Phase 0 — inventory and gap analysis (2026-10-27)
What existed before this phase
Nothing for WP 26. Reusable assets: WP 22 processed Vrai Prix snapshot (../sharp-hpi-quebec/data/processed), WP 22 geography, WP 16 comparables validation (method reference), WP 20 Claude extraction client (pattern reference). UQO template _modele-working-paper/.
Environment
Laptop: Python 3.14.4 (polars 1.40, pandas 3.0, numpy 2.4, sklearn 1.6, lightgbm 4.6, xgboost 3.2, catboost 1.2.10, statsmodels 0.14, pyfixest 0.60, pydantic 2.13, anthropic 2.3.0, openai 3.40.0), TeX Live (latexmk). Cluster nodes available for heavy baseline tuning (M3U96b).
Access
- Vrai Prix: local snapshot copied (
data/raw/, SHA-256 pinned); live DB readable (771,439 sales to 2026-10-13 → prospective set possible). - Immo-Ka: audited (479,265 listings since 2026-08-12); not used in arms (see
docs/data_access_checklist.md); ancillary geospatial DBs usable. - Providers: OpenAI key works (
/v1/modelslists gpt-5.x/6.x); the Anthropic key supplied in the task is invalid (authentication_error); the WP 14 key authenticates and is used. Smoke test of structured outputs passed onclaude-sonnet-5-5,claude-haiku-5-5,gpt-6.1-sol,gpt-6-luna.
Gaps and decisions
| Gap | Decision |
|---|---|
| No mortgage-rate / macro series in the snapshot | optional external series; agents told the field is unavailable |
| No inspection/condition data | role 5 must report "no inspection available"; condition is unobserved for every arm |
| No human appraisers | arm marked pending; no fabricated benchmark |
| Listings start 2026-08 | not used in arms |
Platform market_index and est_* use future data |
excluded from evidence |
| WP 25 absent from the Recherche-UQO index | number 26 used as instructed; noted in README |
Artefacts produced
CLAUDE.md, pyproject.toml, .env.example, configs/{models,data,treatments,budgets}.yaml, src/agentic_valuation/{config,schemas}.py, providers/ (base, ledger, mock, anthropic, openai), data/synthetic.py (+ data/synthetic/synthetic_transactions.parquet), docs/{architecture,data_access_checklist,research_plan,preregistration}.md, data/raw/README.md.
Commands
python3 -m pip install -e . ; python3 -m agentic_valuation.data.synthetic