Research
The evidence layer under the kit’s designs: what was checked, how the stress test works, and what its results will (and will not) show.
- Literature notes — every source the kit cites, with access dates and verification status (including the sources that could not be fetched directly).
- Stress-test protocol: the method of record for the AI stress test: the planned full design (three prompt strategies × two models, blind cross-judging) and the smaller agent-simulated run actually carried out on 2026-11-11.
The raw outputs and checks live in the repository (not on this site) under research/results/, and evaluation/METHODS.md describes every evaluation method, what was run, by what, and what was not run. No fabricated numbers anywhere.