POSI Global Benchmark Collection

2026-08 publisher-catalog expansion: 3296 additional records were added from full active-journal exports (Elsevier's jnlactive.csv, Frontiers' title list) to validate the ingestion/identity pipeline against real, messy publisher data at scale — not because every title in a publisher's complete catalog is individually "internationally established." These records are unrated (no AJR score is computed for them) and their presence here never makes a journal ranking-eligible on its own — see the identity/registry documentation in posi-data.

Evidence acquisition finding: the first benchmark run scored only 5 of 200 journals. Extending the evidence crawl to check policy subpages (not just the homepage) and widening the ethics-keyword set raised this to 22 of 200 (Nature now scores 71/100) — a real improvement. Expanding the corpus to 600 journals held at the same ~9% rate, ruling out sample size as the cause. A direct investigation found the real bottleneck: 73% of a sampled batch of still-failing journals return HTTP 403 on the first request— Elsevier, Wiley, ACS, APS, AIP, Oxford, and IOP platforms among them. No amount of subpage-guessing or keyword-widening can fix that; it's a structural ceiling of a plain-HTTP-only crawl, not evidence those journals lack real governance. "Not Yet Rateable" on a bot-protected platform means POSI's crawl was blocked — not that no evidence of governance exists. Published here rather than quietly smoothed over, per POSI's own no-hidden-adjustment principle.