feat(scout): Norway outside Oslo via NAV + dedicated boards, language check, reputation, region split

Norway search (user request 2026-10-21/22: Trondheim, Bergen or similar, not Oslo)
- New location policy norway_no_oslo: keeps a role if any Norwegian location
  is outside the Oslo area. Equinor and Telenor move to it.
- New `nav` adapter for the NAV national register (arbeidsplassen.nav.no):
  keyword queries on a fixed request budget, paced; on HTTP 429 it stops and
  returns a PARTIAL run with a warning instead of failing. Full-ad text checks
  are cached in state/nav_detail_cache.json (ignored). Tested against
  fixtures only: the live API was rate-limiting during the build.
- Dedicated boards where NAV was measured thin (2026-10-22: Nordic 0, DNV 2,
  Equinor 2 vs ~10, NATO 0): Nordic Semiconductor (teamtailor custom domain),
  DNV (new `algolia_jobs` adapter; app id/key/index read from the page each
  run), Arm Trondheim (new `radancy` adapter), Microsoft Norway (pcsx).
  Verified live: 7 / 6 / 7 / 6 roles outside Oslo.
- NAV rows that duplicate a dedicated board's posting are dropped
  (employer alias + normalized title).

Language / citizenship check
- classify_language_requirement reads the requirement SENTENCE, not the
  posting language: "norsk og/eller engelsk" is English OK, "kommunisere
  godt ... på norsk og engelsk" is Norwegian required. Only an explicit
  Norwegian requirement or Norwegian-citizenship requirement filters a role;
  clearance is noted, never a gate. Runs on full ad text (NAV detail page,
  untruncated Teamtailor/PCSX descriptions). Telenor Lead Platform Engineer
  now reads NORWEGIAN REQUIRED (it read UNCLEAR on truncated text).
- Norwegian title keywords added (they all scored 0 and were never checked).

Employer reputation (user requirement 2026-10-22)
- state/employer_reputation.json (now tracked): dated Glassdoor lookups with
  sources and the approved thresholds (good >= 3.8 on >= 50 reviews; ok
  3.5-3.79 or thin-but-backed; avoid < 3.5; Vizrt excluded). Every report
  row shows the verdict; unlooked-up employers show UNCHECKED.

Report
- Sections per region: Switzerland, Norway, Europe-remote / Denmark (user
  preference); `--region=ch|no` runs one region only.

Fixes
- Microsoft links: apply.careers.microsoft.com<positionUrl>?jobId=<id>
  with a host alias so all 7 logged Microsoft decisions still resolve; the
  old jobs.careers.microsoft.com form redirected to a generic search page.
  PCSX calls retry on 429.

CLAUDE.md corrections log: design/architecture gap (2026-10-21) and the
Microsoft link fix; the EMEA-suffix location leak stays OPEN.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
dennisthiessenandClaude Opus 5.5 committed 2026-10-22 09:19:05 +02:00
1 parent 7d8166ddd2
commit 92c3c005b4
4 files changed
+867 -73

No files matched your search

+1
View File
@@ -30,6 +30,7 @@ job_scout/state/*
!job_scout/state/decisions.json
!job_scout/state/application_cohort.json
!job_scout/state/applications.json
!job_scout/state/employer_reputation.json
!job_scout/state/history/
job_scout/state/history/*
!job_scout/state/history/application_status_before_2026-09-11.json