Changelog

What changed in the datasets, baselines, and published results — recorded per edition, so you can see exactly what differs from year to year.

All notable changes to the DISO-OAEI datasets, baselines, and published results are recorded here, per edition. Dataset archives are identified by their SHA-256 checksums (see the downloads table); each edition’s machine-readable results snapshot is archived under results/<year>/.

2026 edition (first edition)

2026-10-04 — Final results published

  • The CodaBench evaluation phase has closed. The final leaderboard snapshot (Task 1: 4 systems; Task 2: 1 system) is archived at results/2026/codabench.json, and the website no longer polls CodaBench.
  • Nine organiser-run systems added to the Task 1 results: ParisMatcher, OLaLaForOAEI, CHARM-localLLM, LogMapKG, LogMapLt, MOSAIC-cli, SECEA, LSMatch-sortedcp and TIM-sortedcp. These are OAEI 2026 participant packages and matchers shipped with MELT, run by the organisers and scored with the unmodified CodaBench Task 1 scoring program. They are archived at results/2026/organiser_runs.json.
  • The results tables report micro-averaged P, R and F1 against the repaired reference R≈+R_{\approx}^{+} (headline) and micro-F1 against the unrepaired R≈R_{\approx}.

2026-07-06 — Finalised datasets published

The 2026 datasets are frozen and published on the Hugging Face Hub at OAEI-ML/diso-oaei, pinned by the v2026 tag: the ten ontologies, the Task 1 dual references (repaired and unrepaired), and the Task 2 candidate pools. SHA-256 checksums for every download are in the downloads table; the archive zips were repacked for release (new checksums), with contents verified identical to the repository’s canonical files. Submission portals opened on CodaBench: Task 1 and Task 2; the evaluation window runs 12 July – 1 September 2026.

2026-06-30 — Provisional materials released

  • Provisional website, task documentation, datasets, and download archives made public for review.
  • Organiser-run baselines published on the baselines page: 13 matching systems for Task 1 (8 covering all six pairs; 5 on the two openly redistributable JC3IEDM pairs) and 2 reference baselines for Task 2.
  • Machine-readable snapshot archived at results/2026/leaderboard.json.

Entry format for future editions

From the 2027 edition onward, entries will record, relative to the previous edition:

  • Dataset changes — ontology version bumps, reference-alignment additions/removals/repairs (with cell counts), candidate-pool regeneration, and the new archive checksums.
  • Protocol changes — metric, validation, or submission-format changes.
  • Results — a pointer to the closed edition’s final archived results under results/<year>/.