UsageReset — evidence-led reset alerts for AI usage limits

Users of Codex, Claude and Grok want to know when usage limits reset. Public signals about this are scattered across tweets and range from rumor to first-party statements.

Original creator and engineer of the project.

Rust and Axum with SQLite.

xAI-assisted interpretation of public tweets, with deterministic validation of source, author, time and scope before a signal counts as evidence.

First-party evidence of an incoming or effective reset is distinguished from plain model/rumor chatter. The NO/MAYBE/YES scale shows status; MAYBE is explicitly an inference, not a confirmed reset. Alerts are idempotent and evidence-linked.

63 automated tests passed on 5 September 2026, including advance notice, late discovery, scope matching and false-positive classification. On 5 September 2026, the public ledger showed four signals with advance notice, eight observed resets without advance notice, seven expired unconfirmed signals and no explicit retractions. Median advance notice was about 9.1 hours to publication of the confirmation source. A defensible overall prediction-accuracy figure is not yet available.

The public evidence ledger pairs stored signals with matching outcome sources. It separates advance notice, retracted false positives, expired unconfirmed signals and observed resets without advance notice. Model rumors remain unscored. These records do not yet support an overall accuracy percentage.

Public service with evidence-linked alerts.

Source checks, human approval and tests determine whether an AI workflow holds up in practice. Before your project launches, we establish acceptable results, the failures that must be caught and who takes over when results are uncertain.