When a tool you depend on starts rate-limiting in the middle of a workday, the only signal most users get is a red banner and a vague countdown. There is no historical record, no comparable metric across providers, and no way to plan around the next ceiling. Throttle exists because that gap is no longer acceptable for professional AI workflows.
What is missing today
Providers expose the consumer-facing constraint (a percentage, a reset clock) without exposing the underlying time series. You cannot answer 'when do I usually hit weekly cap?' or 'how does provider A pace differently from provider B for the same workload?' from any dashboard you have access to. Throttle records those numeric reads locally, with optional cloud sync, so the question becomes answerable.
What this is not
It is not a prompt logger, not a content monitor, not a behavioral profiler. The extension reads numeric usage signals already present in the page. No prompt content, no chat content, no generated artefact ever leaves the browser. See the privacy page for the full data inventory.
Why publish findings
Aggregate, anonymized counts get published as field notes so they can be cited and challenged on the merits. Research-grade transparency is the only credible posture for measurement infrastructure.