Extraction diagnostics
Parsing specs have versioned validation and precise field paths. Validate one
without a network request using newsworker spec validate spec.yaml or
FeedSpec.validate().
Use --explain for a short human trace or --explain-json for the stable
diagnostic model. It reports attempted strategies, selector/item counts, reason
codes, timing, warnings, effective URL, and a heuristic confidence score. The
score is a review aid, not a correctness probability: complete repeated items
raise it, while zero items and missing core fields lower it. Server clients may
opt into bounded confidence and warning headers with diagnostics=1.
Configured credentials are redacted before diagnostics or batch errors are serialized.