test(scintillation): replay and falsify recovered notebook peak

[jakobtfaber]

What changed

• records the end-to-end Python 3.8.19 replay of the recovered scint_freya.ipynb
• preserves the newly rendered 190 kHz ACF panel and the notebook's archived 3.8 MHz panel side by side
• adds a notebook-parity matched-window falsification runner and arithmetic-parity tests
• runs 24 matched off-pulse windows, shifted windows, lag/fit-window scans, and split-band controls
• preserves the machine-readable results and reviewed four-panel diagnostic figure

Why

The old notebook had been cited as a possible direct route to a successful CHIME scintillation fit. A literal replay was needed to distinguish a reproducible historical result from saved notebook output.

The notebook runs, but the surviving same-named pickle and helper module are not the assets that produced the saved cells: the replayed array is (13082, 6250) after cropping versus (6144, 1562) in the stored output, and the closest recovered helper runs 1,000 MCMC steps versus 200 historically.

Result

The surviving-file replay returns gamma1 = 190.329 +/- 11.999 kHz (R^2 = 0.8999), but the follow-up battery falsifies it as scintillation:

• the notebook's inherited 725:875 integration is off-pulse after the surviving capture is centered at bin 3125
• the exact notebook estimator reproduces 190.329 kHz on that off-pulse window
• 0/24 matched off-pulse ACFs are statistically white
• matched off-pulse widths have median 189.33 kHz and span 151.87--224.82 kHz
• the actual burst window gives 264.18 kHz, only 1.40x the off-pulse median

The recovered-notebook route is therefore closed as a CHIME measurement. The next scientifically credible route is earlier-chain injection/recovery; otherwise CHIME should be treated as an upper-limit or scintillation-quenching constraint.

Validation

• source notebook SHA-256 verified against the recovered ARC and h17 archive copies
• input and helper hashes recorded
• full 16-cell notebook executed twice under Python 3.8.19, including an inline-rendered pass
• FFT ACF checked against the archived overlap-normalized helper arithmetic to machine precision
• ruff check and ruff format on the new runner/test
• 48 passed, 1 xfailed across the falsification, CHIME artifact-guard, and reference-parity tests
• git diff --check
• agent-closeout-check: pass