Skip to main content

​
Where the numbers come from

Pitboard asks Anthropic about Claude Code accounts and OpenAI about OpenAI’s Codex CLI accounts, all at once, with the login in use or the account’s parked login. An answer counts the account’s work on every machine. For OpenAI, Pitboard sends Codex’s own usage request, which spends no limit (read in Codex 0.154.0). Pitboard asks nothing with a login it cannot read or identify, since the numbers could land on the wrong account. It takes no numbers from an OpenAI answer that names another account than the login asked about. As read in Codex 0.164.0, OpenAI’s answer also gives a model’s own limits, which Pitboard shows under the model’s name.

​
How often Pitboard asks

Pitboard asks about an account again only after 1% of its shortest limit’s period: 3 minutes for a five-hour limit, and 1 minute with no reading yet. After a rate limit or an error, it waits longer, as asking.json records. The app and pitboard watch read when they start and every 5 minutes, and see a switch or numbers another run recorded as soon as it is written. Where the system cannot watch Pitboard’s directory, such as on a network drive or a Windows drive in WSL, they look for them every 2 seconds instead. The app also reads when the computer wakes, and when you open its menu or Accounts once its numbers are a minute old. pitboard status --fresh and Refresh skip the 1% floor and a wait after an error, but not a wait the service asked for.

​
One reading per account

The command line, the app and the status line share one reading per account. Anthropic’s answer lists every limit the account has (read in Claude Code 2.1.294), so it replaces the reading whole and can lower a percentage. A higher percentage a session recorded since stands. Where an answer may leave a limit out, Pitboard keeps that limit until it resets:
  • An Anthropic answer without its list of limits, or with a row Pitboard cannot read.
  • Every OpenAI answer. Either of its two periods can be empty. A Pro plan has no five-hour limit, as OpenAI’s Codex pricing said on 23 October 2026. For other plans the reason is not known.

​
When a period starts

A Claude Code account’s five-hour period starts with its first request after the last one reset, one on the smallest model included. It starts at the start of the ten minutes that request fell in: a request at 17:52 starts a period that resets at 22:50. Until then, Anthropic gives the limit no reset. The weekly limit runs on a fixed cycle: the next week starts at its reset, at 0%, whether or not anything is sent. While one of the account’s own limits is used up, Anthropic refuses every request, and a refused request starts no five-hour period. A model’s limit used up alone was not measured. All three were measured on 12 October 2026 with Claude Code 2.1.296 on a Team and a Max account.

​
Pace

pitboard status shows each limit’s pace, and the app shows none. An even pace uses a limit at one rate from the start of its period to its reset: 50% by halfway through a week. Pitboard gives the difference in percentage points. Over pace is more than 5 points over: at its rate so far, the limit runs out before it resets. Under pace is more than 5 points under. Anything closer is on pace. There is no pace in a period’s first 3%, for a limit used up, or where its reset or length is unknown. Pace is as of the reading, so old numbers do not drift towards under pace and hide a warning. Between readings, only a run-out time counts down.