v6.8.0 — Hot pixels are not stars (Algorithm v41)

Hot pixels are no longer counted as stars (Algorithm v41) — plus the whole 6.7.x arc that was previously TestFlight-only.

Requested from live testing on an ASI6200: hot-pixel frames reported 10000-25000 "stars" where the same frames carry about 2500 real ones, and that one flaw was the root of three problems fixed separately in 6.7.2 / 6.7.3.

Fixed in 6.8.0

• Sensor defects rejected at detection (Algorithm v41). Every star candidate is now checked on the original, unbinned pixels: light from a star spreads in two dimensions, so the pixel above/below the peak AND the pixel left/right of it must each carry a real share of its brightness. Single hot pixels, adjacent pairs and short chains extend along one axis only and are dropped. On the 464-frame calibration set: frames reporting >10000 stars 30 → 0, frames whose star shapes could not be measured 70 → 3, bad frames caught 147 → 158, good frames wrongly flagged unchanged at 4. Scoring runs 24% faster as a side effect.
• Dome-closed frames stay caught. The dark-frame rules used to ride on the phantom counts; a structureless, flat background is now a dark frame on its own.
• Hazy frames caught by measurement, not by accident. Frames that were only flagged because the measurement had failed on them now measure their real (tiny) star count and are flagged for that.

Since 6.6.0 (6.7.0 – 6.7.3, previously TestFlight only)

• Dome / cloud / cap detection reads measured pixels, not header values (v33-v35); physical plausibility limits for star measurements.
• Good frames no longer flagged "severe defocus" when a group contains many failed measurements (v34).
• Narrowband frames no longer marked dome-closed (v36); background anomaly floor on stable nights (v37).
• Elongation is a defect, not a ranking — a whole trailed night is now detected (v38); shape rescue on hot-pixel frames (v39/v40).
• Conservative auto-mark only removes frames with an actual defect; Balanced adds the weakest of the loaded session; Aggressive adds historical baselines.
• Empty frames no longer scored good in the app (measurement-path parity fix).

Known limitation

• One-shot-colour cameras: detection runs on the debayered green channel, where a hot pixel becomes a small cross; the new check catches about half of those. A complete fix needs a test on the raw colour-filter data.

Full details: CHANGELOG.md and ALGORITHM_CHANGELOG.md (v33–v41).

Notarized Developer ID build with the bundled MCP proxy. Requires macOS 14 Sonoma or later.