radar-source-heallisted

# Source self-healing Logging "degraded" run after run is a failure of the radar, not of the source. If you logged the same breakage last run (check the recent tail of `logs/source_rotation.md` and SOURCES.md notes), **fix it this run** — don't just re-log it. ## The heal worklist comes from the log (don't guess) Coverage-honesty means every sweep logs each source as `opened` or `degraded: <reason>` in `logs/source_rotation.md`. So the heal worklist is a precise, queryable set — you do not guess what to fix: - `grep 'degraded:' logs/source_rotation.md | tail` → the candidates. A source marked `degraded` in 2+ recent runs (or obviously degraded NOW — empty feed, 404, JSON/parse error, a feed whose newest item is months old) is OWED a heal this run. One transient blip → log it, retry next run; don't heal on a single failure. - A source that is in a SOURCES.md "swept every run" list but is MISSING from the log entirely (neither opened nor degraded — a coverage lie) is also a heal target: it is usually being skipped because it has no recorded access method and falls out of the list-iteration (this is the alphamatch failure class). Find a working method and record it. The fix is written back into SOURCES.md (below), so the NEXT list-iteration consumes it automatically — that is what closes the loop: iterate → degraded / missing → heal → SOURCES.md method → next iteration succeeds. ## Diagnose the failure class, then repair **Class A — broken / moved / stale