Repository navigation
Quality report: count files not sessions, and don't assert a cause for links-dropped #509
Description
Iris Maintainer Agent here.
#503 showed three problems in how the weekly quality report describes a rate. PR #508 fixed only the detector.
- It counts sessions as documents. "4 of 74 documents" was 4 sessions, and 3 of them were one file uploaded three times. This affects every rate in the tally. Sessions store no hash of the uploaded file, so distinct files can't be counted yet.
- The
links-droppedtext in.github/scripts/quality-body.jqasserts a cause. It says "this is content loss… the delivered HTML does not [have it]" and namesagents/copy_editor.mdas "the instruction that is not holding". In Quality regression: links from the source document are being dropped #503, 9 of the 10 links were repairs. The rate can only show that anhrefchanged or disappeared. editor_links_droppedlogs only the old URL. Telling a loss from a change took a production read. The event could also log any newhrefin the same block.
Proposed fix:
- Reword (2) to state what was measured.
- Add (3).
- Leave (1) open until there's a way to store a file hash per session. This needs a schema change, so I'll ask before starting it.
🤖 Generated with Claude Code
Activity
fix(quality): cap `added`, and put droppedHrefs' comment back on it (#…
fix(quality): links-dropped states what it measured, and logs what wa…
bbertucc commentedon Oct 4, 2026
Iris Maintainer Agent here.
#510 merged items 2 and 3. Item 1, counting files instead of sessions, is still open. It needs a file hash stored per session, which is a schema change, so I'm leaving it until it's asked for.
🤖 Generated with Claude Code
Count distinct uploaded files in the quality tally (#509) (#550)
bbertucc commentedon Oct 17, 2026
Iris Maintainer Agent here.
#550 (merged as 7370bd0) did item 1. Each new session stores a hash of its upload, and the weekly report now says how many distinct uploads the documents came from. Documents from before the change aren't hashed, and the report counts them separately. Items 2 and 3 were merged in #510, so this issue is done.