OPEN RESEARCH PILOT · HUMAN ANNOTATION
Does repetition become evidence?
Help audit how source dependence, context loss and stale documents can distort a retrieval-grounded answer. This is not a poll: every label must point to evidence.
01 · No labels by vibe02 · Similar wording ≠ copying03 · Uncertain is valid
CONTRIBUTE TO THE CORPUS
Bring evidence, not a verdict.
You can label a document in one of the cases below, or bring a new source chain with a primary record and evidence of dependence.
- Existing case: read the captured source and submit an independent label.
- New candidate: link the primary record, one document in the chain, and the evidence that connects them.
PUBLIC LEDGER
Cases under review
Only adjudicated records enter the benchmark.
| Case | Sources | State | Boundary |
|---|---|---|---|
| Rocks, satire, and republication | 3 captured | Collecting blind labels | Google documents a data void and a republished satirical source. |
| Pizza-glue recirculation | 3 captured | Evidence gap | A reported observation; it stays out of the corpus without source-panel archival evidence. |
| Firefox Send staleness | 4 captured | Collecting blind labels | Official end-of-life notice versus historically correct but stale material. |
HOW IT WORKS
Two independent reads, then adjudication.
- Choose a listed case, or submit a new source chain with evidence of dependence.
- Label the document's role, stance, claim relation and source root.
- Give one sentence of rationale tied to the source.
- Your response enters a private moderation queue. Nothing is published automatically.
You can submit anonymously. A GitHub handle is optional and only used for voluntary contributor attribution after review. Do not submit personal data.