Skip to content

Restore full historical Claim payloads for review backlog - #125

Merged
basicmi merged 1 commit into
masterfrom
restore-historical-claim-review-backlog
Nov 17, 2026
Merged

basicmi merged 1 commit into
masterfrom
restore-historical-claim-review-backlog

Conversation

@basicmi

@basicmi basicmi commented Nov 17, 2026

Copy link
Copy Markdown
Owner

Problem

The persistent review cache currently tracks 2,525 unresolved Claim IDs, but most historical entries retain only timestamps and a few metadata fields. The latest discovery run has no changed candidates, so the review artifact lists a large backlog without the underlying Claim statement, source excerpt or technical context.

I inspected the retained Actions artifacts and recovered a complete match for all 2,525 IDs across eight historical runs. This PR makes that recovery automatic and durable.

Changes

  • Add an idempotent migration that downloads eight retained Actions artifacts using a short-lived read-only GITHUB_TOKEN.
  • Validate each archived discovery, Claim Extraction and evidence-graph bundle; recompute the existing deterministic priority score at the original run timestamp.
  • Match recovered IDs to the persisted cache, preserve original first-seen timestamps, and retain the full unverified Claim payload, source quote and priority metadata.
  • Preserve Claim payloads across future change-only discovery runs; newly extracted Claims are also stored in cache.
  • Include full Claim statements and source excerpts in the unresolved review-aging artifact and Markdown summary.
  • Add offline tests for hydration, partial/missing data, idempotence, network failure and cache persistence.
  • Update the retrieval documentation to reflect the two-page experiment from PR Extend targeted arXiv retrieval experiment to two bounded pages #124.

Safety and operations

  • Adds only actions: read to the scheduled discovery workflow.
  • Does not change the human decision ledger, Claim review state, factual verification status or production registries.
  • Priority scores remain routing signals, not truth-confidence scores.
  • Empty/already-hydrated caches skip network access. Unavailable artifacts are reported without discarding the cache.
  • Historical artifacts have normal retention limits; the next scheduled or manually dispatched Discovery run will hydrate the cache while they remain available.

Expected result

The next successful discovery run should report 2,525/2,525 Claim payloads available and produce a genuinely reviewable backlog instead of metadata-only counters.

@basicmi
basicmi merged commit 909efac into master Nov 17, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant