# Research log — WP22 SHARP-HPI | When (2026-10-21) | Phase | What | |---|---|---| | 00:15 | 0 | Discovery: Vrai Prix production DB on M3U96a, QWHPI engine in PhD/recherche, UQO WP template; plan written (docs/PLAN.md) | | 00:30 | 1 | Immutable snapshot (`sqlite3 .backup`, SHA-256 pinned); four earlier vintages collected for the publication-lag audit | | 00:40 | 1 | Extraction → parquet layers; profiling; spatial join to SDA V2026-07; MAMH register; classification and exclusion ledger (632,006 headline sales) | | 00:50 | 1 | Finding: roll `unite_voisinage` is a unit-level code → fine location = nested metric grid; finding: provider `valeur_role` = assessment in force at sale | | 01:00 | 2–3 | Hedonic engine (pyfixest, Huber-IRLS), benchmarks A–K, state-space module (local level / LLT, robust innovations, grid-EM empirical Bayes), SHARP nested hierarchy, panel schema | | 01:10 | 3 | Literature agent: 115 references verified via Crossref; HMTS and RTD method notes; novelty scan | | 01:15 | 4 | Monte Carlo framework (9 scenarios incl. 2 counter-examples), 80 replications × 9 scenarios launched (12 workers) | | 01:20 | 5 | Full panel: 2,696 cells × 69 months; reliability bundle; representative prices | | 01:25 | 6 | Real-time vintages (July 2022 → October 2026, 10-day delay) launched; validation: aggregation, composition shock, placebo, Moran, repeat sales, bootstrap, ablations, trimming | | 01:30 | 6 | External indices (Teranet, CREA, StatCan) obtained by agent; comparison on annual means; Montréal gap diagnosed (type composition, richer CREA attributes); Québec City agreement | | 01:40 | 7 | Manuscript sections 1–6, 8, 10–11 and appendices drafted; figure and table generators; tests (9 passing) | | 14:00–16:30 | 4–7 | Laptop slept 01:38→13:43 (battery); Monte Carlo resumed and completed (720 reps); vintages regenerated with filtered + smoothed modes; benchmarks and HMTS completed; sections 7 and 9 written; 49-page manuscript compiles with 0 missing macros |