Eclipse Calibration Stacking

---
name: eclipse-calibration-stacking
description: Calibrate and stack total solar eclipse RAW, Bayer or linear RGB data using measured registration, calibration, saturation and geometric coverage.
---

# Calibrate and stack

Read [setup and operating instructions](../../docs/USAGE.md) before running AEL. This plugin holds instructions only: run the AEL 1.4.2 engine installed and checked as that guide describes. Respond in the user's language. Work on user-selected inputs and write new output directories.

Read [stacking and extinction recipes](../../docs/DEVELOP.md) and the template at [stack_config_template.json](../../examples/stack_config_template.json). The CLI executes a declared recipe; registration and calibration must be measured beforehand.

- Group by optical train and ISO. Check exposures against metadata and sequence, totality membership, black/white levels and the mosaic's two green planes. Use matched darks/flats if supplied; never invent a flat or call an unvalidated response fit a calibration.
- Preserve linearity: subtract black, apply declared calibration and channel ratios, and avoid gamma or a raw-converter tone curve. Measure the linear ceiling. Separate geometric coverage from saturation/exposure weights.
- Register the corona in the solar frame. The moving Moon is not a fixed registration target. Preserve manual lunar geometry. Reference-to-input affines map native pixel centres; do not add a second half-pixel offset. Correct each optical train's distortion before fusion.
- Keep the user's chosen full canvas. Record a reference or union canvas explicitly; do not choose a crop to hide coverage boundaries. No extrapolation or reconstruction where every exposure clips.
- Statistical hot-pixel rejection requires supported dithering. Clouds require measured spatial transmission/background maps frozen before testing and independently fitted for independent witnesses; one scalar per frame is insufficient.

Run `stack --config RECIPE --out NEW_DIR` with the installed engine (`PYTHON -m ael`), into a new directory. Check the linear result, coverage, support, weights, alpha and geometry at the limb and all boundaries. Keep input hashes and the receipt. The package's variance model is not a complete photometric uncertainty budget.

For differential extinction under a low Sun, read the extinction section and [template](../../examples/extinction_config_template.json). Subtract the sky and use a measured extinction coefficient on the linear stack; sky/extinction degeneracy must be declared. Rebuild dependent layers. See [field lessons](../../docs/FIELD_LESSONS.md) for failed controls and limitations.