TeX 59%
Python 40.8%
1# Data sources and sampling-frame rules (WP20). Paths point to read-only copies outside the repository.2sources:3 immoka_db: /Users/simon-pierreboucher/Desktop/Cluster/apps-laptop/toveli/data/src/immoka.db4 vraiprix_db: /Users/simon-pierreboucher/Desktop/Cluster/apps-laptop/toveli/data/src/vraiprix.db5 copy_date: "2026-10-17" # sqlite3 .backup taken from M3U96a:~/apps/vrai-prix/data/ for the Toveli project6 listings_window: ["2026-08-12", "2026-10-17"]78frame:9 house_property_types: ["Maison", "Détachée", "Jumelé", "Maison de ville"]10 max_match_m: 15 # metres between listing geocode and roll unit11 min_images: 412 price_min: 10000013 price_max: 300000014 excluded_sources: ["kijiji", "fb_marketplace"] # no broker structured fields, median 6–10 photos15 # Image hosts whose robots.txt disallows automated retrieval for all agents ("User-agent: * / Disallow: /");16 # verified 2026-10-20 (results/raw/robots_check.txt). The Anthropic URL fetcher also refuses them. Excluded on17 # data-rights grounds; the selection effect is reported (results/tables/robots_exclusion_balance.csv).18 robots_disallowed_hosts: ["media.remax-quebec.com"]1920sample:21 seed: 2026102022 n_listings: 120023 max_images_per_listing: 1224 strata: ["cma", "age_band", "price_tier"] # proportional allocation with a floor of 20 per non-empty stratum25 subsample_opus: 200 # third extractor tier on a nested random subsample26 holdout_cma_for_geo_oos: "quebec" # geographic holdout used in the cost-model calibration27