Methodology

Principles

• Every published fact carries a source link and the time we last checked it.
• We aggregate; we do not predict. There is no “will Codex reset today” probability here.
• We do not run model evaluations or a model leaderboard. Models appear on plan rows when they affect a quota or a default.
• Nothing is rendered without data. Empty sections say so instead of showing placeholders.

Reset events

A Codex extra reset enters the timeline when a primary post from the Codex team explicitly describes a completed or scheduled reset, a banked-reset grant, or a compensation. We keep that primary link. Announcement time is shown; “landed” is recorded when the source confirms propagation.

“Typical gap” is the median of gaps between consecutive verified extra resets in our own record.

Plans and changes

Plan facts come from official pricing and help pages. Official page diffs stay behind plan rows as provenance; they are not the public changelog. The first changelog desk is Codex: CLI releases, ChatGPT desktop / web notes that name Codex or Work, Codex cloud, and API entries that mention a Codex harness. Official notes are checked on a schedule. We rewrite the note, link the original, and add “Affects” / “Do this” only when the source states a quota, model, or required action. Nightly and alpha CLI builds stay out. Other vendor desks come later.