studies for human review" introduces a human-in-the-loop step without specifying who or when. "vision/LLM-assisted classification" implies paid model calls with no budget or recording requirement. "read in full when legally accessible" for ≥1,000 full texts is not checkable. R "where superior" is a judgement call.',132statements_added_during_implementation='none identified',133analyst_notes='Densest format: 15 numbered paragraphs, a single Markdown heading (automated n_headings = 1), 1 114 words. Only mandate whose primary object is the literature itself; "Autonomous scientific research assignment, not a narrative review."')134135R("WP22",document_type="claude_md",136research_domain="real estate economics / price-index econometrics (quality-adjusted housing price indices, Quebec transactions)",137research_type="mixed (empirical index construction + Monte Carlo simulation + methodological contribution)",138explicit_question='none — stated as objectives, e.g. "Investigate an adaptive estimator that optimizes the trade-off between local information, borrowing strength, temporal responsiveness, bias, and variance." (not phrased as a question)',139explicit_estimand_or_hypotheses='yes — dedicated section: "## 4. Core estimand Define I[g,k,t] as the quality-constant market price index for geography g, building type k and month t. Establish a reference property distribution and common-support rules. Set 2021 annual average = 100 for the primary index" ; no hypotheses as such',140stated_assumptions='partial — "Enforce identifiability constraints and explicitly describe priors, likelihood, covariance, temporal evolution and estimation algorithm." ; "Make explicit whether the headline is a price index for the stock, for transacted properties, or for a representative constant-quality basket."',141falsification_or_null_result_conditions='yes — "Include deliberate counterexamples where SHARP-HPI underperforms simpler alternatives." ; "Report when reconciliation introduces bias or excessive smoothing." ; "Conduct placebo tests, synthetic quality-composition shocks, …" ; "Test whether estimated changes reflect new transaction coverage rather than market dynamics."',142literature_or_novelty_requirement='yes — "Derive and test an adaptive pooling/shrinkage mechanism; do not call it novel unless a literature review establishes novelty." ; "Explain exactly how the proposed contribution differs from each close predecessor." ; anchor references named (Hill et al. 2022; Ishaak et al. 2024; Hill 2026)',143data_access_controls='yes — "Discover local Vrai Prix sources, read-only credentials, schemas, backups … Never assume credentials, copy secrets into logs, or modify production." ; "Keep raw transaction records private; publish only legally permissible aggregate results. Never commit credentials or personally identifying records."',144temporal_information_controls='yes — "Never use post-transaction information or future data in real-time estimates." ; "(2) real-time vintage series estimated using only information available at each historical publication cutoff." ; "Use forward-chaining and rolling-origin backtesting, never random train-test splits for temporal index claims." ; "Prevent image-derived signals from encoding the sale price, future information, listing outcome or date-dependent model artifacts."',145baseline_requirement='yes — "## 5. Benchmark estimators Implement and compare: A. Raw median and arithmetic mean prices. B. Stratified median. C. Standard pooled hedonic time-dummy OLS. D. Rolling time-dummy … K. Existing QHPI method as a documented benchmark."',146n_phases_declared='8 (Phase 0–7 in "21. Work sequence")',147dependency_or_gate_structure='explicit — "## 12. Simulations — mandatory before paper claims" ; "## 22. Final quality gate No paper can be declared complete until all figures and tables are reproducible, every empirical claim is linked to a generated result, all citations are verified, benchmarks have run, Monte Carlo tests are documented, uncertainty is calibrated, geographic coverage is disclosed, sparse cells are labeled, leakage tests pass, the manuscript compiles without substantive errors, and the results have been critically reviewed for overclaiming." (one terminal gate; phases 0–7 are ordered without per-phase pass criteria)',148stopping_rules='partial — fallback, no stop: "If data access is unavailable, implement testable code and synthetic-data simulations, but never pretend real-data analysis has been completed."',149recovery_or_resume_instructions='partial — "cached intermediate artifacts" ; "At every phase save machine-readable artifacts and an explicit decision log." ; no resume instruction',150artifact_requirements='data_dictionary.md; provenance_report.md; cleaned transaction pipeline + exclusion audit; model specification and derivation; reproducible benchmark and SHARP estimators; Monte Carlo framework and results; monthly index panel in Parquet and CSV with a 23-field row schema; coverage/reliability tables; index vintage/revision database; tables and figures from actual outputs; optional interactive geographic diagnostics; limitations and reproducibility documentation; English LaTeX manuscript with BibTeX; unit/integration tests; versioned configs; environment lockfiles; decision log',151provenance_requirements='yes — "Build an immutable extraction snapshot." ; "provenance_report.md" ; index rows carry "data_cutoff, publication_vintage, model_version, index_method" ; "versioned configurations, unit tests, integration tests, data contracts, environment lockfiles and cached intermediate artifacts"',152review_or_self_audit_stage='yes — "## 1. Audit and provenance" (data audit) ; final gate requires "the results have been critically reviewed for overclaiming" ; "final reproducibility audit" in Phase 7',153disclosure_or_AI_use_statement='no',154budget_or_resource_limits='no — "Profile memory and runtime." and "Use streaming/chunked computation when necessary." are engineering guidance, not limits',155integrity_prohibitions='yes — "Do not invent data, empirical results, references, statistical significance, or novelty claims." ; "Never equate predictive accuracy with index accuracy." ; "A model estimate with zero local transactions is not an observed monthly price movement." ; "Avoid fabricated precision." ; "Do not silently substitute assessment values for market transactions."',156contradictory_or_non_operational_requirements='Terminal-gate conditions that cannot be checked from the document alone: "uncertainty is calibrated" (requires truth, available only in simulation), "critically reviewed for overclaiming" (no reviewer defined), "a publishable contribution". Data-dependent requirements that may be unsatisfiable: "aggregation weights, ideally grounded in eligible housing stock", "Compare aggregate series to suitable external indices", "historical boundary crosswalks". Optional/unverifiable: "interactive geographic diagnostics if practical". Factual assertion inside an instruction: "The October 2026 observation is partial through the actual last reliable transaction date". No explicit escalation or stop rule despite "Begin immediately … Continue autonomously".',157statements_added_during_implementation='none identified',158analyst_notes='Only mandate with a dedicated "Core estimand" section and a prescribed output data schema. 22 numbered sections, 1 966 words. Strongest temporal-information discipline in the corpus (real-time vintages, revision policy).')159160R("WP23",document_type="claude_md",161research_domain="real estate economics / machine-learning forecasting (spatiotemporal, hierarchical and probabilistic price forecasting, Quebec)",162research_type="mixed (empirical forecasting evaluation + simulation study)",163explicit_question='"do newly designed forecasting architectures substantially improve residential real estate price predictions?"',164explicit_estimand_or_hypotheses='partial — forecasting targets defined, no hypotheses: "Task A — future transaction price: P(i,t+h) = f(X(i,t), S(i,t), M(t), h); address transaction selection explicitly." ; "Task B … horizons 1, 3, 6, 12 months" ; "Task C — probabilistic forecasting: predictive distributions (mean, median, intervals, quantiles, downside risk, P(decline), …)"',165stated_assumptions='no — none stated or required beyond "Never assume a variable exists."',166falsification_or_null_result_conditions='partial — the question admits a negative answer and the simulation must show failures, but no disproof condition: "(8) when complex models beat conventional ones" ; "when methods succeed or fail; explain mechanisms" ; "honest novelty statement"',167literature_or_novelty_requirement='yes — "Never manufacture findings or claim novelty without verifying the literature." ; "# 13. LITERATURE REVIEW — … verify every reference; classify contributions (existing / adapted / new combination / original)." ; "literature search first"',168data_access_controls='partial — read-only rule on the upstream project only: "WP22 (`../sharp-hpi-quebec`) is read-only for this project." ; "never modify WP22" ; no credential/secret rules',169temporal_information_controls='yes — "Every feature computed from information available at the forecast origin; no leakage." ; "strict pseudo-real-time expanding/rolling windows; for each origin: information set → estimate → forecast → store → compare. Never random splits." ; "regimes from information available at the origin only" ; "Prevent leakage from WP22: reconstruct index vintages from transactions observable at the origin; audit hedonics, clustering, normalisation, feature selection, imputation, target encoding, graph estimation, tuning, revisions."',170baseline_requirement='yes — "# 6. BENCHMARKS — econometric (hedonic OLS, ridge/EN, SAR, ARIMA/SARIMA/ARIMAX, VAR/BVAR, DFM, local linear trend, state space), ML (RF, XGBoost, LightGBM, CatBoost, SVR, MLP), deep (LSTM, GRU, TCN, TFT, modern TS), simple (random walk, no change, mean growth, seasonal naive, momentum); fair common origins/information sets"',171n_phases_declared='none (18 numbered topical sections, no execution phases; only "literature search first" and "SIMULATION STUDY" ordering cues)',172dependency_or_gate_structure='explicit (terminal gate only) — "# 18. QUALITY GATE — reproducible figures/tables, verified references, leakage audit passed, strict chronology, uncertainty reported, honest novelty statement, manuscript compiles."',173stopping_rules='no',174recovery_or_resume_instructions='no',175artifact_requirements='complete academic LaTeX manuscript (15 sections + appendices on architectures, hyper-parameters, tests, supplementary tables); reproducible Python research pipeline; original forecasting models; empirical results; 15 listed figure types (vector, reproducible); 13 listed table types; supplementary material; simulation study',176provenance_requirements='partial — "reproducible Python research pipeline", "vector; reproducible" figures and a "leakage audit", but no manifest, hash, seed-logging or number-to-file traceability requirement',177review_or_self_audit_stage='partial — "leakage audit passed" inside the terminal quality gate; no dedicated results audit or referee stage',178disclosure_or_AI_use_statement='no',179budget_or_resource_limits='no — a "computational cost" table is required as a reported outcome, not as a limit',180integrity_prohibitions='yes — "Never manufacture findings or claim novelty without verifying the literature." ; "no granular results for inadequate cells" ; "predictive importance ≠ causality" ; "no over-complex benchmarks the data cannot estimate"',181contradictory_or_non_operational_requirements='The document itself declares that it is partly reconstructed: "the pasted text ended at section 17 \\"11. Prob\\"; sections 17–end were completed from the UQO working-paper conventions" and "[Sections 11–15 and the remainder of the protocol were reconstructed by the author\'s assistant because the pasted protocol was truncated.]" — part of the mandate is therefore not the author\'s original instruction. Six named "original" architectures (HSHN, DSGFN, RAREF, BHPF, QANSS, QREFA) are prescribed in advance while novelty is to be "verified" afterwards and "keep only components with incremental out-of-sample value" — pre-specified novelty versus post-hoc selection. "substantially improve" has no effect-size threshold. Heavy methods ("VI / Laplace / HMC on subsets") with no compute budget. Section numbering is internally inconsistent (headings "# 17." and "# 18." versus the note "ended at section 17 \\"11. Prob\\"").',182statements_added_during_implementation='Two explicit markers: (1) header "> Protocol as received on 2026-10-21 (the pasted text ended at section 17 \\"11. Prob\\"; sections 17–end were completed from the UQO working-paper conventions, see docs/PLAN.md). WP22 (`../sharp-hpi-quebec`) is read-only for this project." ; (2) in section 17: "[Sections 11–15 and the remainder of the protocol were reconstructed by the author\'s assistant because the pasted protocol was truncated.]"',183analyst_notes='Telegraphic style (1 132 words, 20 headings). The only mandate whose provenance is partly non-authorial by its own admission. No phases, no stop or recovery rules, no provenance manifest, no self-audit beyond the leakage audit.')184185R("WP24",document_type="claude_md",186research_domain="macro-finance theory (AGI, task-based automation, general equilibrium and asset valuation; extension of Restrepo 2025)",187research_type="mixed (theoretical model + quantitative simulation + empirical calibration/short-run design)",188explicit_question='"How does increasingly capable and abundant AI reallocate factor income from labor to reproducible and scarce capital, change the ownership distribution of aggregate wealth, and jointly determine the prices and expected returns of financial, real-estate, and other assets?"',189explicit_estimand_or_hypotheses='yes — nine target results: "T1. Conditions under which Restrepo\'s labor-share asymptote survives multi-sector production with complementary physical inputs and scarce factors; counterexamples with binding energy/land. T2. Sufficient conditions for factor-income reallocation …" through T9 ; "Each result needs explicit assumptions, mathematical statement, rigorous proof or counterexample, economic interpretation, and simulation check."',190stated_assumptions='yes — primitives are stated and a numbered assumptions protocol and register are required: "Restrepo\'s central primitives include task output X_t(ω)=L_t(ω)+Q_t(ω)/α_t(ω), …" ; "Proof protocol: write assumptions as numbered statements; verify differentiability, convexity/concavity, equilibrium existence, transversality, boundedness, market clearing, timing, stochastic integrability, and no-arbitrage." ; `docs/assumptions_register.md`',191falsification_or_null_result_conditions='yes — "Do not assume that AI automatically raises profits, all asset prices, inequality, or aggregate welfare. Derive conditions and counterexamples." ; "Tests must fail on deliberately inconsistent accounting." ; "identify parameter regimes where asset valuations fall despite higher output" ; "Provide a falsifiable short-run empirical design: … with pre-trends, confounding, measurement error and alternative mechanisms addressed."',192literature_or_novelty_requirement='yes — "Prepare a literature matrix: … Verify exact references from authoritative sources. Mark all unverified claims." ; referee question "(1) is this genuinely new relative to Restrepo and task models?" ; "Explicitly separate the original author\'s conclusions from our proposed extensions."',193data_access_controls='yes — "Do not present proprietary datasets as accessible unless actually available." ; "company annual reports/Compustat if licensed" ; "`data/README.md`, `data/raw/`, `data/processed/`: sources, licenses, provenance and transformations; no fabricated observed data."',194temporal_information_controls='partial — vintage awareness and pre-trend checks, no leakage rule: "Check exact definitions, vintages, geographic scope, units, time frequency and licenses." ; "Validate against pre-AGI historical patterns without claiming the historical sample identifies post-AGI structural parameters." ; "Do not use speculative AGI dates as facts."',195baseline_requirement='yes — "Reproduce a minimal baseline numerically or analytically before adding features." ; "Implement a real baseline first." ; paper section "3. Benchmark Restrepo replication and limitations." ; "Define a transparent benchmark and add layers with tests at every stage."',196n_phases_declared='7 (Phase 0–6) plus a 12-step "EXECUTION ORDER" A–L',197dependency_or_gate_structure='explicit — "## EXECUTION ORDER AND STOP CONDITIONS A. Audit source PDF and write source notes; B. derive smallest exact benchmark; C. … L. perform referee audit and final replication." ; "Never declare a theorem proven without a proof; never declare a model solved without numerical equilibrium checks; never declare a PDF compiled without verifying the file." ; "Implement deterministic perfect-foresight transition first, then aggregate shocks" (no numbered gates)',198stopping_rules='partial — the heading says "STOP CONDITIONS" but the content is a fallback: "If a block cannot be solved, deliver a verified simpler model and clearly document the limitation." ; "start implementing without waiting for further permission unless a genuinely blocking external input is required."',199recovery_or_resume_instructions='partial — "At each stage update `docs/research_log.md` with completed steps, assumptions, residuals, unresolved problems and next actions." ; "Record seeds, solver tolerances and convergence diagnostics." ; no resume/checkpoint instruction',200artifact_requirements='paper/main.tex (35–60+ pages); paper/appendix.tex; paper/references.bib; slides/main.tex (30–45 Beamer slides); src/ (solver, simulations, calibration, tests); data/README.md + raw/ + processed/; results/ (tables, figures PDF+PNG, parameter files, machine-readable outputs); docs/{model_dictionary,assumptions_register,identification,research_log,replication,referee_report}.md; Makefile; requirements.txt or pyproject.toml; environment lockfile; README with one-command reproduction; compiled PDFs "if TeX available"',201provenance_requirements='yes — "sources, licenses, provenance and transformations" for data ; "Build a source-to-parameter table with observed moments, uncertainty and identification limits." ; "Clearly separate externally identified, calibrated and scenario-only parameters." ; "every result traceable to output" ; "Record seeds, solver tolerances and convergence diagnostics."',202review_or_self_audit_stage='yes — "## PHASE 6 — CRITICAL SELF-REVIEW Write an adversarial referee report addressing: (1) is this genuinely new relative to Restrepo and task models? (2) is the model too large to identify? … (12) can the core mechanisms be established in a simpler model? Revise after review." ; docs/referee_report.md',203disclosure_or_AI_use_statement='no',204budget_or_resource_limits='no',205integrity_prohibitions='yes — "no invented citations, DOIs or results" ; "no fabricated empirical estimates, no invented statistical significance" ; "Never declare a theorem proven without a proof; never declare a model solved without numerical equilibrium checks; never declare a PDF compiled without verifying the file." ; "Label conjectures as conjectures, not theorems." ; "Avoid hype words like \'revolutionary\' or \'legendary\' in manuscript."',206contradictory_or_non_operational_requirements='Scope tension: an 8–12-sector IO economy with heterogeneous households, full asset-pricing block, optional monetary/fiscal/open-economy modules and a 2026–2100 simulation versus "Use a nested, modular economy, not an intractable all-at-once model" and "Prioritize a correct publishable contribution over a grand but incoherent system" (resolved only by priority ordering). Length targets ("35–60+ page", "30–45 Beamer") with no anti-padding clause other than the priority statement. External dependency: "Download and read the complete original PDF" presumes network access. "STOP CONDITIONS" heading without any stop condition. Conditional deliverables ("Compile PDFs if TeX available") make completion ambiguous. Non-verifiable: "publication-quality", "disciplined findings".',207statements_added_during_implementation='Explicit post-hoc section: "## UQO conventions (added 2026-10-27) This project is **UQO Working Paper No. 24** in `Recherche-UQO/finance/`. Follow `Recherche-UQO/HARMONISATION.md` (canonical title page, `\\WP…` macros, README template, `paper/Makefile` + `.latexmkrc`, git remote `gitsrv:srv/git/wp24_uqo.git`)."',208analyst_notes='Mandate was written without UQO framing and later attached to the series (addition dated 2026-10-27). Contains a "FIRST RESPONSE EXPECTED FROM CLAUDE CODE" section specifying the first turn. Only mandate requiring seminar slides.')209210R("WP25",document_type="claude_md",211research_domain="economic theory / computational economics (agentic general equilibrium, LLM agents as delegated decision-makers)",212research_type="mixed (theoretical model + heterogeneous-agent numerics + LLM-agent simulation laboratory)",213explicit_question='"How does delegation of economic decisions to autonomous AI agents change equilibrium wages, interest rates, asset prices, capital accumulation, production, credit, systemic risk, and wealth inequality?"',214explicit_estimand_or_hypotheses='yes — six falsifiable hypotheses: "H1: Delegation creates an endogenous adoption threshold determined by information gains, execution costs, compute prices and agency losses. H2: Shared foundation models correlate trades and may reduce market depth under stress, conditional on common information. H3: …" through H6 ; "Pre-register outcomes."',215stated_assumptions='partial — required at every stage and sketched as a non-binding starting point: "(b) write down assumptions" ; "Provide propositions and rigorous proofs under explicit assumptions, or label unsolved statements conjectures." ; "Illustrative starting point, NOT an assumed final model: V_i(s)=max_{c,l,a\',d}{…}"',216falsification_or_null_result_conditions='yes — "For each: mathematical mechanism, formal or numerical prediction, falsification condition, isolation experiment and limitation." ; "Science gate: controlled experiments, uncertainty, negative results, sensitivity and mechanism isolation." ; "Supply counterexamples and boundary cases."',217literature_or_novelty_requirement='yes — "Before developing the model, search and verify literature on … Create `docs/literature_matrix.md` with verified bibliographic details, URLs/DOIs, modeling differences, and the precise remaining gap. No invented references." ; "1. **Novelty gate:** verified literature matrix, precise contribution and competing explanations."',218data_access_controls='yes — "Support local OpenAI-compatible inference and optional BYOK providers; no embedded credentials." ; "Request user input only for indispensable private data, external credentials, expensive compute authorization, or major scope changes." ; "Use verified public macro, labor, wealth, interest rate and financial data where possible."',219temporal_information_controls='yes — "Prevent look-ahead and information leakage." ; "validate surrogates on held-out actions" ; "No final-test-informed" logic implied by "Pre-register outcomes."',220baseline_requirement='yes — "Controls: rational optimized agents, rules, reinforcement learning, single LLM, multiple heterogeneous LLMs, shared LLM with varied prompts, with/without tools, memory and negotiation." ; "Benchmark model with AI switched off against known analytical cases." ; "zero-AI limiting case" (Numerics gate) ; scenario "no AI"',221n_phases_declared='6 (six numbered "Milestones and acceptance gates"; the "Start now" paragraph adds an ordered sequence; no "Phase N" labels)',222dependency_or_gate_structure='explicit — "## Milestones and acceptance gates 1. **Novelty gate:** … 2. **Theory gate:** … 3. **Numerics gate:** … 4. **Agent gate:** … 5. **Science gate:** … 6. **Writing gate:** … Do not proceed to extravagant multi-agent simulations before gates 1–3 are met. If a gate fails, report why and repair the design." ; "Build tiny verified economies first; scale only after tests pass."',223stopping_rules='partial — gate-failure and escalation rules, no stop: "If a gate fails, report why and repair the design." ; "Request user input only for indispensable private data, external credentials, expensive compute authorization, or major scope changes." ; "If internet or TeX is unavailable, state the limitation and continue with honest local deliverables."',224recovery_or_resume_instructions='yes — "Support deterministic seeds, config-driven CLI, checkpointing, experiment manifests, Git revision and reproducible environment locks." ; "Cache reproducibly."',225artifact_requirements='repository scaffold (configs/, data/raw+processed, src/age/{economy,agents,equilibrium,markets,banking,llm,experiments,analysis}, tests/{unit,integration,property}); docs/{literature_matrix.md,model_specification.md,parameter_ledger.csv,research_log.md,limitations.md}; experiments/pre_registration.md; results/{figures,tables} (13 named figure types, 9 named table types, vector quality); paper/main.tex + sections/ + appendices/ (proofs, algorithms, prompts, schemas, replication instructions) + references.bib; README; pyproject.toml; "replication package"; "2–3 page research blueprint"',226provenance_requirements='yes — "Record model/version, prompt, context, temperature, seed, response, latency, tokens and cost. Cache reproducibly." ; "experiment manifests, Git revision and reproducible environment locks" ; "figure/table provenance" (Writing gate) ; "`docs/parameter_ledger.csv`: parameter, units, source, year/geography, moment, method, range, uncertainty."',227review_or_self_audit_stage='yes — "6. **Writing gate:** complete paper, figure/table provenance, verified references, replication package and hostile-referee critique." ; execution protocol step "(e) discuss threats to interpretation"',228disclosure_or_AI_use_statement='no',229budget_or_resource_limits='partial — costs recorded and expensive compute requires authorisation, no numeric budget: "expensive compute authorization" ; "Record … tokens and cost" ; figure "inference cost frontier" ; "Start on a single workstation; optional distributed inference only after a working local baseline."',230integrity_prohibitions='yes — "Never fabricate results, citations, proofs, datasets, or calibrations. Label conjectures and illustrative simulations. Verify novelty before claiming it." ; "LLM output is never trusted as accounting truth." ; "Never claim a test passed unless it ran." ; "Do not call arbitrary LLM-generated market outcomes general equilibrium without verifying feasibility, market clearing and convergence." ; "Never treat simulated LLM preferences as verified human preferences."',231contradictory_or_non_operational_requirements='"Pre-register outcomes." and `experiments/pre_registration.md` without any timestamping/registry mechanism — pre-registration by the same agent that runs the experiments is not verifiable. "hostile-referee critique" is produced by the same model. Scope tension between "Aim for a parsimonious groundbreaking mechanism, not complexity for its own sake" and three nested models (A, B, C) plus an LLM laboratory with eight roles and banking. Negative design constraint stated as judgement: "The key innovation MUST NOT be simply adding AI to a Cobb–Douglas production function." Non-verifiable: "Aim at top economics/finance/computational economics journals without assuming acceptance."',232statements_added_during_implementation='Prefatory note: "(UQO Working Paper No. 25 — repository `Recherche-UQO/finance/agentic-general-equilibrium`, spbgit `wp25_uqo.git`. Programme supplied by the author on 2026-10-27; form follows `Recherche-UQO/HARMONISATION.md`.)" — repository mapping added around the author\'s programme (dated 2026-10-27, same day as the programme).',233analyst_notes='Gate structure is the most formal of the theory mandates (six named acceptance gates with contents). Explicit per-stage protocol "(a) state the precise scientific question, (b) write down assumptions, (c) implement and test, (d) show numerical evidence or mathematical proof, (e) discuss threats to interpretation, (f) log decisions and next tasks."')234235R("WP26",document_type="claude_md",236research_domain="real estate valuation / multi-agent AI (LLM agent deliberation for property valuation, Quebec transactions)",237research_type="mixed (empirical evaluation + design-science multi-agent system + stylised welfare model)",238explicit_question='"RQ1: Do specialized multi-agent systems outperform single LLMs, hedonic models, repeat-sales approaches, gradient boosting, and conventional AVMs on genuinely unseen Quebec transactions?" (RQ1 of six)',239explicit_estimand_or_hypotheses='yes — pre-specified primary outcome: "Pre-register primary endpoints and hypotheses before running final test evaluations. Primary outcome: out-of-time absolute log-price error on arm\'s-length residential transactions. Co-primary or secondary: MAE in CAD, MdAPE, RMSE log, tail error (P90/P95), interval coverage, interval width, CRPS or pinball loss, cost and latency." ; "Propose falsifiable propositions about diversity, correlation, communication topology, and aggregation weights."',240stated_assumptions='partial — "State identification assumptions and develop multiple operational definitions of value." ; "Transaction prices are noisy proxies, not ground-truth fundamental values." ; "Do not claim general existence/uniqueness theorems without proof."',241falsification_or_null_result_conditions='yes — "A negative finding is publishable if the design is rigorous." ; "Report failure as rigorously as success." ; placebo arm "I. Random/shuffled-role or placebo deliberation" ; "Explicitly separate improvements from more tokens, more tools, more models, or more data." ; "Derive testable conditions under which deliberation reduces or increases mean-squared error"',242literature_or_novelty_requirement='yes — "Phase 1: literature review and novelty map; identify close prior multi-agent real-estate/AVM papers and avoid unsupported priority claims." ; "Search and verify the relevant academic literature if browsing is available; otherwise flag bibliography for verification."',243data_access_controls='yes — "These are potential sources, not automatically accessible. Begin with a data-access audit" ; "Do not bypass authentication, scrape restricted endpoints, or infer consent. Do not send personal identifiers, private listing remarks, occupant information, or confidential records to external LLM APIs without authorization and suitable safeguards." ; "use environment variables and a local secrets manager, never put secrets in source control or manuscript" ; "Publish only de-identified or synthetic examples if redistribution is not authorized."',244temporal_information_controls='yes — "Only use features observable as of the valuation timestamp; prevent post-sale information and target leakage." ; "Comparable retrieval must be strictly as-of-date; no same-property future transfers, post-sale listings, or near-duplicate records leaking across partitions." ; "Use a chronological split with train/validation/test and a final untouched holdout" ; "Build a leakage-audit suite and adversarial tests." ; "No final-test-informed aggregation weights."',245baseline_requirement='yes — "Implement and evaluate at least: A. Hedonic OLS / regularized regression (proper spatial/time controls) B. Repeat-sales or hierarchical hedonic … C. CatBoost/XGBoost/LightGBM tuned on validation only D. Strong tabular AVM … E. Single LLM, evidence-matched F. Independent LLM agents aggregated without communication … J. Cost-matched single-agent repeated reasoning and cost-matched ensemble"',246n_phases_declared='9 (Phase 0–8 in "Execution phases and acceptance gates") plus seven "Immediate first actions"',247dependency_or_gate_structure='explicit — "## Execution phases and acceptance gates Phase 0: inventory … Phase 5: run provider-balanced, information-matched, compute-matched randomized ablations on validation; freeze protocol. Phase 6: evaluate final temporal/spatial holdouts once, with clustered uncertainty and cost accounting. … Phase 8: produce manuscript" ; "After each phase, produce a concise `reports/phase_XX.md` with completed artifacts, failed checks, decisions, outstanding assumptions, and exact commands." (ordering and freeze/once-only conditions; no per-phase pass criteria beyond the phase reports)',248stopping_rules='partial — escalation rule and debate-round limits only: "Stop and ask targeted questions only for indispensable access or decisions; otherwise proceed with explicitly documented defaults. Never fabricate completion." ; "Limit debate to pre-specified rounds (e.g. 0, 1, 2, 4)" ; src/debate/ "stopping rules"',249recovery_or_resume_instructions='partial — "privacy-aware caching" ; "bounded retries" ; "store response artifacts securely and measure run-to-run variation" ; no resume/checkpoint instruction',250artifact_requirements='README.md, CLAUDE.md, pyproject.toml, .env.example, LICENSE; configs/ (splits, model IDs, treatments, budgets, seeds); src/{data,features,baselines,agents,debate,evaluation,experiments}; tests/ (unit/integration/leakage/mock-provider); notebooks/; paper/ (LaTeX manuscript, BibTeX, figures, tables, appendices, compilation script, PDF); reports/ (experiment cards, failure analyses, model cards, limitations, phase_XX.md); research_plan.md, data_access_checklist.md, preregistration.md, architecture.md; data dictionaries; clearly labeled synthetic dataset and mock LLM adapters; CLI (ingest, validate, train, run agents, evaluate, generate tables, build PDF); limitations register; API-cost budget',251provenance_requirements='yes — "Store raw snapshots immutable with data dictionaries, versioned transformations, provenance, licenses, and record-level lineage." ; "Record model IDs, API parameters, prompt hashes, temperature, tool calls, token usage, cost estimates, and latency." ; "All figures/tables must come from versioned outputs." ; "Pin dependencies, set seeds, capture versions and hardware"',252review_or_self_audit_stage='yes — roles "11. Data-quality, leakage, and provenance auditor 12. Independent skeptical reviewer / red-team agent" ; "Phase 7: human benchmark if authorized, welfare model, robustness, interpretability audits." ; "Build a leakage-audit suite and adversarial tests."',253disclosure_or_AI_use_statement='no — "Agents are **simulated analytic roles**, not licensed Quebec appraisers or real lenders; never misrepresent professional certification." concerns role representation, not AI-use disclosure for the paper',254budget_or_resource_limits='yes — "Present an API-cost budget with small pilot, medium validation, and final experiment tiers, without inventing vendor prices." ; "token budgets" ; "compare all-Claude, all-OpenAI, and mixed-provider committees at matched cost and information budgets" ; "async concurrency with strict limits"',255integrity_prohibitions='yes — "Never invent data, results, citations, human assessments, API capabilities, or model performance." ; "Never fabricate empirical tables: use placeholders clearly marked `NOT YET ESTIMATED` until generated by scripts." ; "No p-hacking or test-set prompt tuning." ; "do not fabricate a human benchmark" ; "Do not claim deterministic reproducibility for hosted LLMs"',256contradictory_or_non_operational_requirements='"Pre-register primary endpoints and hypotheses before running final test evaluations" and `preregistration.md` have no registry/timestamp mechanism; the same agent writes and executes. "evaluate final temporal/spatial holdouts once" — the once-only rule has no enforcement mechanism. Human-dependent arm: "recruit licensed appraisal professionals and/or trained human appraisers under an approved protocol" cannot be executed autonomously (mitigated: "mark this arm as pending and do not claim it was performed"). "Respect Quebec privacy requirements and organizational research-ethics review where applicable" is not operational for an agent. Tension between "make outputs reproducible" and "Do not claim deterministic reproducibility for hosted LLMs" (resolved by storing response artifacts). Scope (12 default roles, up to 20, ten treatment arms, three provider configurations) versus an API budget that is to be proposed rather than given.',257statements_added_during_implementation='Explicit post-hoc section: "## Repository conventions (added by the implementation, 2026-10-27) - UQO Working Paper No. 26, folder `Recherche-UQO/immobilier/agentic-valuation-quebec`, git remote `gitsrv:srv/git/wp26_uqo.git`. - Python package: `src/agentic_valuation/` (installed editable: `pip install -e .`). CLI: `av <command>` (see README). - Secrets only in `.env` (gitignored). Snapshot provenance in `data/raw/README.md`. Every result file is produced by a numbered script in `scripts/`."',258analyst_notes='Most complete empirical-design mandate in the corpus: pre-specified primary outcome, placebo arm, cost-matched controls, leakage suite, API budget tiers, phase reports. Research-only clause: "No live autonomous trading, lending decisions, or real estate offers".')259260R("WP28",document_type="claude_md",261research_domain="econometric theory / statistical computing (distributed maximum likelihood, higher-order bias correction, communication-computation trade-off)",262research_type="mixed (mathematical theory + Monte Carlo simulation + distributed-systems benchmark)",263explicit_question='"For regular nonlinear likelihood models with `N` observations and `K` distributed partitions, how can additional local computation replace global network synchronization while preserving the first-order inferential properties of centralized maximum likelihood?"',264explicit_estimand_or_hypotheses='yes — candidate theorem and proof queue: "K = o( N^[ 1 - 1 / ( 2^(q+1) * r ) ] )" ; "**Theorem 3 — Exact exponential boundary.** Derive all Gamma moments needed and establish the nonzero limit `-4 theta_0 c^4` for half-block jackknife + one global Newton step when `K ~ c N^(7/8)`." ; Lemmas 1–7, Theorems 1–4, a failure Proposition and a Corollary are enumerated',265stated_assumptions='yes — baseline model assumptions are listed in the mandate and an assumptions file is required: "### 5.1 Baseline model Start with i.i.d. observations and a correctly specified fixed-dimensional regular likelihood: - `X_1,...,X_N ~ P_{theta_0}` independent; - `theta_0` interior to an open parameter set in `R^p`; - information matrix `I(theta_0)` positive definite; …" ; "State precisely which theorem uses which assumption." ; theory/ASSUMPTIONS.md',266falsification_or_null_result_conditions='yes — "A dead end with a mathematical explanation is better than a fabricated breakthrough. A strong negative result or counterexample may itself become the contribution." ; "**Gate 4:** at least one nontrivial theorem beyond a rephrasing of known one-step-plus-jackknife results is fully proved, **or** the report explicitly concludes that novelty is not yet proved and reframes the paper around validated insights and counterexamples." ; "a counterexample or violation for at least one essential assumption" ; "**Proposition — Failure/counterexample.**" ; "Track E — Lower bounds, counterexamples and impossibility"',267literature_or_novelty_requirement='yes — "Do not claim that distributed gradients, one-step Newton, split-sample jackknife, Richardson extrapolation, `K = o(N^{7/8})`, or `K = o(N^{11/12})` are intrinsically novel. Investigate who has already obtained each idea or bound." ; `literature/PRIOR_ART_MATRIX.csv` with "novelty_threat" column ; "**Gate 1:** no unqualified novelty language is allowed until the closest competing theorems have been inspected and cited." ; six anchor references with DOIs to verify',268data_access_controls='yes — "never scrape or transmit private material to external services without authorization" ; "Do not expose API keys, SSH keys, passwords, endpoint secrets or private datasets in source files, logs, paper, artifacts or commits." ; "Use the researcher\'s available CPU/GPU/cluster systems only when provisioned and authorized." ; "Avoid destructive commands, forced pushes and unrelated repository modifications."',269temporal_information_controls='no — not applicable (synthetic Monte Carlo); only a literature cutoff: "**Current date for literature cutoff:** 2026-10-28; search current material rather than assuming bibliographic currency."',270baseline_requirement='yes — "Full centralized maximum likelihood, using a trusted optimizer and tight tolerances." ; "Credible implementations of at least two strong literature baselines where practical, such as surrogate likelihood/CSL and another relevant distributed estimator." ; audit items "New method comparisons include ordinary local averaging + two Newton rounds." and "A baseline with exact sufficient-statistic aggregation is shown for applicable families."',271n_phases_declared='13 (Phase 0–12, each closed by Gate 0–12)',272dependency_or_gate_structure='explicit — "**Execute in order, completing one phase\'s acceptance tests before starting the next.**" ; "**Gate 0: pass only when** all old claims and artifacts are inventoried, every existing number is tagged as replicated / not replicated / unavailable, and a WP28 workspace exists without clobbering other work." ; "**Gate 6:** all critical tests pass and the code reconstructs original-paper figures within stated tolerances before large experiments begin." ; pass/fail rubric in section 16',273stopping_rules='yes — "Complete the project from initial inspection through PDF validation; stop only when a real blocker makes an indispensable step impossible. Document blocked items, and continue with whatever can be completed." ; "**Gate 11:** produce `FINAL_REVIEW.md` with a final verdict `ready for human review / theoretically incomplete / novelty unresolved / computational evidence insufficient`" ; "Never assert results from jobs that have not finished." ; "Do not publish a preprint or submit to a journal without explicit instruction."',274recovery_or_resume_instructions='partial — phase status and run manifests, failure tracking, no resume instruction: "After completing each phase, append to `PHASE_STATUS.md`: Phase: Start/finish timestamp: What was actually done: Artifacts created: Tests passed: Tests failed: …" ; "Track optimization failures rather than discard failed replications silently." ; "straggler failures" to be bounded',275artifact_requirements='article (~25–40 pages) + mathematical appendix; paper/{main.tex,preamble.tex,sections,appendices,references.bib,figures}; literature/{PRIOR_ART_MATRIX.csv,PRIORITY_AND_GAP.md}; theory/{THEOREM_AUDIT.md,ASSUMPTIONS.md,proofs_check.py,symbolic_checks.py}; src/pamle/ (10 modules) with common-interface estimators; experiments/{configs,run_monte_carlo.py,run_benchmarks.py,run_ablations.py,EXPERIMENT_MANIFEST.csv}; tests/ (7 files); results/{raw,processed,tables,figures,resource_profiles}; scripts/{setup.sh,reproduce_all.sh,compile_paper.sh,validate_pdf.sh}; PROJECT_MAP.md, SOURCE_PROVENANCE.md, PHASE_STATUS.md, research_log.md, LIMITATIONS_AND_CLAIMS.md, FINAL_REVIEW.md, FINAL_DELIVERY.md; proof ledger; 8 figures + 6 tables programme; delivery bundle (PDF, .tex, source-and-reproducibility zip, Research_Summary, Theorem_Audit, Prior_Art_and_Novelty, Experiment_Results.csv, Final_Review); nine-point closing report',276provenance_requirements='yes — "Extract the existing archive to an immutable `references/initial_pamle/` directory if available; record checksums." ; run manifest "run_id,git_commit,python_version,dependency_lock_sha256, seed,N,K,m,p,DGP,method,optimizer,tolerance,replicates, parallel_backend,node_count,hardware,OS,wall_seconds, communicated_bytes,rounds,status,raw_output_sha256" ; "Every table value has a traceable input/output path." ; "Project checksums and commit state are captured."',277review_or_self_audit_stage='yes — "## 14. PHASE 11 — Adversarial referee and proof audit Simulate at least three independent review perspectives; log objections and responses" (Referees A, B, C) ; "For every objection, respond with (a) exact location in paper/code, (b) correction made, (c) new proof/test if needed, and (d) residual limitation." ; THEOREM_AUDIT.md with "proved / conditional / open / false" labels',278disclosure_or_AI_use_statement='yes — "Confirm the applicable journal/venue authorship and AI-disclosure policy before a submission version; never falsely represent an AI system as a legally accountable human author." ; "The journal submission version may need GPT-6.1 removed from the formal author list and placed in transparent AI-use disclosure; **flag this for the human author rather than concealing the difference**." ; audit item "AI contribution disclosure and human review requirements are explicit." ; rubric row "Attribution | Simon-Pierre Boucher properly named; AI role transparently recorded"',279budget_or_resource_limits='partial — resources bounded by availability, no numeric cap: "Continue to a publishable-quality, internally consistent draft within the resources and access actually available." ; "`N`: logarithmic sequence … chosen based on available RAM/compute" ; blocker category "(iv) exhausted compute budget" ; "where measured, electricity and dollar-cost assumptions"',280integrity_prohibitions='yes — "No invented theorem, proof, assumption, dataset, benchmark, reference, DOI, parameter estimate, speedup, significance level, or journal acceptance." ; "Do not call an in-memory mathematical partition a distributed cluster benchmark." ; "Never assert results from jobs that have not finished." ; "Do not silently describe these as 1,000 replications or as physical-cluster experiments." ; "Never explicitly invert the Hessian in production code"',281contradictory_or_non_operational_requirements='(1) "Simulate at least three independent review perspectives" versus "do not have the same model merely declare its own work accepted" — the simulated referees are by construction the same model, so independence is not operational. (2) Authorship: "Preserve this provenance and the user\'s requested GPT-6.1 attribution in the working draft" (GPT-6.1 in the author block) versus "never falsely represent an AI system as a legally accountable human author" — resolved only by deferring to the human for the submission version. (3) "Research program funding line to include **only if confirmed** for this particular WP28 project" — confirmation requires the human; the agent cannot satisfy it. (4) Compute: "target at least 1,000 per **main statistical design**" across ≥4 N values, several K, several p, ≥3 models and DGP conditions with no budget beyond "where practical". (5) Conditional bonuses ("If Lean is installed, formalize at least one deterministic algebraic lemma as a bonus, not as a blocker"). (6) Rubric criteria such as "substantive new result or defensible negative discovery" and "journal-style typography" are not script-checkable. (7) "The researcher may have access to a MacLustr collection" together with "never assume the existence … of a MacLustr node" is consistent but leaves hardware scope undetermined until run time.',282statements_added_during_implementation='none identified as post-hoc; the header records prior provenance rather than an addition: "**Prior exploratory draft credits:** Simon-Pierre Boucher and GPT-6.1. Preserve this provenance and the user\'s requested GPT-6.1 attribution in the working draft."',283analyst_notes='Longest mandate in the corpus (8 809 words, 76 headings, 27 code blocks, 2 Markdown tables). The only mandate with an AI-use/authorship disclosure instruction, a per-phase acceptance-test rule ("completing one phase\'s acceptance tests before starting the next"), a pass/fail rubric table, a multi-referee audit phase and a verdict vocabulary. Explicitly a re-examination of a prior AI-assisted draft ("This is NOT an instruction to paraphrase the exploratory note.").')284285R("WP27_BRIEF",document_type="design_brief",286research_domain="finance / market microstructure (agent-based limit-order-book simulation with LLM trading agents)",287research_type="simulation (agent-based market simulator with a stylised one-period theoretical model)",288explicit_question='none — the brief describes "what the simulator *actually* implements" and lists hypotheses H1–H6; no research question is stated',289explicit_estimand_or_hypotheses='yes — "Hypotheses the simulator tests: H1 informativeness falls in the share using a single shared model (holding information quality fixed) and rises in M_eff; H2 order-flow correlation and synchrony rise in φ; H3 bubble/crash frequency and liquidity spirals rise with φ × leverage; H4 prompt/model diversity restores part of the loss; H5 communication channels raise synchrony; H6 circuit breakers and position limits cut crash depth at a cost in time-to-absorb." ; objects defined: "price informativeness τ = 1 − Var(v | Y)/σ_v²; pricing error Var(P − v); order-flow correlation ρ_x = Corr(x_i, x_j); synchrony; effective number of models M_eff = 1/Σ_g φ_g²"',290stated_assumptions='yes — model assumptions stated in the text: "One-period, one asset, liquidation value v ~ N(0, σ_v²) around the public PV. N traders receive s_i = v + ε_i with ε_i = η_{g(i)} + u_i" ; "β > 0 a fixed response intensity; the behavioral assumption mirrors rule-like LLM policies and is relaxed in a CARA extension" ; "A competitive, risk-neutral market maker sets P = E[v | Y]"',291falsification_or_null_result_conditions='partial — directional hypotheses are testable and a seed/power design is sketched ("≥ 30 seeds exploratory, power-based for confirmatory"), but no null-result condition is stated',292literature_or_novelty_requirement='no — the brief is addressed to "the sub-agents writing the literature review, theory and pre-analysis plan" but imposes no literature or novelty requirement',293data_access_controls='no — synthetic data; providers listed without access rules: "Providers: OpenAI (Responses API, json_schema), Anthropic (Messages API, json_schema), mock (seeded rule with injected invalid answers)."',294temporal_information_controls='yes — information-set discipline inside the simulator: "Agents never see V_t." ; "No agent sees future draws." ; "orders arrive at decision time + latency and are processed in arrival order"',295baseline_requirement='partial — control arm and matched design rather than named baseline methods: "Matched pairs across `llm_share` share every stream except the LLM slot draw." ; "mock (seeded rule …)" provider ; "frozen decision library replay vs online calls"',296n_phases_declared='none',297dependency_or_gate_structure='none',298stopping_rules='no',299recovery_or_resume_instructions='no',300artifact_requirements='none stated explicitly; implied: literature review, theory write-up ("to be written up by the theory agent, must map to the above") and a pre-analysis plan; the simulator reports "Realized LLM share of agents, orders and volume"',301provenance_requirements='partial — "Randomization: master seed → independent streams (fundamentals, news, slots, policy assignment, LLM slot assignment, activation, tie-breaks, per-agent RNG)." ; "Invalid decisions = hold, logged." ; "exact conservation checked after every fill" ; no manifest/hash requirement',302review_or_self_audit_stage='no',303disclosure_or_AI_use_statement='no',304budget_or_resource_limits='partial — budgets appear as experimental treatments, not as resource limits: "equal token / USD / decision-opportunity budgets"',305integrity_prohibitions='no',306contradictory_or_non_operational_requirements='Not a mandate, so few directives. Non-operational: "power-based for confirmatory" without a power target or effect size; "must map to the above" asserts a correspondence between the one-period theory and the multi-tick simulator (fixed β market orders versus a continuous double auction) that the brief does not check.',307statements_added_during_implementation='none identified — but the whole document is by nature post-implementation ("It describes what the simulator *actually* implements (src/market_agents)").',308analyst_notes='Supplementary design brief, not a CLAUDE.md; coded separately and excluded from all "of 11" counts. WP27\'s own CLAUDE.md returned HTTP 404 and is not represented in the corpus. 820 words, 3 headings.')309310R("WP15_PROMPT",document_type="duplicate",311research_domain="see WP15", research_type="see WP15",312explicit_question="see WP15 (byte-identical duplicate; not coded separately)",313**{k:"n/a — duplicate of WP15_CLAUDE.md (same SHA-256 2dd19c59fd320ee5fcdbb25f9ebd41b08b612703573cd8af119858e21a521b77)" for k in F[9:28]},314analyst_notes='docs/PROMPT_INITIAL_CLAUDE.md in the wp15_uqo repository is byte-identical to docs/CLAUDE.md in the retrieved snapshot. Recorded once in the corpus manifest for completeness; counted as one mandate (WP15) in all tallies. The two paths suggest the initial prompt was preserved verbatim as the working CLAUDE.md, but the snapshot alone cannot establish which file came first.')315316order=["WP15","WP18","WP19","WP20","WP21","WP22","WP23","WP24","WP25","WP26","WP28","WP27_BRIEF","WP15_PROMPT"]317rows.sort(key=lambda r:order.index(r["wp"]))318out=os.path.join(ROOT,"corpus_coding.csv")319with open(out,"w",newline="",encoding="utf-8") as f:320    w=csv.DictWriter(f,fieldnames=F,quoting=csv.QUOTE_ALL,lineterminator="\r\n")321    w.writeheader()322    for r in rows:323        missing=[k for k in F if k not in r]324        assert not missing,(r["wp"],missing)325        w.writerow(r)326# validate327back=list(csv.DictReader(open(out,newline="",encoding="utf-8")))328assert len(back)==13 and all(len(r)==len(F) for r in back)329print("ok",len(back),"rows",len(F),"columns ->",out)330