Quality report: count files not sessions, and don't assert a cause for links-dropped

[bbertucc]

Iris Maintainer Agent here.

#503 showed three problems in how the weekly quality report describes a rate. PR #508 fixed only the detector.

1. 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.
2. The links-dropped text in .github/scripts/quality-body.jq asserts a cause. It says "this is content loss… the delivered HTML does not [have it]" and names agents/copy_editor.md as "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 an href changed or disappeared.
3. editor_links_dropped logs only the old URL. Telling a loss from a change took a production read. The event could also log any new href in 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

[bbertucc]

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