Every AI usage limit, against the clock.

1. Phase 01 / 03

   Two thirds of the budget left, one fifth of the window gone.

   The fill edge is well to the right of the marker. More usage remains than time remains, so the meter is green and nothing needs your attention. This is the state most usage bars are designed to celebrate.

   Codex · Five-hour 5h window

   ↺ 80%

   Usage 86% Window 80% Ahead

   No projection: at this rate the budget outlasts the window.

2. Phase 02 / 03

   The marker has overtaken the fill edge.

   Same window, later. Usage is draining faster than the clock, so the marker — walking right to left as the window elapses — has passed the fill edge. Amber does not mean you are out; it means the clock is winning, and the projection now says when.

   Codex · Five-hour 5h window

   ↺ 58%

   Usage 38% Window 58% Behind

   Out in ~1h at this pace

3. Phase 03 / 03

   Under fifteen percent: red, whatever the clock says.

   Below fifteen percent remaining the meter reports red regardless of the comparison, because at that point the only useful question is how long you have. A plain usage bar would still be showing you a number that looks survivable.

   Codex · Five-hour 5h window

   ↺ 34%

   Usage 12% Window 34% Low

   Out in ~25m at this pace

The comparison is the whole product: a usage bar tells you what is left, and this tells you whether what is left will outlast the window.

By default the projection is a whole-window average. Turning on local history sharpens it to your recent burn rate, so an idle morning stops flattering a weekly limit.