Registry · Writing & editing · Evidence-grounded editing
ParticleMedia/RAGTruth
Live FHK-0088
Identity
- Repository
- ParticleMedia/RAGTruth
- Repository id
- 735798936
Also filed on the survey's other hub.
Description
Editorial draft A corpus marking unsupported and contradictory spans at word level across generation tasks; adversarial evidence for grounded editing.
The campaign that admitted it
2026-07-30__evidence-grounded-editorial-agents ·
run type (the campaign's own designation) high-recall-map
High-coverage, effort-bounded, not saturated as of 2026-07-31: an evidence-grounded editorial agent is architecturally a versioned artifact whose paragraphs are projections of a typed claim graph (source packet to evidence objects to typed claims to schema/plan to claim-covered paragraphs to deterministic gates to human redline to versioned page), not a prose generator with citations attached.
The claim is the campaign's own record — its coverage and its judgments — quoted word for word. It is not this site's assessment of the repository. Claim wording is the campaign's own, including its shorthand — see the methodology glossary.
2 ledger rows cite this repository.
Current metrics
- Stars
- 260
- Forks
- 35
- Watchers
- 10
- Language
- Python
- License
- MIT
- Archived
- No
- Created
- Last push
- Topics
- None declared
- Snapshot
- · retrieved UTC
Counts only. This site never publishes stargazer, watcher or contributor identities.
Weekly movement
First snapshot 2026-08-10 — weekly movement begins with the second.
Star history
Not yet backfilled. Source will read: an independently collected archive of the same platform's public events, plus our snapshots.
Trust marks
- Snapshot
- · retrieved UTC
- Entries hash (sha256, first 12)
-
40e9e9c1efa4— this hub's served data file — verifiable at /api/entries.json - Snapshot hash (sha256, first 12)
-
5029a1dd0e77— raw capture, shared by both hubs (not yet published) - Method version
- supervised beta (v3.3 taxonomy)
- Sealed record
- 2026-07-30__evidence-grounded-editorial-agents — named, not linked; the record is not public.
What we collect
Collected per entry, weekly, append-only and hash-attested — the full field list is on the data card. Aggregate public metrics only; no stargazer or contributor identities. The posted record is public and quotable; the collection run itself is not published.