house-scrape: re-check huispedia, huislijn, huurportaal and rentola rows and mark clearly gone ones removed

docs/house-tracker.md

@@ -19,6 +19,24 @@ night and is logged as `challenged`; nothing is retried or worked around.
 | Huurstunt              | Cloudflare challenge                                          | skipped                              |
 | Mijndak Midden-Holland | OutSystems app behind a login, no public listing feed         | skipped; use their own e-mail alerts |
 
+**Re-check of imported rows** (`src/lib/house-tracker/recheck.ts`). Rows added by the CLI or
+one-off backfills from huispedia, huislijn, huurportaal and rentola are not in any search
+feed, so after the sources the scrape fetches each row's own page (all four allow detail
+pages in robots.txt), up to 400 per site, oldest `last_seen_at` first, within
+`HOUSE_RECHECK_MINUTES` (default 8). Each site has its own one-at-a-time 6–10 s queue.
+Alive sets `last_seen_at`; gone sets `removed_at` and needs a clear signal:
+
+| site                 | gone when                                                                       |
+| -------------------- | ------------------------------------------------------------------------------- |
+| huurportaal, rentola | redirect to a search page with `?inactive_property_id=`, or 404/410             |
+| huislijn             | redirect to the province overview (id gone from the URL), 404/410, woningruil   |
+| huispedia            | 404/410 ("niet meer beschikbaar"), or Status verhuurd / verkocht / niet te koop |
+
+Errors, timeouts, off-site redirects and unfamiliar pages are `unknown` and change nothing; a
+403/429/503 or challenge stops that site for the night. If more than half of a site's checks
+in one night (min. 5) say gone, none are applied and the run is logged as `suspicious`.
+Try it with `--dry-run --source recheck --recheck-limit 5`.
+
 Same home on two sources: `address_key` (street + number + city, normalised, see
 `src/lib/house-tracker/address.ts`) links the copy to the original via `duplicate_of`; a
 manual entry wins. The page shows one row with "also on …" links. Kamernet never shows house