A new optical music recognition benchmark ships more than claims: ground-truth MusicXML for 31 engraved pages, phone-degraded photocopies of the same scores, and raw output from three engines, a commercial pipeline, Audiveris 5.11.0, and oemer. The included scorer, score.py, needs zero dependencies, uses strict note-level F1, and rebuilds the published table in about ten seconds. Compact at 2.4 MB under CC BY, it turns OMR figures from trust-me into run-it-yourself.
CrossRef's annual snapshot of every registered DOI is out, now nearly 180 million records in JSON-lines format and free to pull via torrent or AWS, no hitting the API 2 billion times a month. This year adds 12.7M records (up 7.6%), with richer metadata: abstracts up 15%, ORCIDs for authors up 20%. Retraction Watch entries are baked in too.
Want the full arc of the US labor market since COVID in one download? This Kaggle dataset tracks daily (Indeed) job postings across 44 sectors and all 51 states from Feb 1, 2020 through today, with pre-computed stats per sector: net change, pandemic trough and peak, and days-to-recovery. It adds an original SOC-2018 crosswalk that maps Indeed's categories onto standard occupational codes, licensed CC BY 4.0. The fully scripted pipeline rebuilds every file from scratch for easy refresh.
A tracker diffed daily snapshots of 998 companies' Greenhouse, Ashby and Lever boards and caught 18,208 postings whose salary band moved after they went live. Because the diff keeps the prior snapshot, it preserves the old band, which vanishes the moment a company overwrites it. Only 48.8% of currently open postings even list a pay range, and parsing one means escaping Greenhouse's double-escaped HTML. Whoever hosts the pre-change salary else, so this is rare ground truth.