Throttle Research Program
Throttle studies aggregate AI usage patterns across opt-in users. The goal is to understand how quota limits, reset windows, provider switching, and usage pacing affect real AI workflows in professional environments.
What we study
Quota depletion patterns, redline frequency, reset behavior, provider mix, usage gaps, and multi-account patterns. Aggregated across opt-in users — never analyzed at the individual level.
What we do not study
Prompt content, chat content, generated code, project substance, identifying information, or company-specific data. The extension does not have access to this material in the first place.
Participation
Opt-in only. Off by default for new accounts. Revocable at any time from the dashboard. Aggregated before any analysis. No individual reporting and no contact based on usage data.
Outputs
Public field notes, benchmark memos, periodic usage reports. All findings are released publicly so they can be cited and challenged on the merits.
First report
The first aggregate consumption report — covering Claude and Lovable usage patterns observed through the Throttle extension — is expected in Q3 2026, once the dataset crosses the analytical threshold for non-trivial inference. Reports will be released publicly, dated, and citable.
Until then, /field-notes carries shorter, single-pattern observations as the dataset matures. As a working rule, no aggregate breakdown is published below N=20 contributing accounts.
Methodology details
- Source — Numeric snapshot rows uploaded by paired Throttle extensions to the tokenmaxx backend. No prompt or chat content is ever ingested.
- Eligibility — Only accounts with research opt-in explicitly toggled ON contribute to published analyses. Default for new accounts is OFF.
- Aggregation rule — No breakdown is published below N=20 contributing accounts for any cell. Cells below threshold are suppressed, not estimated.
- Bias disclosure — The dataset is opt-in and skews toward heavy users. Findings are not representative of all AI usage and never claim to be.
Get in touch
Suggestions, contributions, early-pairing requests, or methodology pushback — send them through the form below. Read by Lucio. Not a waitlist. Not a newsletter.