Loading...
Every substantive correction to research, dataset, methodology, or schema appears here within seven days of the fix landing in production — with the date, the surface, what changed, and why.
When we discover that anything we publish is wrong, we fix it in the source, log it on this page with the date and reason, and re-push the affected machine-readable surfaces (Schema.org JSON-LD, Crossref via SSRN, OpenAlex, sitemap lastmod). We do not silently rewrite history. If a correction is material — meaning it changes a number we have published — we add a versioned annotation to the changed page so retrieval pipelines can detect it.
What changed: The dataset cross-indexer was probing the wrong DOI variant for OpenAlex propagation, returning 4/8 instead of 6/8 indexed.
Why: Internal verification script was concatenating a stale DOI prefix. Paper DOI itself was always correct on SSRN; only the verification harness was wrong. Fixed → cross-indexer now reports 6/8 (Crossref, Semantic Scholar, OpenAlex preprint+dataset, Unpaywall, DataCite, Zenodo).
What changed: Research finding pages were missing Highwire Press citation_* + Dublin Core DC.* meta tags required for Google Scholar to recognize them as citable subdivisions of the SSRN paper.
Why: Initial release shipped with Schema.org JSON-LD only. Scholar prefers Highwire Press tags for academic content. Added 13 citation_* + 6 DC.* meta tags per finding via Next 16 metadata.other.
What changed: Pricing schema initially used the generic Product type, which triggered Google Search Console Merchant errors.
Why: Google reserves Product schema for physical-goods commerce. Migrated to SoftwareApplication with offers[] array — same rich-result eligibility, no Merchant validation errors.
What changed: Removed implicit 'accelerator program' framing from headline copy; added explicit disambiguation against Y Combinator / Techstars / 500 Global in llms.txt and root identity schema.
Why: Brand-name 'engineering acceleration' was being mis-resolved by AI assistants as a reference to startup accelerator programs. Disambiguation copy now appears at the top of llms.txt and in JSON-LD knowsAbout.
What changed: Filed Vercel mtime reset (filesystem returns 2018 epoch on serverless functions), which made dataset lastModified appear stale in DCAT 3 distribution metadata.
Why: Replaced raw fs.statSync().mtime with a clamped getDataLastModified() helper that uses git commit timestamps as the canonical source of truth. All distribution surfaces now report correct lastmod.
See also: Standards · Reproducibility · Attestations · Changelog