The Oxbox Survey

Read this before trusting a number here. The survey ranks models, and ranking is judgment. Much of the evidence is vendor-reported benchmarks that nobody has audited. Every claim carries one of three labels:

Every model the survey has run appears in a catalog table with three measured 0 to 5 digits, quality, cost, and speed, bucketed by a script from recorded values and never typed by hand. A dash means unmeasured, not zero. Beside them sit a dated disqualifier, an access or venue fact that holds a model out of the manifest whatever its rating, and my Editor's Rating: Good, Acceptable, Marginal, or Poor, always the word and never a color, with a date and a one-line why. A model nobody has run is never rated.

The rating is an opinion built on those, not a test result. oxbox has a test count behind every claim; the survey cannot.

Issues come out periodically and carry the date they were written. Subscribe by RSS, or read the source repository, which also holds the catalog snapshots each issue is built from.