Two fixes to the Rabi tune-up:
- Replace the noise-sensitive argmax peak-finder with a WHOLE-CURVE model fit
(search the candidate pi-amplitude minimizing Sigma|measured - sin2(pi*a/2a_pi)|,
coarse then fine) — what real tune-ups do, and robust where argmax fails: near a
flat maximum shot noise moves the peak sample, but the model fit uses every point.
Browser check: recovered pi-amp 1.428 vs hidden truth 1.426 (was ~0.11 off).
- Add TuneResult::artifacts() so all four routines seal uniformly via seal_artifacts.
- Rename the module's public FRAC/ONE to CAL_FRAC/CAL_ONE: cbindgen exports pub
consts into wai.h, and the bare names collided with quantum.rs's FRAC=30/ONE
(duplicate #define). Now namespaced; header clean (FRAC 30 + CAL_FRAC 20 distinct).
9 lib tests green.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
2c8e5665cb