-
released this
2026-11-14 12:38:59 +00:00 | 1 commits to main since this releaseFixes the "identifying 51495s" report: a sample placed on the detector produced no result, ever.
Every defect below was reproduced offline before being changed, and the fixes were verified on real hardware against a uranium ore source.
The evidence window never ended
The analysis window only ever moved on a rate step of roughly +8 cps, so a strong sample rolled it and a weak one did not. Anything the step detector missed stayed diluted in however many hours the detector had been running. On a 14-hour accumulation a weak ore is 0.03% of the spectrum and could not be identified at any acquisition length.
- The accumulation adopted at startup is now analysed once and kept only if it identified something. Counts are never discarded when they hold a result; hours of plain ambient no longer dilute the next sample.
- The window is capped at 3600 s and slides, always keeping the newest counting.
Weak ore (0.183 uSv/h) presented onto 51495 s of ambient: never identified -> URANIUM at 1800 s, which is what the same ore needs on a clean window.
The ambient floor was learning the sample
With real ore at 3.6-4.1% of counts and 5 sigma sitting on the detector, the learned U-238 floor climbed 1.1% -> 2.1% until the sample stopped counting as elevated, and the device reported BACKGROUND with the ore in front of it. The result persisted in NVS, so it outlived the session.
A floor tracks the minimum, not the mean, so it now only rises toward readings it cannot tell apart from itself.
RELEARN AMBIENT
New fourth button on the Background screen, for the legitimate case the change above does not cover: a detector carried into a genuinely brighter place. It adopts the measured room level rather than resetting to defaults, which may be lower than the new room and so are no remedy at all.
Battery, and the data decoder
BAT --never resolved. The data-buffer decoder loses byte alignment, and the old code gave up on the first bad record, discarding the rest of that buffer - which is where the battery record lives. It now resynchronises on the sequence byte.Measured on hardware: unknown data types 20-75 per 150 s -> 0, sequence jumps 7-12 -> 0, and the battery reads correctly.
Display
- Elapsed time is formatted (
15h 05m) instead of a raw second count. - The age of a resumed run is labelled as such: it is the detector's clock, not how long this sample has been examined.
- Live text is clipped to its card. SOURCE and DEVICE STATUS share a row, so an over-long line painted into the neighbouring card and was never erased.
- The status card now clears the full height of its own glyphs instead of two rows short.
Note for anyone updating
This is the first published release since v1.0.7. Versions 1.0.8 through 1.1.0 were built and tested but never released, so this update covers all of them.
Offline regression unchanged: selftest 55/55, background 900 s, forest 3600 s, thorium 120 s, uranium 300 s.
Downloads