Useful tools.
Claims that earn their place.

What left the shelf

The external series contained 34 generated discoveries and 4 early recommendations. Most generated entries summarised a technology-news article. The early recommendations included first-hand-use claims and strong product verdicts without linked test records.

The old Cursor write-up, for example, called it the best AI code editor “right now”. That is not a comparison we can substantiate from this site. The later generator explicitly forbade first-hand-use claims, but still published proposals directly without an editorial review step.

We are no longer maintaining that recommendation directory. Retirement does not mean those products are bad, unavailable or unsafe. Their claims have not been revalidated. Original titles, dates and source text remain available below.

A reachable link is not a reviewed tool

The old link-check job looked at HTTP responses. It did not test the product, inspect a benchmark or review the article's claims. A last-verified stamp could therefore sit beside content that still needed an editorial check.

Retirement notices keep any historical link-check date and explain that distinction. Retired write-ups leave active search, tag listings and the sitemap. Their original URLs still explain what happened.

What stays, and what can return

The context planner guide now describes whole-request budgets and separate output caps. The text-cost guide removes the unsupported 15 percent accuracy claim. The publishing guide distinguishes checked GitHub jobs from the separate unverified server.

A new external write-up needs a bounded task, primary sources, a clear distinction between reported and performed checks, and a reason to belong here. The bot now proposes an unpublished draft. A generated proposal cannot set its own review state. Updating its source does not restart the old host.

The retired write-ups

38 source records. Original publication dates are retained.