Company matching relied on normText(), which kept legal-form suffixes,
spacing and umlaut spelling — so "Bosch GmbH", "Bosch AG" and "bosch gmbh"
did not match the same entry, making the blacklist unreliable.
Introduce firmaSlug(): transliterate German umlauts, strip diacritics and
trailing legal-form tokens (GmbH/AG/SE/KG/…), then kebab-join. Add a
firma_slug field to jobangebote_blacklist (schema + ALTER/backfill migration)
and match on it for typ firma/firma_stelle/auto, falling back to firma_norm
for legacy rows. POST /joboffers rejects blacklisted offers via matchBlacklist
automatically since offerSignature now carries the slug.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Offers are now de-duplicated by normalized URL (tracking params stripped)
in addition to (quelle, external_id), so the same posting never lands
twice — even re-scraped under a new id. Deleting an offer (web or API)
auto-blacklists it, so it can never reappear.
lib/blacklist.js provides shared normalization + matching. Manual entries
can block a URL, a whole domain, a company, or a company+title posting
(gender-marker tolerant). New /blacklist page lists and manages entries.
REST API: GET/POST /joboffers/blacklist, DELETE /joboffers/blacklist/{id};
POST /joboffers returns 409 when blacklisted; DELETE /joboffers/{id}
auto-blacklists (opt out with ?blacklist=false). Swagger updated with the
new paths and schemas.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
thomasandClaude Opus 4.8