Data audit (P1)
| file | sha256 | size |
|---|---|---|
| data/sales.csv | 0a53f5e851e31130083cfc70aafd2e2a644eadbc734d6928782d6d6cb60caf53 | 1201 lines (header + 1200 data lines), 5 declared columns |
| data/metadata.json | 2aefdcfaf936c7a301d4fa32aa05cb04bff46ad2fd34c423a358e423ec47ac8f | JSON: n_records=1200, sum_price_cad=471287000, n_by_municipality (5 munis) |
Schema: sale_id (S+5 digits), municipality, sale_date (ISO), price_cad (CAD integer), area_m2 (m², 1 decimal). Units per metadata note: CAD, m².
Findings (from naive awk split and csv-module parse):
- 36 garbled data lines at file lines 701–736: random characters, field counts 1–12 under CSV parsing; ids S00700–S00735 are absent from readable rows. Unreadable → excluded, not reconstructed.
- 25 readable rows have price written as a quoted "280,000"-style string (awk sees 6 fields; RFC-4180 parse gives 5). Unambiguous → parsed by stripping separators.
- Readable rows (1164): no missing values, no duplicate sale_ids, no duplicate full rows, all municipalities in control list, dates 2021-01-01..2022-12-31 all valid, price 92,000–1,363,000 (all multiples of 1000), area 39.8–335.9 m², price/m² 1,321–6,923. No unit anomalies detected.
- Per-municipality readable counts vs control: Gatineau 335/346, Laval 274/282, Sherbrooke 209/216, Levis 159/162, Longueuil 187/194 (shortfall 36 total = garbled lines).
- Availability: no input violates the rules; metadata is provider-supplied and allowed. Only rule-relevant risk is reconstructing garbled rows — explicitly not done.