Evidence Bundle v1

The evidence bundle is the action layer behind a candidate page. It gives a reviewer enough structure to inspect, replicate, challenge, or extend a record without guessing where the claim came from.

The readable page tells the story. The evidence bundle carries the working materials.

A candidate page is readable, but a bundle is actionable. It puts the public snapshot, evidence dossier, artifact manifest, and contribution routes into one packet for outside review.

The public packet exposes source metadata and provenance fields. It does not automatically publish every licensed full-text document; rights-restricted material stays behind the evidence boundary while the identifiers and claims remain auditable.

Chunk IDs are pointers into TWOG processed research records. They let the internal system re-open the exact supporting material while the public page shows the readable claim and source label.

If MD, docking, omics, or another compute lane informs the candidate, the bundle should include the method reference, input contract, settings, artifact IDs, and run ledger identifiers needed for review.

Snapshot identity

Candidate ID, display ID, content hash, pipeline version, and method reference.

Source dossier

Citation records, identifiers, links, duplicate handling, and supported claims.

Chunk provenance

Research object IDs and chunk IDs that connect public claims to stored source material.

Artifact manifest

Plots, payloads, compute outputs, method versions, container images, and reproducibility notes when available.

Open gaps

Negative coverage, missing identifiers, unresolved citations, and blockers that need follow-up work.

• A bundle is for audit and follow-up work, not medical decision-making.
• Source metadata can be public even when source text has access restrictions.
• Unresolved references are useful defects, not decoration.
• Artifacts should be hashable or otherwise traceable back to their generation run.