errors.csv: what each column means One row per error on the error log page, in the page's order. 54 rows. id The error's number. E002 is not in the file. It restates E023: the source listed one defect twice, and it is counted once. stage caught_before_shipping caught before it shipped (the page's first list) shipped it shipped before it was found (the page's second list) period The period the source record files the error under. A: to 2026-03-28. B: to 2026-08-20. C: to 2026-10-07. D: after 2026-10-07. Each end date belongs to the period it closes. For a shipped error the period is the source's, not computed from the dates: five shipped errors came in outside their period's dates and keep the source's period (E023, E035, E037, E049, E052). came_in The date the error entered the work. Blank for E048, which came in on several dates. found The date it was found. days found minus came_in, in whole days. 0 means the same day. Blank where came_in is blank. found_by check a check that runs on its own, or a planned sweep (see sweep) reviewer_agent a separate reviewer agent agents_later_work the agents themselves, in later work owner the site's owner (the page says "me") not_recorded the record does not say sweep yes for the five checks that were sweeps the agents ran on purpose; those do not run unasked. no otherwise. reach work_files stayed in work files owner_notes_or_briefing reached the owner's notes or a briefing to the owner live_orders reached a live run's orders No error reached anything public, so there is no code for it. No error was found by someone outside the project, so there is no code for that either. text One or two sentences on the error. Three rows are withheld: their text says so, and they stay in every count. How the page's numbers are computed A percentage is rounded to the nearest whole percent. "Found by a check or reviewer" counts found_by = check and found_by = reviewer_agent. The median and the longest days in a period leave out E048.