List Errors

Errors grouped by first-error code over the window and filters (spec §5.1.3).

occurrences = Σ error_count; affected_views = views whose FIRST error had the code; pct_of_views = 100 × affected / all views in the window (same filters); message = the most recent first-error message for the code; trend = affected views per org-local day over the last ERRORS_TREND_DAYS days of the window, oldest first, gap-filled. Bucketing is done in Python from view_start so SQLite and Postgres agree.