How this archive is built

Written so you can judge the data rather than trust it.

How automation is used

Collection, extraction and publishing run automatically once a day. A scheduled job queries public APIs for new announcements, extracts the product name from each headline, pulls migration mentions out of the attached public discussion, rebuilds the site, runs a technical audit, and deploys only if that audit passes. No text on a product page is written by a language model. Every sentence is either a fixed template, a field from a dated record, or a verbatim quote with a link to its source.

What gets recorded

An event is archived when a public announcement matches a shutdown, end-of-life, deprecation, acquisition, pricing-change, breach, or layoff pattern. Political news, physical infrastructure, and general media stories are filtered out — imperfectly. If you find one that slipped through, tell us.

What gets published

Recording is not publishing. A product page appears only when there is something worth reading: migration options named by more than one person, multiple independent reports of the same event, a substantial preserved discussion, or a first-party archive notice from a widely used project. Single undiscussed mentions stay in the dataset but do not get a page. Pages that only restate a source without adding to it are marked noindex — they remain readable, but we do not ask search engines to rank a copy.

How migration options are chosen

They are not chosen. They are extracted. A replacement is listed only when at least two independent commenters named it in the public discussion of that event. Each is shown as a verbatim quote, attributed, and linked to the original comment. Ordering reflects how often each was mentioned — nothing else. Where nothing was named often enough, the page says so rather than filling the gap.

Known limitations

Product-name extraction is pattern-based and sometimes wrong. Discussion-derived migration data skews toward developer tools, because that is where the discussion happens — business software is badly under-covered. Absence from this archive means nothing was detected, not that nothing happened. Support dates are re-published from a third party and should be verified against the vendor before any compliance decision.

Corrections

Errors are fixed on report, and the underlying record is kept rather than quietly deleted. See corrections.