The state of the corpus that "Journal Atlas: Venue Fit Scoring Under
Incomplete Evidence, and the Corpus Method It Requires" (ACM/IEEE JCDL 2026,
Demonstrations track) reports on.
The corpus keeps growing and a paper's numbers do not. This release exists so
that a reader who finds different figures on the live demonstration can
retrieve the ones the paper reports, rather than conclude the paper was wrong.
What is reproducible from this tree
| Claim in the paper | How to check it |
|---|---|
| 399 entries — 11 Tier 1, 152 Tier 2, 236 AI-Researched | python skills/journal-atlas/scripts/coverage_by_tier.py |
| Evidence coverage by tier: 100% / 100% (55–100) / 40% (40–60) — median and range over entries, three in-domain profiles | same script |
| Entailment audit: 518 claim–source pairs, 30 sampled under a fixed seed, 12 sources unreadable to an automated client, 15 of the remaining 18 entailing their claim | docs/ENTAILMENT_AUDIT.md |
| 17 conference entries whose AI-policy claim was withdrawn after that audit, each carrying a recorded blank and its reason | grep -rl "withdrawn 2026-08-15" skills/journal-atlas/references/journals |
Relationship to the live demonstration
This is also the state deployed to https://journal-atlas.chroniclecore.com on
2026-09-03, so at this commit the paper, the repository and the running service
agree. They are not expected to agree afterwards. The demonstration tracks
main and states which commit it is serving in its coverage panel; this
release is what that panel points back to.
Licence
Corpus and documentation CC BY-SA 4.0. Code MIT.