What it labeled
The run’s population: which posts it was given, in the facts the run records.
- Readings it labels for
- 2026-W40 (28 Sep – 4 Oct 2026), 1 release
- Posts
- 200 posts. The count only: the posts themselves are never listed or downloadable.
- Judgments made
- 200 judgments
- Population drawn from
- The output of run 541
- Dates
- Started 8 Oct 2026, 14:26 UTC; finished 8 Oct 2026, 14:26 UTC
Who labeled it
Exactly which labeler and which instructions, so a judgment can be traced to what made it.
- Labeler
- claude-opus-5-5
- Version
- subject_link_check v1 · two blind judges · carried from run 5
- Prompt hash
759ffe7afbb6c33bf23ad3f22b0d5714- Prompt
- Recorded by its hash only: its text isn’t published.
- Judges
- 2, each blind to the others
- Run it checks
- Run 541
- Judgments carried from
- Run 5, where it judged the same links by the same method
Agreement
Link precision, as the readings that rest on it state it: correct ÷ (correct + incorrect) among links judged blind. Unclear links are left out of the ratio and counted beside it. Precision is per linking method and per period.
| Period | Link class | In the readings | Precision | Links judged | Checked by |
|---|---|---|---|---|---|
| V2 | ambiguous alias, rule (run 541) | Counted | 98.3% (95% interval 91.1% to 99.7%) | 60 links judged; 0 unclear, left out of the ratio | This run |
Readings that rest on it
Cited by 97 stored readings in 1 release; all current.
| Release | Week | Readings citing it | Current | List |
|---|---|---|---|---|
| 2026-W40 | 28 Sep – 4 Oct 2026 | 97 | 97 | List 2026-W40’s readings |
A run can rest under thousands of readings, so they’re listed one release at a time.
Runs don’t change
A run is never edited or re-run in place. Labeling the same posts again adds a new run with its own page; the earlier labels stay, and every reading keeps naming the runs it used.