Image Integration

The table records the primary exact-path inventory. The list below names the remaining files; together, the table and list are the exhaustive source/test inventory:

• Model/core: src/AfterNightProcesses/processes/ImageIntegration/model/GroupingKeys.h; src/AfterNightProcesses/processes/ImageIntegration/model/GroupingKeys.cpp; src/AfterNightProcesses/processes/ImageIntegration/model/GroupingEngine.cpp; src/AfterNightProcesses/processes/ImageIntegration/model/RegistrationGroupEngine.cpp; src/AfterNightProcesses/processes/ImageIntegration/model/IntegrationGroupEngine.cpp; src/AfterNightProcesses/processes/ImageIntegration/model/FrameMetadata.cpp; src/AfterNightProcesses/processes/ImageIntegration/model/FrameCollection.cpp; src/AfterNightProcesses/processes/ImageIntegration/core/FileImport.cpp; src/AfterNightProcesses/processes/ImageIntegration/core/FilenameUtils.h; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/CalibrationStackInputs.cpp; src/AfterNightProcesses/processes/ImageIntegration/core/StackingParamsUtils.cpp; src/AfterNightProcesses/processes/ImageIntegration/core/NormalizationOptionParsing.cpp; src/AfterNightProcesses/processes/ImageIntegration/core/RejectionMapUtils.h.
• UI builders/dialogs: src/AfterNightProcesses/processes/ImageIntegration/tabs/CalibrationTabUiBuilder.h; src/AfterNightProcesses/processes/ImageIntegration/tabs/RegistrationTabUiBuilder.h; src/AfterNightProcesses/processes/ImageIntegration/tabs/StackingTabUiBuilder.h; src/AfterNightProcesses/processes/ImageIntegration/tabs/SummaryTabUiBuilder.h; src/AfterNightProcesses/processes/ImageIntegration/tabs/BiasCalibrationPanel.cpp; src/AfterNightProcesses/processes/ImageIntegration/tabs/DarkCalibrationPanel.cpp; src/AfterNightProcesses/processes/ImageIntegration/tabs/FlatCalibrationPanel.cpp; src/AfterNightProcesses/processes/ImageIntegration/tabs/LightCalibrationPanel.cpp; src/AfterNightProcesses/processes/ImageIntegration/controls/RegistrationSettingsDialogUiBuilder.h; src/AfterNightProcesses/processes/ImageIntegration/controls/StackingSettingsDialogUiBuilder.h; src/AfterNightProcesses/processes/ImageIntegration/controls/LocalNormalizationSettingsDialogUiBuilder.h; src/AfterNightProcesses/processes/ImageIntegration/controls/StackingRecommendationDialog.cpp; src/AfterNightProcesses/processes/ImageIntegration/dialogs/ManualRegistrationDialog.cpp; src/AfterNightProcesses/processes/ImageIntegration/dialogs/CometTrackingDialog.h; src/AfterNightProcesses/processes/ImageIntegration/dialogs/BlinkComparatorDialog.cpp; src/AfterNightProcesses/processes/ImageIntegration/dialogs/ImportProgressDialog.cpp; src/AfterNightProcesses/processes/ImageIntegration/dialogs/WorkflowProgressDialogV2.cpp.
• Preview/diagnostic/control implementations: src/AfterNightControls/controls/CalibrationPreviewConfig.h; src/AfterNightControls/controls/CalibrationPreviewUtils.h; src/AfterNightControls/controls/ImageViewControl.cpp; src/AfterNightProcesses/processes/ImageIntegration/widgets/GroupingOptionsWidget.cpp; src/AfterNightProcesses/processes/ImageIntegration/widgets/FrameStatisticsDashboard.cpp; src/AfterNightProcesses/processes/ImageIntegration/widgets/FrameGroupingIssuesWidget.cpp; src/AfterNightProcesses/processes/ImageIntegration/widgets/FrameQuickActionsWidget.cpp; src/AfterNightProcesses/processes/ImageIntegration/widgets/CalibrationCoverageWidget.cpp; src/AfterNightProcesses/processes/ImageIntegration/widgets/CalibrationIssuesWidget.cpp; src/AfterNightProcesses/processes/ImageIntegration/widgets/CalibrationPolicyWidget.cpp; src/AfterNightProcesses/processes/ImageIntegration/widgets/CalibrationQuickActionsWidget.cpp; src/AfterNightProcesses/processes/ImageIntegration/widgets/FrameQualityDashboard.cpp; src/AfterNightProcesses/processes/ImageIntegration/widgets/QualityDistributionWidget.cpp.
• Workflow/stages: src/AfterNightProcesses/processes/ImageIntegration/workflow/WorkflowManifest.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/ManifestDiff.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/DirtyStateTracker.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/ManifestHashing.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/ArtifactTracker.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/WorkflowRunStateStore.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/WorkflowExecutionController.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/RegistrationWorkAnalyzer.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/BiasStackingStage.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/DarkStackingStage.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/FlatStackingStage.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/LightsCalibrationStage.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/LightsRegistrationStage.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/LightsAlignmentStage.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/LightsNormalizationStage.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/LightsDrizzleStage.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/LightsStackingStage.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/stages/StageUtilities.cpp.
• Persistence/reporting/engine: src/AfterNightProcesses/processes/ImageIntegration/persistence/AutosaveManager.cpp; src/AfterNightProcesses/processes/ImageIntegration/persistence/IntegrationDatabase.cpp; src/AfterNightProcesses/processes/ImageIntegration/persistence/RegistrationMetricsRepository.cpp; src/AfterNightProcesses/processes/ImageIntegration/reporting/WorkflowReportSectionBuilders.cpp; src/AfterNightProcesses/processes/ImageIntegration/reporting/WorkflowReportArchiveLoader.cpp; src/AfterNightEngine/src/algorithms/cosmeticcorrection/CosmeticCorrection.cpp; src/AfterNightEngine/src/algorithms/stars/StarAlignment.cpp; src/AfterNightProcesses/processes/core/RegistrationRun.cpp; src/AfterNightProcesses/processes/core/LinearChannelMatch.cpp; src/AfterNightEngine/src/algorithms/drizzle/DrizzleCanvas.cpp; src/AfterNightEngine/src/algorithms/drizzle/DrizzleMapDescriptor.cpp; src/AfterNightControls/app/PlateSolveService.cpp; src/AfterNightEngine/src/algorithms/astrometry/solver/PlateSolver.cpp; src/AfterNightProcesses/processes/ImageIntegration/workflow/PreviewThumbnailGenerator.cpp.
• Additional focused tests: src/AfterNightTests/Test_ImageIntegration/unit/test_file_import.cpp; src/AfterNightTests/Test_ImageIntegration/unit/test_grouping_engine.cpp; src/AfterNightTests/Test_ImageIntegration/unit/test_integration_data_model.cpp; src/AfterNightTests/Test_ImageIntegration/unit/test_calibration_matcher.cpp; src/AfterNightTests/Test_ImageIntegration/persistence/test_autosave_recovery.cpp; src/AfterNightTests/Test_ImageIntegration/persistence/test_presets.cpp; src/AfterNightTests/Test_ImageIntegration/workflow/test_calibration_stages.cpp; src/AfterNightTests/Test_ImageIntegration/workflow/test_dirty_state_tracker.cpp; src/AfterNightTests/Test_ImageIntegration/workflow/RegistrationStageTest.cpp; src/AfterNightTests/Test_ImageIntegration/workflow/AlignmentStageTest.cpp; src/AfterNightTests/Test_ImageIntegration/workflow/NormalizationStageTest.cpp; src/AfterNightTests/Test_ImageIntegration/workflow/StackingStageTest.cpp; src/AfterNightTests/Test_ImageIntegration/workflow/test_partial_success.cpp; src/AfterNightTests/Test_ImageIntegration/workflow/test_workflow_cancellation.cpp; src/AfterNightTests/Test_ImageIntegration/workflow/test_workflow_execution_controller.cpp; src/AfterNightTests/Test_ImageIntegration/workflow/test_workflow_run_state_store.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_registration_settings_dialog.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_stacking_recommendation_dialog.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_blink_comparator.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_import_progress_dialog.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_workflow_progress_dialog_v2.cpp.
• Preview/diagnostic surface tests: src/AfterNightTests/Test_ImageViewControl/test_static_preview_behavior.cpp; src/AfterNightTests/Test_ImageViewControl/test_calibration_preview_utils.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_calibration_preview_tooltips.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_frame_source_tab_right_panel.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_frame_statistics_dashboard.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_frame_grouping_issues_widget.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_frame_quick_actions_widget.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_quality_dashboard.cpp; src/AfterNightTests/Test_ImageIntegration/ui/test_quality_distribution_widget.cpp.

Purpose and expected result

The Lights source view shows the populated groups, source controls, calibrated preview, workflow tabs, and selected output directory in one workspace.

The Summary view combines the integration details, retained run history, resource estimate, and report action for the selected project.

Image Integration is AfterNight's persistent, end-to-end project surface for turning still astrophotography frames into calibrated, registered, aligned, normalized, and stacked linear masters. It is intended for multi-frame projects that benefit from one shared frame inventory, consistent grouping and matching, incremental reruns, and a reproducible report.

You load light frames and any available bias, dark, and flat frames, verify the generated groups and assignments, choose registration and stacking settings, review the execution plan, and run it. A successful full run leaves one stacked FITS master per integration group, intermediate products required by the selected options, registration metrics, incremental state, and a run archive with an HTML/Markdown report. Source frames are never modified.

Implementation facts below are distinguished from suggested starting points. In particular, the guide does not claim that a warning-free plan guarantees scientifically good calibration or registration.

At a glance

Item	Current behavior
Menu path	Process > Preprocess > Image Integration. The Dashboard Image Integration card is a second entry point.
Enabled when	Always available as a top-level process; no active image or sequence is required. Execute Integration is enabled only when model validation has no blocking errors.
Inputs	Still FITS, TIFF, PNG, JPEG, camera RAW, and XISF files accepted by the engine image-codec extension check; optional existing FITS/XISF calibration masters; metadata such as dimensions, frame type, filter, object, session, gain/ISO, offset, exposure, temperature, capture time, CFA, WCS, and SIP.
Outputs	Generated master bias/dark/flat FITS files, optional calibrated and CA-corrected lights, registration metrics, registered lights or drizzle descriptors, normalization coefficients, stacked FITS masters, optional rejection/quality maps and RGB/comet products, incremental-state databases/JSON, session notes, run archives, logs, and HTML/Markdown reports. Source files remain unchanged.
Recommended image state	Linear acquisition data. RAW white balance is forcibly disabled and workflow calibration is performed in the raw-before-demosaic domain. Final masters remain linear.
Preview support	The source tabs provide a read-only static frame preview, and the Lights preview adds an off-by-default Calibrated toggle that applies the currently matched, already available bias/dark/flat masters and Light Calibration Settings to the preview only. It can preview a usable subset of calibration; complete coverage is not implied. There is no real-time process preview and no Apply-to-active-image action. Manual Registration, Track Comet, and Blink Comparator provide additional task-specific inspection. Execution writes files.
Sequence support	Unsupported. Import accepts engine-supported still-image extensions, not SER, image-list, or video sequences. Extract frames first.
Mask support	Unsupported. The process is file/project based and does not read the active image's mask.
External dependencies	No external tool is required for basic star matching and stacking. Auto-solving, placing a Mosaic panel star matching cannot register, or solving a final master needs a configured local/online plate solver; Gaia enrichment needs an installed catalog. Network use depends on the chosen solver, and Mosaic alignment calls it whatever Auto-solve unsolved frames says.

Before you start

• Put original frames somewhere stable. Project JSON and autosave records paths; they do not embed image data.
• Supply at least two usable light frames for a normal stack. The UI requires lights and a writable output directory; downstream stages can still fail when too few compatible frames remain.
• Prefer complete metadata. File import takes a frame's type from its header - IMAGETYP, then INDI's FRAME when IMAGETYP names no kind, or an XISF image's type; OFFSET anywhere in the value, as in MasterOffset, names a bias - and only when the header states none from a recognizable directory name (bias, darks, flats, lights, and related plural/offset names); a light frame in a folder named darks stays a light. Unknown auto-sorted files fall back to the current source type. A frame's session is the UTC date of its exposure start (DATE-OBS), or of the time the file was written (DATE) when it states no start, or of the file's modification time when it states neither.
• Confirm that frames intended for one registration group share usable geometry and channels. The current registration key is resolution plus effective object name; channel count and binning are not part of that key.
• Decide whether calibration masters should be built or supplied. Marking one included calibration file as a master makes that group use the file directly and skip master stacking.
• Select a dedicated writable output directory with enough free space. Existing products can be staged and replaced; the previous copies go into run history when the run commits.
• Treat changes to source grouping carefully. Regrouping changes group keys and clears or remaps group-specific settings and selections.
• For CFA/RAW data, verify the CFA pattern and demosaic method in Registration settings. The workflow's raw load policy uses no white balance before calibration.
• Configure plate solving in Settings before enabling Auto-solve unsolved frames during Registration, choosing Mosaic alignment - which sends the panels star matching cannot register to the configured solver whatever Auto-solve says - or relying on Plate-Solve Final Master If Needed.
• Capture requests are pending; screenshots on this page are not runtime evidence.

Procedure

1. Open Process > Preprocess > Image Integration or the Dashboard card. Confirm that the window shows Frame Sources and Workflow, with source tabs Bias, Darks, Flats, and Lights.
2. Add files with the per-tab Add Files.../Add Folder... buttons or the Frames menu. Use a typed action when the files must be treated as that type; use Auto-sort by file type... when metadata/folder detection should decide. The checkpoint is a populated tree with the expected file types and no import issue you intend to fix.
3. In each source tab, inspect metadata, inclusion checkboxes, and grouping. Use Hierarchical to inspect group membership and Flat to inspect individual files. In Lights, enable Coverage to isolate lights without calibration coverage. After a matching generated or existing master is available, select a light and toggle Calibrated to compare its raw and calibrated static previews; the toggle never modifies the source. The checkpoint is that each intended frame belongs to the correct source type and group.
4. Choose Output Directory with the ... button or type a path. The Summary status must no longer report an invalid or unwritable output path.
5. Open Workflow > Calibration. Review master sources, matching assignments, fallback warnings, Matching Policy, and per-light calibration settings. A missing calibration assignment is a warning when calibration frames exist, not automatically a blocker; decide whether that is intentional.
6. Open Workflow > Registration. Select a registration group, open Registration Settings, choose the method/reference/transform/alignment/interpolation, and configure manual points or comet observations if that method requires them. Run registration early if you need quality metrics, photometric weighting, or drizzle recommendations. The checkpoint is a group status of Complete or usable Partial with metrics for retained frames.
7. Review Frame Quality and the optional Quality Filters. Exclude only frames you have inspected. Global exclusion removes a frame from registration and every stack; the Stacking-tab dashboard can exclude it from only the selected stack.
8. Open Workflow > Stacking. Verify mandatory grouping by resolution, filter, and object; optionally split by session, custom tag, or exposure. Select each group and open Stacking Settings... to choose normalization, rejection, weighting, maps, and drizzle. The checkpoint is one intended output row per final master.
9. Review Execution Options. Leave in-memory light calibration enabled unless you need intermediate calibrated files or the workflow selects its low-memory fallback. Set CPU Threads and a memory limit appropriate for the machine.
10. Open Workflow > Summary. Resolve every error reached through Go to and review warnings. Confirm the Integration Summary, estimated disk use, and output-specific Session Notes.
11. Select Execute Integration or press Ctrl+Enter. In Execution Plan, inspect Stage, Group, Decision, Reason, Dependencies, ETA, and Outputs. Recompute performs work, Reuse keeps valid work, and Skip means the stage is not applicable. Select Run only when the plan matches the intended scope. A registration_metrics.db in a format this version does not read is never changed: the run does not start and the window says why.
12. Monitor the stage timeline, current operation/file, progress, and Process Output. Cancel is cooperative; it changes to Cancelling... while the worker reaches a safe stop.
13. On completion, use Open Report, Run History > Report, or Run History > Open All. Confirm the expected master files and report. A completed-with-warnings run can contain valid products plus recoverable per-frame/group failures; read the warnings before using the masters.

Complete parameter and action reference

Persistence abbreviations used below:

• Model — saved by the five-second app-local autosave and by JSON project export/import.
• Preset — included in configuration-only presets when it belongs to grouping, calibration policy/overrides, registration, stacking, or light-calibration settings; frames and output directory are not included.
• UI — saved separately in QSettings as view layout state.
• Run — only the current open window/run unless an output artifact records it.
• Artifact — written under the selected output directory and available to later runs.

Window, project, and source-frame actions

UI label/action	Type and unit	Runtime default	Persisted-default exception	Valid values, step, or constraints	Effect of increasing, decreasing, or selecting	Dependencies and enable/visibility rules	Quality, artifact, runtime, memory, or disk trade-off	Scope: preview/apply, channel, mask, still/sequence	Persistence
Frame Sources / Workflow	Main tab selector	Frame Sources	Last logical tab is not restored as model data	Frame Sources contains Bias/Darks/Flats/Lights; Workflow contains Calibration/Registration/Stacking/Summary	Changes the visible editor only	Always visible	None	Project UI only	Run
Output Directory / ...	Path field / folder chooser	Restored autosave value or empty	New state is empty	Existing writable directory, or a creatable path whose parent is writable	Sets the root for all workflow products, state, notes, and history	Execution blocks on empty/invalid/unwritable paths	Long paths and slow disks affect every stage	File workflow; no mask/sequence	Model; not Preset
Add Files...	Per-tab button	—	—	Supported still files; multiple selection	Adds files as the current tab's type	Available on every source tab	Metadata reads can take time for large selections	Still files only	Model
Add Folder...	Split button	Non-recursive default action	—	This Folder Only or Recursive	Imports supported still files from the chosen tree as the current type	Available on every source tab	Recursive scans and metadata reads scale with file count	Still files only	Model
Frames > Add Files... > Add Bias... / Add Darks... / Add Flats... / Add Lights...	Menu actions	—	—	Supported still files	Forces the selected type even when metadata differs	Always available	Forced type can create incorrect groups if chosen wrongly	Still files only	Model
Auto-sort by file type...	Menu action	Ctrl+O for files; Ctrl+Shift+O for recursive folder	—	Supported still files	Uses the header's frame type first, then directory-name detection; unknown files fall back to the current type	File, this-folder, and recursive variants	Incorrect/missing metadata may need Change Type to	Still files only	Model
Search frames...	Text filter	Empty	Search text is not model data	Matches filename, path, filter, object, session, resolution, tag, exposure, gain, offset, and temperature	Narrows visible rows; does not exclude processing	Ctrl+F; source tabs only	No workflow effect	View only	Run
Hierarchical / Flat	Radio display mode	Flat	Stored independently per source tab	Two modes	Hierarchical shows grouping nodes; Flat shows files	Source tabs only	No workflow effect	View only	UI
Frame tree checkboxes	Include/exclude action	Included	Loaded state may restore exclusions	Checked/unchecked; groups can be mixed	Unchecked frames are globally excluded from calibration, registration, and stacking	Tree rows	Reduces data and runtime but can weaken rejection/registration	Whole still frame; all channels	Model; undoable
Coverage	Lights-only filter toggle	Off	Stored independently for Lights	On/off	Shows lights missing required calibration coverage	Lights tab only	Diagnostic only	View only	UI
Frame preview	Read-only image view	Flats/Lights auto-stretched; CFA auto-debayered for display	Display state is not processing state	Zoom/pan/view controls supplied by the image viewer	Inspects the selected source; does not change workflow pixels	One selected frame	Preview demosaic/stretch/calibration is not an output guarantee	Selected still only; no mask	UI/view cache
Calibrated	Lights-preview checkbox	Off	Not persisted	On/off	Reloads the selected Light preview after applying the currently matched bias/dark/flat master paths and that group's cosmetic-correction, dark-optimization, flat-auto-level, pedestal, and CCD-bias settings. The raw file is unchanged.	Visible only on Lights. Enabled for a selected non-Pre-Calibrated light when at least one usable matched master remains and every selected master path already exists. A missing or manually unassigned individual type is omitted, so the preview can be partial; normal dark use suppresses bias, while CCD mode with a Dark requires Bias. No usable calibration, a missing selected master file, or the CCD Dark-without-Bias case disables it with an explanatory tooltip.	Loads/caches masters and runs a background static-preview calibration; useful for inspection but not proof of the eventual batch output	Selected still preview only; before viewer statistics/histogram/autostretch; no Apply/mask/sequence	Run/view cache only
Remove Selected	Button/menu action	—	—	One or more selected rows; Delete	Removes selected records from the project, not disk	Current source tab	Can invalidate group settings/assignments	Project records only	Model; undoable
Remove All / Remove All Frames...	Button/menu action	—	—	Confirmation; menu shortcut Ctrl+Shift+Delete	Per-tab button clears that type; menu clears all four types	Requires loaded frames	Does not delete files	Project records only	Model; undoable
Go to Group...	Search/jump action	—	—	Ctrl+G; current tree groups	Selects and reveals a matching group	Current source tab	No workflow effect	View only	Run
Frame context actions	Context menu	Selection-dependent	—	Copy File Path, Include in Processing/Exclude from Processing, Show in File Browser, Show in Calibration Table, Lights: Show in Registration Table/Show in Integration Table, Blink Selected, Remove Selected, Change Type to, Set as Master/Unset as Master, Lights: Set as Registration Reference	Performs the named navigation, inclusion, type, master, reference, compare, or removal action	Actions appear only for applicable selection/type; Blink needs at least two files	Type/master/reference changes can invalidate downstream work	Selected still records	Model for data changes; Run for navigation
Metadata override context actions	Context submenu	Metadata value	—	Set/reset Session, Object, Custom Tag; reset Session + Object; reset all overrides	Replaces effective grouping metadata without editing source headers	Selected frames/groups; prompts for values/confirmation	Can split/merge groups and invalidate per-group settings	Selected still records	Model; undoable
Group context actions / drag-and-drop	Context/drag actions	—	—	Show matching calibration table; expand/collapse; include/exclude/remove group; change mixed selection type; drag between groups	Group moves create metadata overrides after confirmation	Hierarchical tree and compatible targets	Regrouping may invalidate settings and assignments	Project records	Model; undoable
Frame Statistics	Lights-only read-only dashboard	Derived; empty until Lights are loaded	—	Total, Included, Excluded, Groups, Exposure, Size, percentage chart; hover for per-group metadata/details	Summarizes the current Light collection and grouping without changing it	Lights right panel	Diagnostic only; totals update after model/grouping changes	Current Light records; no processing scope	Derived from Model
Issues (n)	Lights-only read-only issue list	No issues detected. when clear	—	Up to five displayed issue types plus an additional-count label and tooltips	Reports missing/fallback/manually disabled calibration, single-frame groups, mixed dimensions, high exclusion ratios, and total exclusions	Lights right panel; derived from validation/grouping	Diagnostic only; use the named source/calibration controls to resolve an issue	Current Light groups	Derived from Model
Include All, Exclude All, Clear Tags	Lights Quick Actions	—	—	Buttons; Clear Tags confirms	Bulk inclusion or custom-tag clearing	Lights right panel	Excluding all lights blocks execution	All light records	Model; undoable
Import Progress: Pause/Resume, Cancel	Non-modal progress dialog	Running	—	Pause toggles; Cancel is cooperative	Pauses worker admission or cancels remaining imports	Large/asynchronous imports	Already completed detection work may have occurred; cancelled import is not committed	Import only	Run
Import Project...	Project menu action	—	—	A project exported as JSON	Replaces model state from the JSON project, read whole: a project holding a value no dialog writes is refused with the field and the text named (“The project file could not be read, and nothing was loaded.”)	Ctrl+I	Any other file is refused with a message, and nothing is loaded	Project state	Model; recent path in UI
Export Project...	Project menu action	—	—	JSON output; Ctrl+E	Atomically writes the complete model, including frames/settings/output path	Writable destination	Path-based project, not portable embedded data	Project state	JSON artifact; recent path in UI
Open Recent > Clear Recent Projects	Project menu actions	Up to 10 paths	Missing entries are removed when opened	Recent JSON paths	Opens a project or clears list	Recent entries only	No image data copied	Project state	UI (QSettings)
Presets > Save Preset... / Load Preset... / Delete Preset...	Project menu actions	No presets	A preset holding a value no dialog writes is refused with the value named, and one in a format this version does not read is refused with a message; either is left as it is	Sanitized JSON filename, max 100 characters; overwrite/delete confirms	Saves/applies/deletes configuration only	Preset directory must be writable	Loading is undoable; frames and output directory stay unchanged	Configuration only	Preset JSON under app-local presets
Reset Settings... / Reset	Confirmed action	—	—	Keeps loaded frames and output directory	Resets grouping, calibration, registration, stacking, overrides, and other configuration	Menu or standard Reset button	Destructive to configuration, not files	Project settings	Model; autosaved afterward
Undo / Redo	Menu actions	Empty stack	Stack limit 100 commands	Standard shortcuts supplied by Qt	Reverses supported model edits, including preset load	Enabled when stack has entries	Does not reverse completed disk workflows	Project edits	Run undo stack
Execute Integration	Menu/button action	Disabled until ready	—	Ctrl+Enter; opens Execution Plan	Builds, validates, plans, then executes after Run	Requires no blocking model/manifest error	Potentially long and disk-intensive	Whole project or requested group	Creates Artifacts
Navigate > Bias ... Summary	Menu actions	—	—	Ctrl+1 through Ctrl+8	Opens the corresponding logical tab	Always available	No workflow effect	View only	Run
Keyboard Shortcuts	Help action	—	—	F1/Help	Opens shortcut reference	Always available	No workflow effect	Help only	None

Source grouping controls

Resolution is always part of every source group and cannot be disabled. Hidden criteria are normalized out by frame type.

UI label/action	Type and unit	Runtime default	Persisted-default exception	Valid values, step, or constraints	Effect of increasing, decreasing, or selecting	Dependencies and enable/visibility rules	Quality, artifact, runtime, memory, or disk trade-off	Scope: preview/apply, channel, mask, still/sequence	Persistence
Group By: Resolution	Mandatory criterion	On	None	Exact width × height string	Separates incompatible dimensions	Always applied; not editable	More groups but safer geometry	Current source type	Model + Preset
Filter	Criterion	On for Flat/Light	Hidden/off for Bias/Dark	On/off	Separates filters	Flat and Light only	Prevents unlike flats/lights from mixing	Current source type	Model + Preset
Object	Criterion	Off for Flat; on for Light	Hidden/off for Bias/Dark	On/off	Separates effective object names	Flat/Light only	Flat grouping by object is optional; Light registration separately always uses object	Current source type	Model + Preset
Session	Criterion	On for Flat/Light	Hidden/off for Bias/Dark	On/off	Separates effective sessions	Flat/Light only	More groups and stricter same-session calibration	Current source type	Model + Preset
ISO/Gain	Criterion	Off	None	On/off	Separates sensor gain/ISO	All source types	Safer sensor matching; more groups	Current source type	Model + Preset
Offset	Criterion	On internally	Effective only with ISO/Gain	On/off	Separates sensor offset	Enabled when ISO/Gain is selected	Safer matching; more groups	Current source type	Model + Preset
Exposure / Tol (s)	Criterion / seconds	Dark: on, 3.0 s; Light: off, tolerance inactive	Off/hidden for Bias/Flat	Toggle; 0–36000 s on source tabs, step 0.1; 0 means exact	Larger tolerance merges nearby exposures; smaller splits them	Dark/Light only; tolerance enabled with Exposure	Wider dark groups improve counts but can mismatch thermal signal	Current source type	Model + Preset
Temperature / Tol (°C)	Criterion / Celsius	Dark: off, tolerance inactive	Hidden for Bias/Flat/Light	Toggle; 0–10 °C, step 0.1; 0 means exact	Larger tolerance merges nearby sensor temperatures	Dark only	Wider groups improve counts but can mix thermal states	Dark source	Model + Preset
Custom Tag	Criterion	Off	None	On/off; user text from metadata override	Separates manually tagged groups	All source types	Useful for optical train or acquisition distinctions absent from headers	Current source type	Model + Preset

Grouping changes are debounced before rebuilding the tree. They can change group keys, so review master/reference assignments and group settings afterward.

Calibration tab and settings

UI label/action	Type and unit	Runtime default	Persisted-default exception	Valid values, step, or constraints	Effect of increasing, decreasing, or selecting	Dependencies and enable/visibility rules	Quality, artifact, runtime, memory, or disk trade-off	Scope: preview/apply, channel, mask, still/sequence	Persistence
Bias Groups, Dark Groups, Flat Groups, Light Groups	Collapsible tables	Derived from sources	—	One row per current group	Selecting a row cross-highlights matching or affected groups. The Dark table's Bias Master column shows - for every group, its tooltip saying why: no bias is subtracted when darks are stacked, and CCD mode subtracts it from the master dark where a light is calibrated	Calibration tab	Diagnostic/edit surface	Project groups	Derived from Model
Calibration Coverage	Read-only coverage chart	Derived; reports that coverage is not computed when no calibration frames are loaded	—	Complete, Warning, Override, and Incomplete Light-group counts/percentage	Summarizes current automatic/manual coverage; Pre-Calibrated groups are omitted	Calibration tab; Light groups required	Diagnostic only; fallback matches count as Warning when Warn on fallback matches is on, manual choices count as Override, and missing required matches count as Incomplete	All Light groups	Derived from Model
Issues (n)	Read-only grouped issue list	No calibration issues detected. when clear	—	Up to five displayed issue types plus an additional-count label and per-group tooltips	Groups current calibration validation warnings/errors by code and affected group	Calibration tab	Diagnostic only; it does not change assignments	Calibration validation	Derived from Model
Calibration master table cells	Assignment/source controls	Auto-generated unless an existing master is selected	—	Auto-detect, Browse..., Do not use, candidates, batch assignment; existing master status	Chooses a generated, detected, user-defined, or disabled master	Available by calibration type and dependent group	Wrong master produces systematic calibration error; existing master saves time	Group-wide	Model + Preset overrides
Stacking Settings...	Group dialog action	Type defaults	—	Opens calibration-mode Stacking Settings	Configures how a master is built	Selected Bias/Dark/Flat group; light-only controls hidden	Strong rejection needs enough frames	Selected calibration group	Model + Preset
+ Add Master	File action	—	—	FIT/FITS/XISF chooser in group panels	Adds and marks an existing master for the group	Calibration groups	Avoids rebuilding; source must be compatible	Group-wide	Model + Preset override
Smart Match	Light-panel action	Current automatic matches	—	Re-evaluates all light assignments	Applies current grouping/policy to lights	Light groups	Can replace manual assumptions; inspect results	All lights	Model-derived matches
Calibration Settings...	Light-group dialog action	Cosmetic on; dark optimization off; flat auto-level on; pedestal 0	—	Opens settings below	Configures per-light calibration	Selected light group	Affects calibrated pixels and every downstream artifact	Group-wide, all channels/CFA samples	Model + Preset
Pre-Calibrated	Light-group checkbox/action	Off	—	On/off	Skips bias/dark/flat/cosmetic calibration for that group and suppresses missing-calibration warnings; the run's manifest names the group and the report's Calibration Summary lists it, so a pre-calibrated group is not read as one no master matched	Light groups	Prevents double calibration; wrong use leaves defects uncorrected	Group-wide	Model + Preset
Matching	Light-group instant-popup menu	Mirrors the current matching policy	—	Allow dark exposure fallback; Dark exposure tolerance... (n%); Advanced: Subtract bias from darks (CCD sensors); Allow flat session fallback; Max flat session days apart... (n); Prefer newer flats; Warn on fallback matches; Reset matching policy to defaults; Clear all calibration overrides. The last two actions confirm.	Edits the same policy and overrides exposed by Matching Policy and Quick Actions. Its reset action resets policy and clears every override, unlike the separate Reset Matching Policy quick action, which preserves overrides.	Dark fallback/tolerance are disabled unless Dark grouping uses Exposure. Flat fallback/day/preference actions are disabled unless Flat grouping uses Session.	Immediate project-wide rematching can change masters and downstream dirty state	All Light matching	Model + Preset; undoable
Allow flat session fallback	Matching-policy checkbox	On	—	On/off	Allows closest compatible flat session when exact session is absent	Calibration frames/lights present	Recovers coverage but can use a temporally different optical state	All light matching	Model + Preset
Max session gap (days)	Integer days	30	—	0–365	Larger permits more distant flat sessions	Flat fallback on	Wider fallback raises dust/rotation mismatch risk	All light matching	Model + Preset
Prefer newer flats	Checkbox	Off	—	On/off	Breaks equally distant fallback ties toward newer flats	Flat fallback on	Tie-break only	All light matching	Model + Preset
Allow dark exposure fallback	Checkbox	On	—	On/off	Permits closest compatible dark exposure	Calibration frames/lights present	Improves coverage but may leave exposure-dependent residuals	All light matching	Model + Preset
Exposure tolerance (%)	Integer percent	20%	—	0–100	Larger permits a wider dark/light exposure mismatch	Dark fallback on	More coverage, less exact dark subtraction	All light matching	Model + Preset
Warn on fallback matches	Checkbox	On	—	On/off	Shows fallback and ambiguous multiple matches as warnings; disabling it does not stop the selected match from being used	Fallback/ambiguity diagnostics	Diagnostic only	Summary/issues	Model + Preset
Subtract bias from dark (CCD mode)	Checkbox	Off	—	On/off	Requires both bias and dark; subtracts bias from the dark before light calibration, once: a master dark the run stacks holds the bias, as a supplied one should, so the dark factor or optimization scales the thermal signal alone	CCD workflow; the Light Matching menu labels the same setting Advanced: Subtract bias from darks (CCD sensors)	Correct for uncalibrated dark masters; inappropriate for already bias-corrected masters	All matched lights	Model + Preset
Clear All Overrides	Quick action	—	—	One click; no confirmation	Removes manual master assignments/unassignments	Calibration tab	Returns to policy matching	All groups	Model + Preset; undoable
Reset Matching Policy	Quick action	—	—	One click; no confirmation	Restores policy defaults above but preserves manual overrides	Calibration tab	Changes fallback results	All groups	Model + Preset; undoable
Disable All Fallbacks	Quick action	—	—	One click	Disables flat/dark fallback and fallback warnings	Calibration tab	Produces stricter missing-coverage warnings	All groups	Model + Preset
Enable cosmetic correction	Checkbox	On	—	On/off	Finds hot and cold pixels in the master dark against each dark channel's whole-image median and population standard deviation, and replaces them in each calibrated light with the median of the pixels around them - of the same colour in a filter-array mosaic; only the replacement is local	Light Calibration Settings; the pixels are found in the dark master, so a light with no dark has none corrected, which the run's warnings say once per calibration, in memory or on disk	Detection/preparation costs CPU; aggressive thresholds can alter real compact signal	Selected/all light groups	Model + Preset
Cold Sigma (σ) / Hot Sigma (σ)	Sigma thresholds	3.0 / 3.0	—	0.1–10.0, step 0.1	A dark sample at or below median - Cold × deviation of its channel is cold, and one at or above median + Hot × deviation is hot, the median and deviation being the whole channel's; lower detects more, higher is more conservative	Enabled with cosmetic correction	Too low can affect faint detail/bright stars; too high leaves defects	Selected/all light groups	Model + Preset
Enable dark optimization	Checkbox	Off	—	On/off	Requests automatic dark scaling by noise minimization	Dark master required	Engine search adds work and may compensate moderate mismatch	Selected/all light groups	Model + Preset
Multiplication Factor	Scalar	1.00	—	0.1–10.0, step 0.1	Multiplies the dark before it is subtracted whenever the dark is not optimized	Always enabled: it scales every light's dark with optimization off, and the dark of each light the optimization cannot measure with it on	With optimization on, a single-channel light of at least 512×512 gets the engine's own factor on 0–2 over its central 512×512 square and ignores this one; a colour (multi-channel) light, one smaller than 512×512, or one whose dark does not vary over that square, is scaled by the entered value instead, and the run's warnings say so once per calibration, in memory or on disk. With optimization off the entered value is applied as a fixed scale.	Selected/all light groups	Model + Preset
Enable flat auto-level	Checkbox	On	—	On/off	Divides each channel of the master flat - each colour of a filter-array mosaic - by its mean over the central third of the frame before the light is divided by it, so the light keeps its level and its colour balance there; off divides by the flat as supplied	Flat master used	Usually required unless the supplied master is already normalized	Selected/all light groups	Model + Preset
Pedestal Value	ADU integer, read as 16-bit codes	0	—	0–10000, step 1	Adds a positive offset before subtraction/division; the value is divided by 65,535 to a fraction of full scale whatever the light's own bit depth	Always available	Can prevent low-end clipping; changes baseline; added as stated and not clamped to the light's range	Selected/all light groups	Model + Preset
Reset to Defaults, Apply to All Groups, OK, Cancel	Dialog actions	—	—	Apply-to-all confirms	Reset affects dialog values; Apply-to-all overwrites all Light groups; OK commits selected group; Cancel discards	Calibration Settings dialog	Apply-to-all has broad scope	Settings only	Model after commit

Exact calibration-matching contract

Manual override and explicit Do not use state are checked before automatic matching for the exact Light group key. Automatic matching then applies these rules:

Master	Exact compatibility	Multiple exact candidates	Fallback selection
Bias	Resolution always; ISO/Gain when enabled for Bias; Offset only when both ISO/Gain and Offset are enabled; Custom Tag when enabled	Uses deterministic group-key order, marks the result ambiguous, and shows a warning when Warn on fallback matches is on	None
Dark	Resolution always; optional ISO/Gain, dependent Offset, Custom Tag, and Exposure. Exposure is compared after each value is snapped to the nearest configured Dark exposure-tolerance interval. Temperature is deliberately not a compatibility gate.	Prefers a finite absolute Light/Dark temperature difference; improvements greater than 0.25 °C replace the current candidate, while differences within 0.25 °C are tied by deterministic group-key order. With a Light temperature, only finite tied candidates make the result ambiguous; missing Dark temperatures do not. Without a Light temperature, any multiple compatible candidates are ambiguous.	Attempted only when Allow dark exposure fallback is on and Dark grouping uses Exposure. Other enabled fields must remain compatible; exposure difference must be no greater than Light exposure × tolerance percent. The matcher minimizes exposure difference with a 0.000001 s tie epsilon, then prefers finite temperature improvements greater than 0.25 °C, then group key. With a finite selected temperature, ambiguity includes exposure-tied candidates within 0.25 °C of it; otherwise every exposure-tied candidate counts.
Flat	Resolution always; optional ISO/Gain, dependent Offset, Filter, Object, Session, and Custom Tag	Uses deterministic group-key order, marks the result ambiguous, and optionally warns	Attempted only when Allow flat session fallback is on and Flat grouping uses Session. Light and Flat sessions must both parse exactly as yyyy-MM-dd; invalid dates do not fall back. Other enabled fields must match, except an empty/- Flat object acts as a wildcard. The closest date within Max session gap (days) wins; an equal-distance tie chooses newer when Prefer newer flats is on, otherwise older, then group key.

Dark temperature grouping controls how Dark frames form source groups; it does not impose a maximum temperature distance during Light matching. Likewise, Warn on fallback matches changes presentation, not which automatic candidate is selected.

Calibration matching is asymmetric: a matched dark normally takes precedence over bias for a light. In CCD mode both are used so the bias is removed from the dark first. A flat group's dark is matched as a light's of the flat's exposure and temperature would be, in the run as in the Flat panel: Bias is not used when a Dark master is available, because the dark holds the bias, so the flat takes the dark alone - in CCD mode too, where taking the bias from the flat and from the dark leaves the same flat; with no matching dark it takes the bias. A calibration group is stacked when a light shares its key or a light or flat of the run is matched to it, so a flat-dark at the flats' exposure and a dark the exposure fallback finds for a light are made too. When no calibration frames of any type are loaded, lights-only processing does not produce missing-calibration warnings. Missing/fallback/manual-unassignment warnings do not by themselves block execution.

Registration tab, quality, and registration settings

UI label/action	Type and unit	Runtime default	Persisted-default exception	Valid values, step, or constraints	Effect of increasing, decreasing, or selecting	Dependencies and enable/visibility rules	Quality, artifact, runtime, memory, or disk trade-off	Scope: preview/apply, channel, mask, still/sequence	Persistence
Registration Groups	Table	Resolution + effective object groups	—	Columns: Group, Frames, Filters, Exposures, Alignment, Algorithm, Reference, Status	Selects the group for settings/run actions	Included lights only	One group shares one transform/reference contract	Group-wide	Derived; settings in Model
Registration Settings	Button	—	—	Opens selected group's dialog	Edits method and detector settings	One group selected	Settings changes invalidate registration and dependent stacks	Group-wide	Model + Preset
Group context: Registration Settings..., Run Registration, Clear Registration..., Select All Frames	Context actions	—	—	Clear confirms	Opens settings, performs a group run, deletes the group's registration records (star detections, metrics, transforms and comet records in registration_metrics.db; the registered image files stay), or selects rows	Group row	Clearing forces recomputation	Group-wide	Artifacts/Model
Run Registration / Run Registration (All Groups)	Action	—	—	Selected group or all groups	Builds a registration-only/incremental plan	Valid group(s)	Produces metrics needed by quality, weights, and drizzle	Registration and dependencies	Artifacts
Set as Registration Reference	Button/context action	Auto-reference by default	—	One included frame	Fixes the selected reference for its registration group	One frame selected; auto-reference may be disabled by method	Reference quality affects every transform	Group-wide	Model + Preset
Exclude Selected, Include Selected, Reset Excluded...	Global frame actions	Included	—	Selection; Reset confirms	Changes global inclusion for registration and stacking	Frame Quality table	Fewer frames reduce work but weaken integration	Whole frames	Model; undoable
Quality table	Metrics table	Empty before registration	—	Status, Frame, Filter, Stars, FWHM (px), SNR, Ecc, Score, Quality	Sorts and exposes measured registration quality	Metrics required	Metrics are inputs to selection and weighting, not a guarantee of image quality	Frame records	registration_metrics.db + Model cache
Score ≥	Enabled threshold / score	Off, 0.00	—	0–1000, step 1	Higher rejects more low-score frames	Checkbox enables editor	Aggressive threshold reduces exposure	Global exclusion preview/action	Run until applied; exclusions Model
Stars ≥	Enabled threshold / count	Off, 0	—	0–100000	Higher rejects sparse detections	Metrics required	Can reject narrowband or cloudy frames	Global exclusion	Run/Model
SNR ≥	Enabled threshold	Off, 0.00	—	0–100000, step 1	Higher rejects low-SNR metrics	Metrics required	Can bias toward brighter backgrounds/filters	Global exclusion	Run/Model
FWHM ≤	Enabled threshold / pixels	Off, 0.00	—	0–100, step 0.1	Lower rejects broader stars	Metrics required	Tight threshold reduces exposure	Global exclusion	Run/Model
Ecc ≤	Enabled threshold	Off, 0.00	—	0–1, step 0.05	Lower rejects elongated stars	Metrics required	Tight threshold can reject tracking/optics variations	Global exclusion	Run/Model
Suggest Filters... / Exclude Outliers...	Analysis/apply actions	—	—	IQR-based suggestion; apply confirms	Fills suggested thresholds, then globally excludes failures	Selected group with metrics	Review visually before applying	Frame records	Suggestions Run; exclusions Model
Frame Quality dashboard: Sigma, Detect Outliers, Exclude Selected Frames, Include Selected Frames	Statistical preview/actions	3.0σ	—	Sigma 1.0–5.0	Lower selects more score outliers; actions commit global or current-stack exclusions according to dashboard mode	Metrics required	Statistical outlier is not automatically a bad exposure	Registration: global; Stacking: selected stack only	Selection Run; exclusions Model
Quality chart: Metric, Plot, Sort By, Norm, overlays, Export	Visualization/export controls	Score; Points; Index; normalization Off	—	Metric: Score/Stars/FWHM/SNR/Ecc; plot: Points/Line/Line+Points/Area/Bars; sort: Index/Value asc/desc/Capture Time; Norm: Off/By Filter; percentile/median/selection/drop-to-baseline toggles; copy/save image/CSV	Changes only diagnostic rendering/export	Metrics required	Large tables/charts add UI cost, not workflow cost	View/export only	Run
Registration Method	Enum	Star Matching	—	Star Matching, Phase Correlation, Manual Points, Comet Tracking	Selects feature matching, translation-only correlation, user point pairs, or moving-target fit	Method controls visible sections/reference behavior	Phase correlation cannot model rotation/scale; manual/comet require preparation	Selected group	Model + Preset
Minimum Star Pairs / Maximum Stars per Frame	Counts	10 / 2000	—	5–10000 / 10–20000	Minimum raises acceptance requirement - the star pairs a light must share with the reference or a light registered near it, however small a part of its stars they are; a match of fewer than ten candidate pairs also needs four fifths of them to agree; a reference holding fewer stars than the minimum lowers it to its own stars for the run, never below five, and the log says so - a reference of four stars or fewer cannot match its lights by their stars, which it says too; maximum caps the stars detected in each frame, which are the candidates for matching, and a changed cap detects the group's lights again	Star-based methods	More pairs can improve robustness and cost more	Group	Model + Preset
Auto-select	Checkbox	On for Star Matching	—	On/off	Performs first-pass quality analysis and chooses the best reference, which the group's record names	Disabled for methods requiring a fixed reference	Adds analysis but avoids an arbitrary reference	Group	Model + Preset
Reference Image / Best	Frame selector/action	First/auto candidate	—	Included frames in group	Chooses fixed reference; Best analyzes candidates; with no reference chosen and auto-select off, the group's first light is the reference, which the log names as "(first in group)"	Enabled when auto-select is off or method requires fixed reference	Reference controls output geometry/relative quality	Group	Model + Preset
Auto-solve unsolved frames during Registration	Checkbox	Off	Under Mosaic it shows on, disabled, with a note: Mosaic sends every panel star matching cannot register - and the reference it is placed against, when its header holds no solution - to the configured solver whatever the choice is. The choice the group stores is kept and comes back with another canvas	On/off	Reuses valid WCS and solves unsolved inputs before registration	Enabled only when a configured solver is available - a ready Gaia catalog, an executable ASTAP or local astrometry.net, or an online key, as the run counts them - and not under Mosaic. Without one it shows off and the run sends nothing to a solver, whatever the group stores, and a light only a solve could place is left unregistered with that reason; the stored choice is kept	Local/online solver cost; online mode can use network, which Mosaic can reach with this off	Group	Model + Preset
Enrich with Gaia catalog	Checkbox	On	—	On/off	After registration, matches the stars of each light detected in the run whose world coordinates are known - its own or solved - against the Gaia catalog, for its photometric zero point, limiting magnitude and catalog-matched star count, which photometric weighting and spectrophotometric calibration read	Offered with or without a catalog; without one installed, a group with such a light enriches nothing and says so once in its warnings, naming the catalog directory	Adds catalog reads per light; turning it off leaves photometric weighting without zero points	Group	Model + Preset
Transform Model	Enum	Auto - a Homography, refined by a polynomial when that is measurably better (stored as mode Auto over a Homography, order 0)	A stored choice reads by its mode: an order stored beside a linear mode is dropped, and two-pass refinement is on for Polynomial and Auto	Similarity, Affine, Homography, Polynomial ⅔/4, Auto	Increases geometric degrees of freedom; Auto keeps a linear model unless polynomial residual refinement is measurably better; a stated polynomial needs as many star pairs as it has terms - 6, 10 and 15 for orders 2, 3 and 4 - and a light matched by fewer, or whose pairs the order's fit refuses, keeps the linear transform its stars matched, which the log says for that light; under Manual Points a target with fewer pairs than the stated order's terms fails, with Manual registration failed for that light	Star/manual registration	Higher order can fit distortion but overfit weak matches	Group	Model + Preset
Standard, Mosaic, Intersection	Alignment radio mode	Standard	—	One mode	Standard uses reference canvas; Mosaic uses union canvas, and places a panel star matching cannot register by its world coordinates - its own, or the configured plate solver's, which Mosaic calls whatever Auto-solve unsolved frames says; Intersection crops to common overlap - and when no area is common to every light, the alignment, the stack that warps the lights as it reads them and the drizzle all use the Standard canvas instead and say so	All methods	Mosaic uses more memory/disk, and can send panels to an online solver; Intersection discards borders	Group outputs	Model + Preset
Interpolation	Enum	Auto	—	Auto, Nearest Neighbor, Bilinear, Bicubic, Pixel Area Relation, Lanczos4, Bit Exact Bilinear, Bit Exact Nearest Neighbor	Selects resampling kernel; Auto is Lanczos-4, and the log names it	Alignment	Sharper kernels cost more and can ring; nearest is fast/blocky	All channels	Model + Preset
Correct chromatic aberration before registration	Checkbox	Off	—	On/off	Aligns red/blue star positions to green before main registration	Colour lights - RGB, or a filter-array mosaic, which registration reads in colour - and enough edge stars	Adds detection/resampling. A group whose every light has one channel is registered from its lights as they are, writes no copy and says once that nothing was corrected (copies an earlier run left are kept current, since every later stage reads them). A colour light whose channels match too few edge stars is written as it was read, and a group in none of whose lights a channel moved says so once	Group	Model + Preset; outputs ca_corrected_lights
Edge Fraction, CA Fit Type, Min Matched Stars, CA Interpolation	CA controls	0.55; Similarity; 6; Lanczos4	—	Edge 0.30–0.80 step 0.05; Similarity/Affine; 3–50; Bilinear/Bicubic/Lanczos4/Pixel Area Relation/Nearest Neighbor/Bit Exact variants	Changes sampled edge region, fit freedom, acceptance - never fewer than five matched stars, and three fifths of a match of fewer than ten candidate pairs - and resampling	CA enabled	Broader region/more complex fit can improve evidence or admit unrelated distortion	RGB group	Model + Preset
Radius, Threshold	Advanced star detector	5 px; 1.0σ	—	Radius 3–32; threshold 0.01–15 step 0.05	Larger radius fits wider stars; lower threshold detects fainter candidates; a stored value the detector refuses - a radius above 32, or a roundness threshold or minimum beta of zero or below - runs at the nearest value it takes, and the log says so once per group - the minimum beta only under the Moffat profile, the only one that reads it	Advanced Settings...	More candidates/work and false positives at low threshold	Registration analysis	Model + Preset
Enable Amplitude Range, Minimum, Maximum	Advanced range filter	Off; 0.0–1.0	—	Min 0–0.99, max 0.01–1, step 0.05	Restricts normalized peak amplitude	Enabled range	Can reject saturated or faint detections	Registration analysis	Model + Preset
Roundness Threshold, Enable Roundness Range, Maximum	Advanced shape filter	0.5; Off; 1.0	—	Min 0.01–0.95, max 0.01–1, step 0.05	Requires rounder profiles or bounds the allowed range	Advanced dialog	Strict values can remove trailed/aberrated stars and too many matches	Registration analysis	Model + Preset
Convergence, Relax PSF Checks	Advanced detector controls	1; Off	—	Convergence 1–3; relax on/off	More convergence passes/refinement; relaxed checks accept less ideal fits	Advanced dialog	More work or more false detections	Registration analysis	Model + Preset
Profile Type, Minimum Beta	PSF controls	Gaussian; 1.5	—	Gaussian/Moffat; beta 0.05–10, step 0.05	Selects PSF family and Moffat acceptance	Beta relevant to Moffat	Model mismatch affects centroid/quality metrics	Registration analysis	Model + Preset
CFA Pattern, Method	Demosaic controls	Auto; Default (current global method)	Global method label can differ	Auto/RGGB/BGGR/GBRG/GRBG; Bilinear/VNG4/AHD/AMaZE/LMMSE/RCD	Selects CFA interpretation and demosaic algorithm used for registration analysis	CFA input	Higher-quality methods cost more; wrong pattern corrupts color/star structure	CFA groups	Model + Preset
Reset to Defaults, Apply to All Groups, OK, Cancel	Registration dialog actions	—	—	Apply-to-all disabled for Comet Tracking	Reset changes dialog; apply overwrites compatible groups; OK commits; Cancel restores snapshot, including Advanced changes	Dialog	Broad apply invalidates many groups	Settings	Model after commit
Edit Points... / Manual Registration controls	Dialog actions	No pairs	—	Target frame; click reference then target; table #, Ref X/Y, Target X/Y; Remove Selected, Clear Target Points, Clear Pending Reference; OK/Cancel	Defines reference-to-target pairs	Manual Points requires explicit reference and at least 3 pairs per target - under a stated polynomial order as many as its terms, 6, 10 or 15 for orders 2, 3 and 4	Manual labor; poorly distributed/incorrect pairs produce bad transforms	Group frames	Model + Preset
Comet Output: Comet only, Track Comet...	Comet controls	Comet only	Experimental Stars+Comet is intentionally hidden	Fixed visible output mode; tracking dialog	Enables moving-target registration/stack mode; the summary under Track Comet... states the observations and the tracking and fit options the registration runs with	Comet Tracking method	Requires reliable timestamps/positions; produces comet-specific products	Group frames	Model + Preset/Artifacts
Track Comet: Use, Mark, Clear Point, Refine, Auto Track, Find Moving Object, Clear, Accept, OK/Cancel	Observation/diagnostic actions	No accepted observations; Refine disabled	—	Frame preview; X/Y points; origin/confidence/state diagnostics; at least two enabled accepted observations for production fit	Marks points, proposes guided tracks, searches diagnostics, accepts candidates, or clears data. The dialog's fallback preview is demosaiced by the global RAW setting and stretched per channel, and Find Moving Object reads frames as fractions of full scale with a residual threshold of one 16-bit code	Auto Track needs accepted seeds; Find Moving Object remains diagnostic until Accept	Tracking accuracy controls comet alignment; incorrect points invalidate the fit	Group frames	Model after OK
Track Comet: Tracking and Fit Options - Search radius, Minimum SNR, Bootstrap frames, Track model, Reject outlier observations, Outlier sigma	Folded section: pixels, ratio, frames, selector, checkbox, multiple	32 px; 3.0; 8; Linear; on; 3.0	A value stored more finely than its control shows - a radius between whole pixels, a threshold with more decimals - is kept while the control shows what it runs as	Search radius 1–64 px, the largest region the nucleus measurement takes; minimum SNR 0.1–100.0; bootstrap frames 0–4096; Linear, Automatic, Prefer quadratic; outlier sigma 1.0–100.0	The nucleus is measured within the search radius of where the track places it in each frame, or of the marked position when Auto Track searches from one observation; a nucleus whose peak is below the minimum SNR times the noise around it is not taken, and the frame is aligned where the track places it (Auto Track lists it as predicted-only); with one accepted observation Auto Track searches at most the bootstrap frames, which do nothing once two are accepted. Linear fits a constant velocity; Automatic an acceleration only when six or more observations remain and it lowers the track's residual by 30% or more; Prefer quadratic an acceleration whenever six or more remain. With rejection on, an observation further from the track than the outlier sigma times the residual of the fit without it is left out when that at least halves the fit's residual, never one within 3 px of the track	Outlier sigma only while outliers are rejected; Auto Track runs with the values shown, before OK	A larger radius tolerates a poorer prediction but can take a star beside the nucleus for it; a lower SNR keeps faint frames measured but admits noise	Group frames	Model after OK; a changed option registers the group again
Blink Comparator: Previous, Play/Pause, Next, Slow–Fast, history table	Inspection dialog	Stopped; current interval mapped to slider	Window size is restored	Arrow keys/Space; speed slider 0–100	Blinks selected frames or compatible historical masters and exposes acquisition/run context; its figures come from each file's header as the engine reads it, and a card that does not read as a number shows nothing	At least two inputs/history comparisons	Visual inspection only	Display only	UI size; no Model change

Stacking tab, stacking settings, and local normalization

UI label/action	Type and unit	Runtime default	Persisted-default exception	Valid values, step, or constraints	Effect of increasing, decreasing, or selecting	Dependencies and enable/visibility rules	Quality, artifact, runtime, memory, or disk trade-off	Scope: preview/apply, channel, mask, still/sequence	Persistence
Integration Groups	Sortable single-selection table	Derived from included Lights and integration grouping	Current group settings/results are restored from the Model/artifacts	Columns: Group, Frames, Calib Groups, Total Time, Stack, Reject, Normalize, Weighting, Drizzle, Output	Selecting a row sets the target for Stacking Settings..., Run This Group, and the stack-specific Frame Quality dashboard; sorting changes display only	Included Light groups; action buttons require a valid selection/model	Selection alone has no processing effect; each row previews the settings and output contract that will affect runtime/disk	One integration group at a time	Selection is Run; group settings are Model + Preset; results are Artifacts
Group by Session, Group by Custom Tag, Group by Exposure / tolerance	Integration grouping	Off / Off / Off; tolerance 3.0 s	Resolution, Filter, Object are mandatory	Toggles; exposure tolerance 0–1000 s step 0.1	Splits final stacks beyond mandatory resolution/filter/object	Stacking tab	More groups reduce frames per stack and produce more files	Light groups	Model + Preset
Stacking Settings...	Button	—	—	Selected group	Opens group settings	One group selected	Settings invalidate the group and relevant artifacts	Group	Model + Preset
Run This Group / Run Integration (All Groups)	Actions	Disabled until valid	—	Selected group or all groups	Runs the minimum scoped/full workflow	Ready model and group	Scoped state has its own hashed last-successful manifest	Project/group	Artifacts
Stacking Frame Quality exclusions	Dashboard actions	Included in stack	—	Exclude from This Stack, Include in This Stack, sigma detection	Removes contributors only from selected integration group; registration and other stacks remain unchanged	Selected stack	Useful when one filter/group needs different contributors; less exposure	Selected stack	Model + Preset
Stacking Frame Quality weights	Dashboard preview: Weight column, histogram, Effective exposure	The selected group's weighting	—	A weight typed into the column overrides the preview only	The weights the group's stack applies, at a mean of one, which the engine derives from the registration's measurements as the stack does: Weighted FWHM penalizes a frame's width for the stars it lacks against the group's reference, Roundness Weighted reads each frame's star roundness, and Number of images reads each light's header without holding the window, saying so until it has. Where the stack would weigh the frames equally - a figure some frame lacks, lights that all count one - the preview is uniform and its banner says why, and a weighting the stack lets go of - beside a combination that takes no weights, under an exposure merge or for a comet stack - previews none, whatever the group stores. The counts Number of images reads are kept until the dashboard is refreshed, so a light replaced on disk while the window is open previews its earlier count until then. Effective exposure counts each frame's exposure in proportion to its weight, the heaviest frame's whole	Weighting other than None	Diagnostic only	Selected stack	Run
Generate 64-bit images	Checkbox	Off	—	On/off	Current implementation has no execution effect. No workflow code reads this checkbox, and calibration/master stack parameters request non-64-bit output.	Always visible	Do not use it as a precision guarantee	Current window only	Run; inert
Reprocess existing	Checkbox	Off	—	On/off	Forces a Full plan and bypasses incremental reuse	Execution option	Recomputes all work and stages existing outputs for archive	Requested run	Run
Open integrated files	Checkbox	On	—	On/off	Opens successful current master outputs in the main workspace	After run	UI/memory cost proportional to masters	Result opening only	Run
Combine RGB images	Checkbox	Off/disabled	—	On/off	Combines compatible monochrome R/G/B masters from the same registration group, matching the channels by the ratio of robust scales and the offset of robust locations over samples between 1e-7 and 0.875 of full scale	Enabled only when eligible R/G/B groups exist	Adds final RGB file/work; no effect on source masters	Eligible groups	Run + Artifact
Save calibrated frames to disk (for debugging)	Checkbox	Off (in-memory)	Low-memory preparation can force disk mode below 2 GiB available memory	On/off	Adds a LightsCalibration stage and writes _cal FITS files, which the registration reads, so the registered images are named from them (<input-stem>_cal_reg); a light with no master and no pedestal has nothing to calibrate it, gets no _cal copy and is registered as it is, as in memory	Lights with calibration	Uses more disk/I/O; makes intermediates inspectable	Light files	Model (lightCalibrationCacheMode) + Preset
Limit memory usage / maximum	Checkbox / MB	Off; 2048 MB	Existing group setting can restore	512–1,048,576 MB	Uses tiled stacking under the limit	Maximum enabled with checkbox	Lower caps RAM. The registered and comet-aligned images the run writes are read a band of rows at a time and once at any cap; lights calibrated in memory, warped as they are read or read from camera files are read whole again for each tile, so a lower cap raises their I/O and runtime	Light stacking groups	Model + Preset
CPU Threads	Integer threads	Detected logical parallel count	Hardware-dependent	1–detected count	Caps the threads each step of the run may use; the viewer and other windows keep the rest	Always visible	Fewer lowers contention/throughput; more uses more CPU/memory bandwidth	Requested run	Run/manifest
Stack Method	Enum	Average	Calibration groups also Average	Average, Sum, Maximum, Minimum, Median, Auto-Adaptive Weighted Average	Selects pixel combiner	Rejection/weighting enabled only where supported; weighting requires Average	Robust/advanced methods cost more; Sum changes scale	Group, all channels	Model + Preset
Enable HDR Exposure Merge	Checkbox	Off	Light groups only	On/off	Buckets exposure lengths, stacks buckets, then blends shorter highlights into long-exposure master; a bucket of one light is that light read as a stack reads it, so a sample its calibration left not a number (its flat zero there) is no sample; where one exposure holds no sample the other's is used, the shorter brought to the longer's scale	Disables normalization, drizzle, rejection/quality maps, weighting, and local normalization; a weighting chosen before the merge was turned on is let go of when the dialog is accepted, and the run's warnings name it	Extra stacks/memory; preserves mixed-exposure highlights at the cost of incompatible diagnostics	Light group	Model + Preset
Normalization	Enum	Lights: Additive With Scaling; Flats: Multiplicative; Bias/Dark: None	Local normalization overrides global choice	None, Additive, Additive With Scaling, Multiplicative, Multiplicative With Scaling	Matches frame offsets/scales before combination	Disabled by HDR; fixed at None and disabled for Bias/Dark groups, whose masters are combined without normalization. While local normalization is on it shows Additive With Scaling (set by local normalization), disabled - the global match local normalization runs - and the stored method comes back when local normalization is turned off	Better matching costs estimation and can alter relative scaling	Group	Model + Preset
Scale Estimator	Enum	IKSS	—	IKSS, MAD	The dispersion each frame's scale is measured by: the iterative k-sigma estimate (IKSS) or the median absolute deviation (MAD)	Light and Flat groups: enabled for Additive With Scaling and Multiplicative With Scaling, and while local normalization is on, whose global match is Additive With Scaling; disabled by HDR and for Bias/Dark groups, which are not normalized	IKSS is the more efficient robust estimate; MAD costs less	Group	Model + Preset
Reference Selection, Reference Frame, Best	Enum/frame/action	Automatic (best quality)	—	Automatic or Manual; included group frame	Chooses normalization anchor(s); Best preselects a manual candidate	Enabled under a global normalization other than None, and under local normalization only with the Explicit strategy, which sets Manual; disabled by HDR. A Manual choice stored before local normalization was turned on stays in force while the controls are disabled, and with Enable Drizzle on this reference is the one the drizzled stack fits the local surfaces against. A Manual reference the group no longer holds - excluded from this stack, or renamed - stops the group, naming it	Poor reference propagates its background/scale	Group	Model + Preset
Enable Overlap Normalization, Min Overlap Pixels, Min Overlap Fraction	Checkbox / pixels / fraction of the smaller light's area	Off; 1000; 0.01	Light groups only	On/off; pixels 0–100,000,000 step 100; fraction 0–1 step 0.01	Matches the lights over the areas they share and solves the matches together, so the panels of a mosaic agree where they meet; a pair whose shared rectangle has fewer pixels than the floor, or less than the fraction of the smaller light's area, is left out	Global normalization only - disabled for None, local normalization and HDR; the floors are enabled with the checkbox. Needs registered lights: a group measured as its lights are read, or whose overlaps do not connect, falls back to standard global normalization with a warning	Reads every light once more to find its overlaps; higher floors leave out small overlaps, which can disconnect a mosaic	Group	Model + Preset
Enable Local Normalization, Configure...	Checkbox/action	Off; Standard preset	—	On/off and detailed dialog	Replaces global normalization with spatial coefficient grids	Disabled by HDR; reference settings adjusted. With Enable Drizzle on, the drizzled stack fits the surfaces itself against the group's reference, and the preset label and the dialog say so (see below)	Handles gradients but costs analysis, memory, and artifact storage	Group	Model + Preset/Artifact
Rejection Method	Enum	Lights: Winsorized Sigma Clipping; calibration: Sigma Clipping	Selection resets low/high values by method	None, Sigma Clipping, MAD Clipping, Median Sigma, Winsorized Sigma Clipping, Percentile, GESDT, Linear Fit	Selects temporal outlier estimator	Average stack only	Stronger estimators need adequate frame count and cost more	Group, per pixel/channel	Model + Preset
Sigma Low, Sigma High	Method parameters	Usually 3/3; Linear Fit 5/4; GESDT 0.30/0.05; Percentile 0.2/0.1 when selected, labelled Low Fraction of Median and High Fraction of Median	Values change on rejection-method selection; a stored pair is kept	0.01–10, step 0.1	Lower rejects more low/high samples. Percentile rejects a sample further below the median than the low fraction of the median, or further above it than the high fraction - a fraction of the median, not a percentile rank	Rejection method that uses thresholds	Too low removes signal; too high leaves artifacts	Group	Model + Preset
Weighting / Weighting Options...	Enum/action	None	Photometric option is unavailable without zero-point coverage	None, Number of stars, Weighted FWHM, Noise, Number of images, Photometric; action opens detailed controls	Converts frame/group metrics to weights and opens the mode-specific options. Weighted FWHM weighs the inverse square of each frame's star width after penalizing the width for the stars the frame lacks against the group's reference, so a hazy frame whose few stars measure narrow is not made heavy. Number of images weighs each light by how many frames its header says it combines - STACKCNT, which every light master the stack writes states, or NCOMBINE - a light that states neither counting once; lights that all count once are stacked unweighted, which the warnings say once	Light + Average; options action is enabled only for modes with detailed controls; disabled for HDR/comet modes	Can improve effective quality; depends on reliable metrics	Group	Model + Preset
Star Count Scaling, Exponential Alpha, Min Weight Floor	Weight options	Quadratic; 2.0; 0.01	—	Linear/Quadratic/Exponential/Roundness Weighted; alpha 0.1–10 step 0.1; floor 0–1 step 0.01	Changes the star-count weight curve; the floor raises the weakest frames to that fraction of the best frame's weight, so none is left out; for Photometric it is a floor on the model's raw weight, on which the most opaque frame's transparency counts one, not a fraction of the best frame's	Scaling and alpha for Number of stars; the floor for Number of stars, Weighted FWHM and Photometric (the photometric model's raw weight) only - Noise and Number of images weights are ratios that keep every frame and take no floor, so it is disabled for them and a stored floor is kept	Aggressive curves concentrate exposure in a few frames	Group	Model + Preset
Enable Per-Channel FWHM Weighting	Checkbox	Off, disabled	A stored choice is let go of when the dialog is accepted	—	Unavailable: the registration measures one star width per frame, not one per colour channel, so each channel can only be weighed by its frame's width. A group whose stored settings hold it on is weighed so and the warnings say it; the Summary tab, the Stacking tab's settings tooltip and the report's Full Processing Parameters list such a stored choice as on, as it is stored	Always disabled	—	RGB channels	Model + Preset
Enable Per-Pixel Weighting, Window Size, Grid Spacing, SNR Exponent	Local-weight controls	Off; 5 px; 64 px; 2.0	—	Window 3–31 odd step 2; grid 4–512 step 4; exponent 0.1–8 step 0.1. A stored value the engine refuses - a grid below 4, an even or smaller window, an exponent that is not positive, from an edited session - runs at the nearest value it takes, and the warnings say which	Builds spatial weight grids at the Grid Spacing asked for, from the pixels each frame's validity admits, and refuses by name a plan that does not fit in memory; smaller spacing/faster exponent reacts more strongly	Supported Average weighting	Fine grids use more CPU/RAM/disk; strong exponent concentrates weight	Group, per pixel/channel	Model + Preset
Create Rejection Maps, Merge Rejection Maps	Checkboxes	Off / Off	—	On/off	Writes low/high maps or one merged map	Rejection enabled; merge depends on map creation	Extra files and memory reveal rejected pixels	Group	Model + Preset/Artifacts
Analyze Spatial Rejection Clusters, Re-Stack Excluding Flagged Frames	Checkboxes	Off / Off	—	On/off	Measures connected rejection clusters and optionally reruns without extreme-rate frames, with the same rejection; the master keeps the group's reference light's cards - or the first light kept, when the re-stack left the reference out - and the report names each flagged light	Re-stack needs analysis; disabled for drizzle/HDR; a re-stack that would leave fewer than three lights is not made - the first stack is kept and the warnings say so	Extra analysis and potentially a second stack	Group	Model + Preset/Report
Create Quality Maps	Checkbox	Off	—	On/off	Writes coverage (the number of frames contributing to each pixel), variance, SNR-gain, and where applicable weight-sum maps (their raw values)	Light stacking	Extra memory/disk	Group	Model + Preset/Artifacts
Allow Registered Images to Be Discarded	Checkbox	Off	—	On/off	On a full run the alignment writes no registered images for the group and the stack warps each light with its recorded transform as it reads it; off keeps the registered_lights files	Light groups only; a reprocessing or incremental run, a drizzled group and a comet group keep their registered images whatever it says	Saves one registered file of disk and write time per light; nothing is left to inspect or to resume from	Group	Model + Preset
Plate-Solve Final Master If Needed	Checkbox	Off	—	On/off	Propagates WCS when possible; otherwise solves a scratch copy with the configured solver before the master is written. Each master's row in the report says where its solution came from: propagated (and from what), solved, or none and why	Solver needed only if propagated WCS is invalid/missing	Solver time/network according to configuration	Final master metadata	Model + Preset
Enable Drizzle	Checkbox	Off	—	On/off	Uses registration transforms and one drizzle descriptor per light during stacking	Light groups; registration transforms required	Higher resolution and disk/RAM cost; needs dithered data	Group	Model + Preset/Artifacts
Scale, Kernel, Pixel Fraction	Drizzle controls	2.0×; Square; 0.9	—	Scale 1.0/1.5/2.0/3.0; Square/Point/Turbo/Gaussian/Lanczos2/Lanczos3; pixfrac 0.1–1 step 0.05	Raises output dimensions, changes flux footprint, and shrinks input pixels	Drizzle enabled	Scale squares pixel/memory growth; low pixfrac needs better sampling/more frames	Group, per pixel/channel/CFA-aware path	Model + Preset
CFA-aware drizzle (drizzle.cfaAware, drizzle.cfaPattern)	Session and preset fields; no control	Off; RGGB	—	On/off; Auto, RGGB, BGGR, GBRG, GRBG	Drizzles each light's filter-array samples into three colour planes: under the stated pattern, placed at each light's origin over whatever its BAYERPAT card says or where it has none, or under each light's own BAYERPAT when the pattern is Auto	Drizzle enabled; single-channel lights; with Auto, a light whose header has no BAYERPAT card is refused, naming the card; a pattern the program does not know is refused before the run starts with Drizzle CFA pattern is invalid	A stated pattern that is not the sensor's routes every sample to the wrong colour	Group	Model + Preset
Recommended Settings...	Analysis dialog	Disabled until frame count exists	—	Shows current/recommended values, rationale, alternate, drizzle viability; Apply Recommended Settings, Apply Alternative, Close	Derives source-based settings from frame type/count and available registration transforms	Selected group	Recommendation is a starting point, not an observed quality result	Settings only	Model after Apply
Reset to Defaults, Apply to All Groups, OK, Cancel	Stacking dialog actions	—	—	Apply-to-all same frame type	Reset restores type defaults; Apply overwrites same-type groups; OK commits - a stored value finer than its control shows (a project edited by hand) is kept as stored unless that control is changed; Cancel discards	Dialog	Broad invalidation on Apply	Settings	Model after commit

Local Normalization Settings dialog controls:

UI label/action	Type and unit	Runtime default	Persisted-default exception	Valid values, step, or constraints	Effect of increasing, decreasing, or selecting	Dependencies and enable/visibility rules	Quality, artifact, runtime, memory, or disk trade-off	Scope: preview/apply, channel, mask, still/sequence	Persistence
Preset	Enum	Standard	—	Quick, Standard, Fine, Custom	Loads grid size 2, 4, or 8; manual edits select Custom	Local normalization dialog	Finer grid tracks smaller-scale variation and costs more	Group	Model + Preset
Strategy / Max Frames	Enum/count under Reference	Integrate Best Frames / 16	—	Explicit, Single Best, Integrate Best Frames; max 1–256	Chooses one explicit/best reference or integrates best references	Max Frames is enabled for integrated strategy; hidden with drizzle on	More references stabilize the reference but cost more	Group	Model + Preset
Grid Size, Interpolation, BG Smoothing	Grid/interpolation controls under Background Model	4; Bicubic; 1	—	Grid 1–32; Bilinear/Bicubic; smoothing 0–10	Finer grid follows smaller gradients; smoothing suppresses local noise	Local normalization	Fine/low smoothing costs more and can fit structure	Group, spatial	Model + Preset
Scale Evaluation: Method, Mode	Enums	PSF Flux; Linked RGB	—	PSF Flux/Residual Ratio; Linked RGB/Per-channel	Selects scale estimator and shared/channel-specific coefficients	Local normalization; Method hidden with drizzle on, whose stack takes the residual ratio	Per-channel may correct color-dependent changes or alter color balance	Group/channels	Model + Preset
Low Clip, High Clip	Normalized thresholds	0.000045 / 0.85	—	0–1; steps 0.00001 / 0.01	Excludes dark/bright samples from estimation	Local normalization	Too narrow leaves too few samples; too wide admits clipped structure	Group	Model + Preset
Enable Advanced Options	Checkbox	Off	—	On/off	Enables the Advanced Options controls below	Dialog	Advanced tuning can overfit	Settings	Model + Preset
Evaluation	Enum under Reference Settings	PSF Signal Weight	—	PSF Signal Weight, Quality Score, PSF SNR, M* Background, Median, Star Count	Ranks reference candidates by the engine's star summary: PSF Signal Weight is the summary's quality, PSF SNR divides by the channel's median absolute deviation, and Quality Score uses the summary's medians	Advanced Options enabled; disabled for the Explicit strategy, which ranks nothing; hidden with drizzle on	Different metrics favor different frames	Group	Model + Preset
Estimator, Percentile, Sigma Clip K, Sigma Iterations	Controls under Background Estimator	Sigma Clipped Median; 0.25; 3.0; 3	—	Percentile/Sigma Clipped Median/Sigma Clipped Mode; percentile 0–1; K 0.1–10; iterations 1–20	Changes robust background estimation	Advanced Options enabled; secondary controls depend on Estimator	More iterations cost more; weak clipping admits structure/outliers	Group	Model + Preset
PSF Flux Settings: PSF Type, Moffat Beta, Max Stars, Min SNR, Allow Clustered / Include clustered sources	Advanced PSF controls	Auto; 4.0; 24576; 40; On	—	Auto/Gaussian/Moffat; beta 1–10 step 0.5; max stars 0–200000; min SNR 1–1000; toggle	Controls star sampling and fit population. The PSF flux scale is the median ratio of the paired stars' fitted volumes, from the stars the engine's detector finds in each channel asked for, and a Moffat PSF Type is fitted with the stated Moffat Beta. Min SNR leaves stars below that signal-to-noise ratio out of the flux scale; it changes neither which stars are detected nor how reference candidates are ranked	Advanced Options enabled; PSF-based evaluation/scale; hidden with drizzle on	Larger populations/lower SNR cost more and may admit bad fits	Group	Model + Preset
Hot Pixel Radius, Noise Reduction	Pixels under Outlier Rejection	0 / 0	—	0–10	Adds preprocessing radii for samples	Advanced Options enabled; hidden with drizzle on	Suppresses defects/noise but can blur small structure used by estimator	Group	Model + Preset
Local Scale (Experimental): Enable / Use local scale k(x,y), Smoothing	Checkbox/pixels	Off / 1	—	On/off; 0–10	Enables experimental local-scale smoothing	Advanced Options enabled	Experimental behavior; validate independently	Group	Model + Preset
Reset to Defaults, OK, Cancel	Dialog actions	—	—	Standard dialog actions	Reset restores defaults; OK commits; Cancel restores snapshot	Dialog	None until committed	Settings	Model after OK

With Enable Drizzle on, a group's local normalization is not measured by the normalization stage: the drizzled stack fits each light's local surfaces and residual-ratio scale against the group's reference as it reads the light. That fit reads the grid, interpolation, smoothing, clipping, background estimator, scale mode and local scale; the reference strategy, the evaluation, the PSF flux scale and the hot-pixel and noise filters do not apply, so the dialog leaves them out with a note saying so, the preset label says the drizzled stack fits it, and the execution plan and the run log say the same.

Summary, execution plan, progress, and history

UI label/action	Type and unit	Runtime default	Persisted-default exception	Valid values, step, or constraints	Effect of increasing, decreasing, or selecting	Dependencies and enable/visibility rules	Quality, artifact, runtime, memory, or disk trade-off	Scope: preview/apply, channel, mask, still/sequence	Persistence
Ready to Execute / Ready with Warnings / Cannot Execute	Status card	Derived	—	Errors block; warnings permit run	Summarizes validation	Always visible	Diagnostic only	Project	Derived
Go to	Issue action	—	—	One per actionable issue	Opens the relevant source/workflow tab/group	Issue metadata available	No workflow effect	Navigation	Run
Integration Summary	Read-only values	Derived	—	Total Exposure, By Filter, Capture Dates, Targets, Source Directories, Camera / Setup	Summarizes included lights	Lights loaded	Diagnostic only	Project	Derived
Session Notes	Plain text	Empty per output directory	—	UTF-8 text; hidden until output set	Saves after 350 ms to output-specific notes file	Output directory required	Not atomic; included in report context	Project/output	Artifact .afternight/image_integration/session_notes.txt
Run History: Open All, Report, Log, Re-run, Delete	Run-row actions	History scan	—	Delete permanently confirms and rejects symlink/out-of-root paths	Opens masters/report/log, reruns embedded manifest, or deletes archive	Corresponding archive file/action must exist	Re-run uses archived settings with current output root and checks it as every run does: a registration database in a format this version does not read refuses it; delete is irreversible	Archived run	Artifact
Group history: Open, Compare	Group-row actions	—	—	Compare needs compatible earlier master	Opens master or launches Blink Comparator	Output must exist	Inspection only	Master files	Artifact/UI
Resource status	Read-only estimate	Disk: -- until output	—	Free space and approximate need for stacks/maps/calibrated/registered/drizzle products	Warns about resource pressure	Output and frames	Estimate does not block execution	Project	Derived
Execute Integration / Fix Issues First	Primary button	State-derived	—	Same workflow as Ctrl+Enter	Opens plan or remains disabled	No validation errors	Long-running/disk writing	Project	Artifacts
Execution Plan Run / Cancel	Confirmation actions	—	—	Decision table: Stage, Group, Decision, Reason, Dependencies, ETA, Outputs	Starts or abandons the prepared immutable manifest	Valid prepared manifest	ETA is an estimate	Requested plan	Run
Progress Cancel / Close / Open Report	Progress/completion actions	Cancel enabled while running	—	Cooperative cancel; Close after finish; report only when generated	Requests cancellation, closes dialog, or opens HTML report; stage progress is the stack's own reports and only moves forward	State-dependent	Completed groups may be committed even after later cancellation/failure	Workflow	Artifacts

How the algorithm works

Conceptual pipeline

flowchart TB
    accTitle: Image Integration conceptual pipeline
    accDescr: Image Integration branches after registration: non-drizzle groups materialize aligned lights, while drizzle groups retain transforms and resample during stacking.
    A[Still files and metadata] --> B[Source and integration groups]
    B --> C[Calibration matching and masters]
    C --> D[Light calibration in memory or on disk]
    D --> E[Registration transforms and metrics]
    E --> Mode{Drizzle?}
    Mode -- No --> Align[Write aligned<br/>lights]
    Mode -- Yes --> Keep[Retain<br/>transforms]
    Align --> Norm[Optional normalization<br/>coefficients]
    Keep --> Norm
    Norm --> StackMode{Drizzle?}
    StackMode -- No --> Standard[Weighted rejection<br/>and standard stack]
    StackMode -- Yes --> Descriptors[Prepare/reuse<br/>drizzle descriptors]
    Descriptors --> Drizzle[Drizzle resampling<br/>and stacking]
    Standard --> Output[Maps, WCS, archives,<br/>and report]
    Drizzle --> Output

1. Import canonicalizes paths, rejects unsupported extensions, reads metadata in parallel, and creates deterministic frame records. Duplicate canonical paths are not added twice.
2. Source grouping applies frame-type-specific criteria. Registration groups use resolution plus effective object name. Integration groups always use resolution, filter, and object, plus the optional final-stack criteria.
3. The matcher builds calibration candidates using the exact field, ambiguity, temperature, exposure, date-parsing, and tie-break contract documented above. Manual assignment or explicit Do not use wins first; otherwise exact compatible assignments precede enabled flat-session or dark-exposure fallback.
4. The manifest freezes canonical inputs, file stamps/fingerprints, group keys, settings, calibration assignments, registration references, stage sequence, output names, execution scope, thread cap, and optional RGB combination. Validation happens before execution.
5. Bias, dark, and flat groups without an existing master are stacked first. Existing masters are reused. Lights are calibrated in memory by default; disk mode writes calibrated_lights/*_cal FITS files, preserving an existing .fit, .fits, or .fts suffix and using .fits otherwise, and introduces a visible LightsCalibration stage.
6. Registration detects/matches stars, correlates translation, uses manual pairs, or fits a comet track according to the chosen method. It records transforms and quality metrics in registration_metrics.db. Optional CA correction precedes the main transform. Optional WCS solving/enrichment supplies astrometric/catalog metrics.
7. Non-drizzle groups materialize aligned images under registered_lights/<registration-group>/, except a group that allows its registered images to be discarded, whose lights the stack warps as it reads them on a full run. Drizzle groups instead keep one reusable drizzle descriptor per light. Local/global normalization coefficients are written before stacking when enabled, except a drizzled group's local normalization, which the drizzled stack fits as it reads each light.
8. Stacking reads the effective calibrated/aligned inputs, applies normalization, optional frame/local weights, temporal rejection, and the selected combiner. Drizzle stacks the lights drizzled onto the canvas its descriptors describe. Optional maps, spatial analysis/restack, RGB combination and comet layers follow; the master's solution is decided before it is written.
9. Each stage reports progress and structured errors. A fatal stage stops later stages; partial success can continue with remaining frames/groups. Cancellation is cooperative.
10. Successful and valid partial products are archived. State files become the comparison baseline for the next scoped or full run, and the report expands the manifest and observed result.

The nominal stage sequence is Bias → Dark → Flat → LightsCalibration (disk mode only) → LightsRegistration → LightsAlignment (non-drizzle outputs) → LightsNormalization → LightsDrizzle → LightsStacking. The manifest omits empty/not-applicable stages.

Model and equations

Calibration

For one pixel/sample, the implemented order is conceptually:

\[ L_0 = L + P,\qquad L_1 = L_0 - B,\qquad L_2 = L_1 - kD,\qquad L_c = \frac{L_2}{F_n}. \]

Here (L) is the light, (P\ge0) is the pedestal, read as 16-bit codes and converted to a fraction of full scale (the value divided by 65,535, whatever the light's own bit depth), (B) is an optional bias, (D) is an optional dark, (k) is the dark scale, and (F_n) is the optional normalized flat. Missing terms are skipped. Every light and master is calibrated as fractions of full scale: an integer file's codes are divided by their largest value, and a floating-point file's samples are taken as they are, so a floating-point light must already hold fractions of full scale. One written in ADU - samples up to 65,535 rather than up to 1 - is calibrated without a warning but 65,535 times off the scale of an integer master and of the pedestal; convert it to fractions first, or give it only masters stacked from floating-point frames in the same units, and no pedestal. With dark optimization, (k) minimizes noise in a central 512×512 crop using golden-section search over 0–2 with 0.001 tolerance; only a single-channel light of at least 512×512 pixels whose dark varies over that square is optimized, and any other light is scaled by the entered factor instead. Cosmetic correction is applied after flat division using the prepared deviant-pixel list. Metadata is copied from the light and the result is marked linear.

If Subtract bias from dark (CCD mode) is enabled, the workflow prepares (D-B) before the light calibration path. Every master dark (D) holds the bias, whether the run stacks it or it is supplied: the run stacks the darks as they are. A master flat is made from flats calibrated by (F - D), with (D) the dark matched at the flats' exposure, in either mode, and by (F - B) with no dark. Flat auto-level divides each channel of the master flat, or each colour of a filter-array mosaic, by its mean over the central third of the frame; the means are computed once per master and reused across the batch. Cosmetic correction finds its hot and cold pixels in the master dark against each dark channel's whole-image median and population standard deviation, once per dark, and replaces them in each light with the median of the pixels around them, of the same colour in a mosaic.

Registration, SIP, CA, and alignment

Star matching estimates a selected transform from matched centroids. Similarity models rotation/translation/uniform scale; affine adds shear/nonuniform scale; homography adds projective terms; polynomial modes add residual distortion terms. Phase Correlation is translation-only. Manual Points solves from user-supplied pairs, and Comet Tracking fits enabled time-stamped positions with robust outlier rejection.

A light is registered by the star pairs it shares with the reference: at least Min Star Pairs, and never fewer than five, however small a part of its stars they are, so a mosaic panel or a light that drifted is registered by the stars it shares. Nearly every star of the smaller field pairs with one of the other's, so a match of fewer than ten candidate pairs, which a few spurious pairs could pass for, also needs four fifths of its candidates to agree on one transform within 3 pixels, unless either light holds 1000 stars or more.

A light matched by fewer star pairs than Min Star Pairs is retried through the lights of its group already registered nearest it in file-path order - the reference aside, at most 24, the nearer first: its stars are matched to that light's by at least Min Star Pairs pairs, that light's matched stars are carried onto the reference through its transform, and the light's transform is fitted to those pairs with the group's model. A light registered so is one the rest are tried through, and the retries repeat while one registers another, so a light that overlaps its neighbours but too little of the reference is still registered. Only a light none of them registers is left to its solution.

When FITS metadata contains SIP A/B coefficients, star matching registers the stars as each frame holds them, so the fitted transform carries the frames' relative distortion and alignment resamples through it alone; a polynomial model holds the curvature a linear one cannot. A frame registered from its own solution - its stars matched neither the reference nor a light registered near it, or a mosaic places it on the sky - is registered by sampling both solutions, distortion included, and fitting a projective model to them, with the polynomial the residual asks for. Only a registration recorded to undo a frame's header distortion has that distortion inverted while aligning, and then the aligned image's world coordinates carry none; the report records how many frames that was.

CA correction detects edge-region stars, fits red→green and blue→green Similarity or Affine transforms, and resamples those channels before main registration; edge samples a shifted plane does not reach are written empty. A channel's match is held to the rule above with Min Matched Stars as its pairs and three fifths for a small match. Mono data or insufficient matches is passed through with diagnostics rather than fabricated correction.

Alignment uses the selected interpolation on a Standard reference canvas, Mosaic union canvas - every frame on the reference frame's plane, carrying the reference's projection - or Intersection common canvas; when no area is common to every frame, Intersection falls back to the Standard canvas and says so - in the alignment, in the stack that warps the lights as it reads them and in the drizzle alike, each in the same words - and the registration records the reference's world coordinates for that canvas, so the master carries them. Auto is Lanczos-4, and the log names the interpolation that ran. When the reference frame is plate-solved, its world coordinates, moved onto the canvas, are written into every aligned image. Changing the interpolation, or switching between the Standard and Intersection canvases, realigns the frames without registering them again; only switching to or from Mosaic registers them again.

Normalization, weighting, rejection, stacking, and drizzle

Global normalization estimates per-frame offset and/or scale relative to the selected reference, from a location and spread computed exactly over the samples. In general form,

\[ x'_i = a_i x_i + b_i, \]

where (x_i) is a sample from frame (i), (a_i) is a multiplicative scale, and (b_i) is an additive offset. The selected normalization mode constrains which coefficients can vary. Local normalization replaces the offset with a spatial grid (b_i(u,v)), and the scale with a grid (a_i(u,v)) when Local Scale is enabled - otherwise the scale stays one constant per frame, or per channel under Per-channel - each interpolated bilinearly or bicubically.

For Average stacking with weights, an accepted sample set (A_p) at output pixel (p) is combined as

\[ I(p)=\frac{\sum_{i\in A_p} w_i(p)x'_i(p)}{\sum_{i\in A_p} w_i(p)}. \]

Global weights use the chosen registration/photometric/noise metric; per-pixel weighting makes (w_i) spatial. The rejection method determines (A_p). No combination clamps the master to 0-1, so a master can hold samples above 1 where hot pixels average above one. Auto-Adaptive Weighted Average weighs every pixel, including one with a sample above 1 after normalization, and a pixel whose samples are all zero or below is their robust mean; a Sum with no positive peak to scale by is written as the raw sums, with a warning. Low/high maps record, for each pixel, the fraction of frames rejected there; the coverage map holds the number of frames contributing to each pixel, and the variance, SNR-gain, and weight-sum maps hold their raw values for the accepted stack.

Drizzle maps each input pixel footprint into a scaled output grid using the recorded registration transform, SIP-aware geometry, selected kernel, and pixel fraction. The engine divides each footprint's flux by (s^2); the stacking stage restates each drizzled frame at its lights' level, multiplying its samples by (s^2) as the stack reads them, so a drizzled master keeps its lights' level at every scale and a drizzled group's normalization coefficients, measured at that level, apply at it. The drizzle accumulates in 64-bit floating point in a fixed frame, row and column order, so its result does not depend on the thread count or the tile height. Scale (s) increases output width and height approximately by (s), so pixel count and uncompressed memory/disk approach (s^2). A CFA-aware drizzle (a session or preset field, with no control) drizzles each light's filter-array samples into three colour planes, under the pattern its settings state or, when that is Auto, each light's own BAYERPAT.

Invalid samples and output boundaries

• Normalization leaves a sample its frame treats as no data exactly as it was, and a normalized sample that would land on exactly zero in a frame that treats zero as no data is written as the smallest normal value instead, so it stays data.
• In the standard non-drizzle stack, a light the alignment wrote, the stack warps as it reads, or a comet stack aligns is read with exact zero meaning no data: those samples are removed before rejection/combination, so a legitimate exactly black sample does not contribute to that output pixel. A group stacked as its lights are read (no registration) treats every finite sample as data, a zero included.
• Calibration writes a non-finite (NaN) sample wherever the flat is zero, in a light calibrated in memory as in a calibrated light written to disk, and a light registered from it carries that sample onto the canvas. No route treats it as data: registration's star detection measures each light's background without it and searches the background in its place, so such a light is registered from its calibrated stars; a group stacked as its lights are read leaves it out, the registration's canvas reads it as the exact-zero marker and removes it with that marker, and drizzle excludes it before mapping. Normalization measures the lights the same way. The master's pixel there is combined from the other frames, and a pixel no frame holds a sample for is written as zero.
• Drizzle distinguishes data from geometric coverage: each light's geometry comes from its drizzle descriptor, and an output pixel no input footprint reaches is marked as having no data, but finite exact-zero image samples remain valid contributors and a Lanczos kernel's negative lobes are kept. This preserves genuinely black calibrated signal instead of treating it as missing drizzle coverage.
• A drizzled output pixel that no valid sample reaches carries no data into the stack, and a master pixel that no frame reaches is written as zero. In Intersection mode the drizzle keeps an edge guard around its canvas, which the stack crops from the master and its maps; an intersection no whole output pixel is common to is drizzled onto the Standard canvas, without a guard. Coverage and quality maps report the contributing-sample distribution; inspect them at transformed edges before accepting a master.
• Invalid images, transforms, settings, or stage-level results still use the structured validation/processing errors below. The sample rules above describe expected per-pixel handling, not silent recovery from a failed stage.

Incremental planning

The current manifest is compared with the exact previous manifest for the same scope; its fingerprints read each choice by its meaning, not its spelling. Changes include added/removed/modified calibration groups and file stamps, master dependencies, per-light assignments, registration light sets/settings/reference, integration light sets/settings/assignments, and normalization references.

• Recompute propagates from a dirty master to affected lights, registration groups, and dependent integration groups.
• Reuse keeps unchanged products when the required file/database row and schema are present.
• Skip marks an inapplicable stage or an external master.
• With no exact previous manifest, the planner inspects existing masters, registration_metrics.db, registered outputs, normalization metadata, and stacked outputs. It uses an artifact-backed incremental plan when some valid work exists; otherwise it chooses Full.
• With an unchanged manifest, missing outputs are still scheduled for regeneration. If nothing is missing, the plan says Nothing to process.
• Reprocess existing forces Full.

The generated bias, dark and flat masters and the stacked masters are recorded in artifacts.db with a SHA-256 digest taken from one full read of the file right after it is committed; the run archive records the same digest as the provenance of what it keeps. The digest is never compared to decide reuse. Preparing a plan reuses such a master when the expected manifest-section hash, file size and modification time match and the file is still the one recorded, or, when it was moved or restored under a new file identity, when its own FITS header still states the identity of the manifest entry it satisfies. Neither preparing the plan nor starting the run reads a master's pixels to decide this.

Generated bias, dark and flat masters and calibrated lights are written whole or not at all, and each carries in its own FITS header (ANARTID, ANRUNID) the identity of the manifest entry it satisfies and the run that wrote it. Calibrated lights and masters are reused from the recorded file identity or, when the file was moved or restored, from that header, without reading their pixels, so a copy restored with the same bytes and modification time is still reused; a moved master whose header no longer states its identity, or whose record holds no identity to compare with, is made again. A master is always a 32-bit floating-point file, also when its group holds one frame; a group of fewer than three frames is combined without its rejection method, and a calibration group's weighting or local normalization, and a bias or dark group's normalization, is not applied, each said in the stage's warnings. A calibration frame that cannot be read, or whose size, channels, bit depth or CFA differs from the group's first readable frame, is skipped and listed against its own file. A master that cannot be read fails the master group that applies it, or only the lights that name it.

CA-corrected lights, registered lights and comet-aligned lights are written the same way: whole or not at all, each stamped with the manifest entry it satisfies and the run that wrote it, and recorded only once the file is complete. Registration positions - transforms, manual control points and comet observations - are stored with the first pixel's centre at (0.5, 0.5). Gaia enrichment, and the matched sources spectrophotometric calibration reads, are produced for every solved frame when the Gaia catalog is installed and the group's Enrich with Gaia catalog setting is on - it is on by default. Where no catalog is installed, a group with a solved light detected in the run says once, in its warnings, that it enriched nothing.

Normalization coefficients are measured where the stack reads the lights: the registered lights, as the alignment wrote them, when every light the registration accepted is current - a light it refused is left out, as the stack leaves it out - the registration inputs warped onto the registration's canvas in memory when they are not, a drizzled group's lights warped onto its drizzle canvas, and a group with no registration as its lights are read. The reference is the group's normalization reference frames, or its first light; several reference frames are combined. A Manual reference the group no longer holds - excluded from this stack, or renamed - stops the group with a message naming it. Local normalization fits each light's background surface, its scale against the reference and then the global match over the locally normalized light. When local normalization ranks its own reference - Single Best Frame - the stack takes that light as its reference too, so the noise and width weights are relative to it as well; a reference of several integrated lights brings every light to the integrated surfaces and to the levels of the group's normalization reference. The coefficients and their .meta.json are written whole or not at all and reused while the sidecar still describes the inputs and every setting - the local normalization values no dialog offers included. The stack applies only coefficients measured under the group's current settings, and otherwise measures its own and says so; a drizzled group's coefficients apply at any drizzle scale, so changing only Scale does not measure them again. The report's stacking quality section names what brought the lights to the reference, and which light that was. Coefficients or drizzle descriptors in a format this version does not read are neither read nor replaced, and the group fails with a message, unless Reprocess existing is on. Drizzle writes one small descriptor per light, holding its geometry and the engine's identity of it, and dense coordinates beside it only when checkpoints are kept.

Stacking reads each group's lights where its registration put them: the registered lights when the group keeps them and each is still the file the registration recorded - a registered light changed since stops the group with a message naming it - the registration inputs warped as they are read when the group's registered images are discardable, the drizzled lights, the comet-aligned lights for a comet stack, or both layers for a stars-plus-comet composite. A stored stacking method, rejection, weighting or star-count scaling the program does not know - for a calibration group, its method or rejection - is refused before the run starts, with a message naming the setting and the group, and no group runs; and every setting a group's stack cannot honour - a weighting a median takes no weights from, a rejection a group has too few frames for, maps of a stack that rejects nothing - is said in the warnings rather than replaced silently. Weights are read from the registration's metrics - Number of images from each light's own STACKCNT or NCOMBINE - and a group missing them is stacked unweighted, with a warning; a light whose stars have no measured roundness is weighed with the others; a stored per-channel FWHM weighting is said, since the registration measures one star width per frame; every light master states in STACKCNT the frames it combines - the sum of its lights' counts, so a master of masters states every frame beneath it, and a colour master its reference channel's count - and a calibration master states none. The normalization coefficients are applied where they describe the frames the stack reads; otherwise the stack measures its own, with a warning. The master's world-coordinate solution is decided before the master is written - from the registration's canvas, moved onto the drizzled canvas for a drizzled master, or by a fallback solve when Plate-Solve Final Master If Needed is on; a comet-aligned master, drizzled or not, carries none from the registration and is solved, like any master without one, only when Plate-Solve Final Master If Needed is on - and a mosaic's pointing cards (RA, DEC, OBJCTRA, OBJCTDEC) state that solution's field centre or, when the master has no solution, the spherical mean of its lights' own pointing cards. The master, its maps, the composite's layers and the colour master are written whole or not at all, each stamped with the manifest entry it satisfies, so a later run reuses them from a header read; a map is shown with its master only while both carry the same stamp and are the files recorded. The quality report keeps the pixels in rejection clusters and every element either map rejected as two separate figures. A stars-plus-comet group cannot be drizzled.

Engine handoffs

Workflow responsibility	Production handoff	Result returned to the workflow
Calibration and cosmetic correction	ImageCalibration.h (calibrateImage), CosmeticCorrection.h, and executeStack with the masters applied as frames are read	Linear calibrated images or per-frame calibration errors
Star matching, phase correlation, metrics, alignment	StarDetection.h (detectStars), StarAlignment.h (alignStarFields), RegistrationFit.h (fitRegistrationTransform), phase correlation, and RegistrationPlan.h (planRegistrationCanvas, the warp)	Transforms, star/quality/astrometric metrics, aligned outputs
SIP geometry and transform bounds	SipDistortion.h, WcsSolution.h, computeRegistrationBounds	Sampled solution-to-solution registrations, inverse warp, output geometry
Global/local normalization	GlobalNormalization.h (measureGlobalNormalization, estimateGlobalNormalization, the overlap solve), LocalNormalization.h (fitLocalBackgroundGrid, estimateLocalScale, fitLocalScaleGrid), DrizzleLocalNormalization.h	Reusable coefficient document and its sidecar
Rejection, weighting, combination, HDR	StackExecutor.h (executeStack) over the workflow's frame source, StackQuality.h (analyzeRejectionClusters, countRejectionMaps), CometComposite.h for the stars-plus-comet composite, ChannelTransfer.h (composeImageChannels) and the channel composition's linear match for the colour master	Master, rejection/quality data, quality report
Drizzle reconstruction	DrizzleCanvas.h (planDrizzleCanvas), DrizzleMapDescriptor and its identity, and DrizzledStackFrameSource under executeStack	Per-light drizzle descriptors, optional dense checkpoints, and a drizzled master
Registration's solve of an unsolved frame	PlateSolveService (the internal solver against the installed Gaia catalog, or the configured external solver)	A solution the frame is registered from, or a warning
Plate-solving fallback	PlateSolveService over a scratch copy of the master, before the master is written	The master's solution, or a warning

Boundary and failure behavior

• Model validation blocks an empty output directory, a non-directory path, an unwritable/missing uncreatable directory, and an empty light list.
• Calibration coverage problems are warnings when calibration material exists. Pre-calibrated groups are exempt.
• Manifest validation checks positive version, output path, registration/integration references, output filenames, comet observations/reference, the stacking method, rejection, weighting and star-count scaling each group names, drizzle scale (1–4), pixel fraction (0,1], kernel/CFA values, light-calibration paths, and consistent readable dimensions for multi-frame groups. Channel compatibility is not currently a manifest check.
• A light integration group needs at least two usable lights: a group of one light, or one left with fewer than two after lights without a registered product are skipped, fails even when the top-level model was otherwise ready. A calibration group of one frame is stacked into a master like any other group, and a flat group of one frame is calibrated by its masters.
• Invalid images, transforms, settings, and stage-level results are reported by the relevant engine/stage. Individual NaN/zero/boundary samples follow the documented standard-versus-drizzle missing-data rules and do not necessarily create an error. Output conversion is handled by the engine codec; generated workflow masters use .fit names and current stage parameters do not request 64-bit output.
• File-not-found, unreadable-file, and invalid-format structured errors are recoverable categories; stages may skip them if enough valid work remains. The light stack itself does not: a light it cannot read once the stack has started fails the group, naming the light, and only a light whose registered product is missing before the stack starts is left out and listed. Other I/O, validation, processing, memory, timeout, and configuration errors are fatal unless a stage explicitly returns partial success.
• An output being overwritten is first renamed with a .tmp_archive[_uuid] suffix. A failed in-progress group restores its old files. Completed groups remain valid if a later group/stage fails or cancellation arrives.
• If a cancelled/failed workflow has completed integration groups, those groups and prior staged versions are archived and a committed-subset manifest becomes the next incremental baseline. If archiving cannot commit, warnings are logged; orphan cleanup restores/removes stale staging files on a later run.
• Cancellation does not terminate an external/native operation mid-instruction; it is checked between safe units of work.
• A comet output that cannot be written is recorded as a failed product, and the stage names the lights.
• A warning two stages give in the same words is stated once in the run's warnings and counted once.

Structured error families are:

Codes	Family	Examples	Default continuation rule
100–107	I/O	Missing/unreadable/unwritable file, missing/unwritable directory, disk full, locked file, invalid format	Only missing, unreadable, and invalid-format are classified recoverable.
200–206	Validation	Missing frames, incompatible dimensions/channels, invalid configuration, empty/insufficient list, invalid output	Fatal unless the stage returns a smaller valid partial group.
300–306	Processing	Stacking, registration, calibration, normalization, drizzle, memory, internal failure	Fatal for the affected stage/group; other completed groups can remain.
400–401	Cancellation	Cancelled, timeout	Stops the requested workflow and preserves committed completed groups.

Parameter interactions and starting points

Combination	Conservative starting point	Too weak looks like	Too strong looks like	Adjustment order
Calibration matching	Keep exact grouping metadata; leave flat/dark fallbacks on but review every fallback warning	Unmatched groups or obvious residual sensor/flat structure	Over-broad grouping uses incompatible masters	Correct metadata/type → grouping → exact masters → fallback limits → per-light settings
Cosmetic correction	On, Hot/Cold Sigma 3.0	Isolated hot/cold pixels remain	Real small stars/detail are replaced	Confirm masters → inspect defects → change one sigma threshold → compare calibrated frames
Registration	Star Matching, Auto-select, Homography/Auto behavior, Standard, Auto interpolation	Too few matches or systematic residuals	High-order model follows bad matches/warps edges	Improve star detection/reference → inspect exclusions → choose transform → change interpolation last
Quality exclusion	Inspect metrics first; use 3σ detection or IQR suggestion only as a selection aid	Clearly poor frames remain contributors	Useful exposure is discarded by a metric-only rule	Run registration → normalize by filter → inspect frames/Blink → apply exclusions
Light stack	Average, Additive With Scaling, Winsorized Sigma Clipping 3/3, no weighting/maps/drizzle	Background mismatch/outliers remain	Rejection removes real signal or too few samples remain	Normalize → rejection method → thresholds → weighting → maps
Local normalization	Standard preset, Integrate Best Frames, Linked RGB	Spatial gradients remain inconsistent	Real extended structure is fitted as background; color balance shifts	Choose references → grid → clipping/smoothing → advanced estimator/PSF → per-channel only if justified
Drizzle	Off initially; when justified, 2×, Square, pixfrac 0.9	No recovery of sub-pixel sampling	Sparse coverage/noisy holes, excessive memory/disk	Verify transforms/dither → enable maps → scale → pixfrac → kernel
Memory/parallelism	In-memory calibration, no explicit memory cap, detected threads	Underuses available CPU or adds disk I/O	Memory pressure, swapping, UI/system contention	Set memory cap if needed → reduce threads → enable disk calibration for inspection/low memory

These are implementation-backed defaults and adjustment orders, not promises about visual improvement.

Inputs, outputs, and metadata

Input and grouping contract

• One project may contain any count of bias, dark, and flat stills, plus required light stills.
• Canonical paths identify frames. Metadata overrides affect grouping in the project but do not rewrite source headers.
• Registration groups are resolution + effective object. Integration groups are resolution + filter + effective object, with optional session/custom tag/exposure.
• Globally excluded frames are absent from registration and integration. Stack-only exclusions affect only one integration group.
• The workflow does not consume an active image, mask, or sequence.

Output names and locations

Output	Current path/name	Created when
Master bias/dark/flat	<output>/master_bias_<sanitized-group>.fit, master_dark_...fit, master_flat_...fit	Generated calibration group is processed.
Calibrated light	<output>/calibrated_lights/<input-stem>_cal.<FITS-suffix> for a FITS input; <input-stem>_cal.fits otherwise	Disk calibration mode. Existing .fit, .fits, or .fts suffixes are preserved; non-FITS sources become FITS intermediates. A run removes the copy of a light it no longer calibrates (its group marked Pre-Calibrated, or no master left for it), so the light is read as it is.
CA-corrected light	<output>/ca_corrected_lights/<input-stem>_ca.<FITS-suffix> for a FITS input; <input-stem>_ca.fits otherwise	CA correction enabled and a corrected product is materialized. Existing .fit, .fits, or .fts suffixes are preserved.
Registration metrics	<output>/registration_metrics.db	Registration runs; rows are keyed by group/settings version and frame.
Registered light	<output>/registered_lights/<sanitized-registration-group>/<stem>_reg.<FITS-suffix>, the stem being that of the file the light was registered from: <input-stem>_ca when chromatic aberration correction wrote a corrected copy, <input-stem>_cal when calibrated frames are saved to disk, otherwise <input-stem>	Non-drizzle alignment materializes registered images, except for a group that allows them to be discarded on a full run (the stack then warps each light as it reads it). A group containing any non-FITS alignment input is written as .fits; otherwise each .fit, .fits, or .fts suffix is preserved. Registered images an earlier run wrote under another of these names - <input-stem>_reg from a light whose calibrated copy it did not read - are left in the folder beside the new ones until removed, and no stage reads them. Comet-aligned lights are written as <stem>_comet_reg.<FITS-suffix>, from the same stem (.fits when any input is not FITS) in registered_lights/<registration-group>/comet_aligned/<comet-run>/.
Normalization coefficients	<output>/.afternight/image_integration/artifacts/normalization/<group>/coefficients.json plus .meta.json	Normalization enabled; not written for a drizzled group's local normalization, which its stack fits as it reads each light.
Drizzle descriptors	<output>/.afternight/image_integration/artifacts/drizzle/<group>/pixmaps/ - one .drizzle descriptor per light, .drizzlemap dense checkpoints when kept, and the group's .meta.json	Drizzle enabled.
Stacked master	<output>/stacked_<sanitized-integration-group>.fit	Integration group succeeds. Filename collisions receive a stable short hash.
Rejection maps	<master-stem>_rejection.fit or _rejection_low.fit and _rejection_high.fit	Rejection maps enabled.
Quality maps	<master-stem>_coverage.fit, _variance.fit, _snrgain.fit, and optionally _weightsum.fit	Quality maps and/or weighting request them.
RGB/comet auxiliaries	Compatible combined RGB master; comet layer/mask and comet-aligned products beside the group outputs	Corresponding execution/method options.
Artifact registry	<output>/.afternight/image_integration/artifacts.db	Stack artifact hashes are recorded.
Last successful state	<output>/.afternight/image_integration/last_successful_manifest.json, scoped last_successful_manifest_<scope>_<16-char-keyhash>.json, and last_successful_run.json	Full/scoped successful or committed-partial run. Sensitive runtime fields are redacted for persistence.
Session notes	<output>/.afternight/image_integration/session_notes.txt	Notes are entered.
Run archive	<output>/history/<execution-id>/	Successful run, or failed/cancelled run with completed integration groups.

A drizzled master states its drizzle in DRIZZLE, DRIZMULT (the scale), DRIZPIXF, DRIZKERN and NDRIZIM (the lights drizzled), and the drizzle step in D001SCAL, D001PIXF and D001KERN; the kernel cards hold the kernel's key in lower case - square, lanczos3 - not the dialog's spelling.

The master's observation cards state only facts the reference holds: a frame type is a quoted string, text that does not begin with a number writes no 0, an exposure, gain or offset of exactly -1 is written, and a non-finite number or a date that names no instant is left out. A mosaic master's pointing cards are read by the engine's angle reader - a card whose text has a fractional field followed by another field is not read, and a full turn folds to zero - and a declination that rounds to zero is written with a plus sign.

The archive can contain workflow_manifest.json (schema 3 with embedded manifest/results), workflow_report.md, workflow_report.html, workflow_log.txt, a snapshot of registration_metrics.db, groups/<group>/group_summary.json, archived/staged outputs, and generated report previews. A report or archive preview is the viewer's presentation of the image shrunk by its largest samples; an image of other than one or three channels has none. Each group's preview is written by the viewer's real-time presentation. History retains the newest runs until both limits are met: at most 20 entries and at most 5 GiB total; oldest directories are removed first.

The report includes masthead/provenance/session notes/index; executive and acquisition summaries; equipment/run focus; stage timeline; stacked masters; calibration and registration summaries; SIP and comet/SPCC diagnostics when applicable; full processing parameters and automatic decisions; quality gate/weighting/targets; comparison with previous runs; metadata completeness; reproducibility fingerprints; failure diagnostics; inputs; warnings/errors; and a technical appendix with optional detailed file lists/log.

The report's statistics are the engine's, as fractions of the display range with clipped samples left out: ASQR and SSQR are the mean and sum of squares, location and scale are the engine's IKSS, Sigma and Std Dev are both the population deviation, ISO, gain and offset are read as integers, and a FITS card is matched whatever its case. Its frame weights are the stack's for every weighting, derived by the engine from the measurements the stack reads and shown relative to the heaviest frame, the photometric weights at their mean of one; the floor is held to 0-1, and counted and marked for Number of stars and Weighted FWHM where the stack raises a weight to it, while the photometric floor, on the model's raw weight, is applied but not counted. Its registration parameters call the per-frame star cap Max Stars per Frame, as the dialog does, and the transform mode's row says whether a polynomial is fitted over the linear model - for a polynomial mode and for Auto.

App-local persistence

• On window creation, the model loads session last_session from image_integration_sessions.db in the platform app-local-data directory. Autosave uses SQLite WAL and writes every 5000 ms while the model is dirty; close triggers an immediate save.
• Sessions, presets and project files store each choice by a fixed key, so renaming a label never changes a saved setting. If the stored session holds a value no dialog writes, the window starts empty, says which value, and does not autosave over the stored session. A session database in a format this version does not read is neither read nor changed: it is set aside beside itself under a new name, which the window names, and new sessions are saved from then on.
• A project's registration_metrics.db in a format this version does not read is refused and left unchanged.
• Presets are JSON files in app-local presets/. They contain only light-calibration cache mode, calibration policy, source/integration grouping, stacking settings, registration settings, light-calibration settings, and calibration overrides.
• Project export writes the entire model to JSON with QSaveFile. Project import reads a project exported as JSON and refuses any other file with a message, loading nothing.
• Recent project paths (maximum 10) and source-tab splitter/header/display/coverage UI state use QSettings.
• The undo stack is window-local and does not survive close.

Performance

Primary cost drivers are total input pixels, number of groups, star count/match cap, alignment canvas size, normalization grid/sample count, rejection method, map generation, drizzle scale/pixfrac/kernel, and any second stack triggered by spatial rejection. Preparing a plan reads no recorded master or light in full: the masters and the lights the workflow writes are checked from file identity, or from a header read when a file was moved or restored.

• Import metadata is parallelized with oneTBB and shows Pause/Cancel/ETA for large sets.
• Workflow stages execute in dependency order; work inside suitable stages is parallel, capped by CPU Threads.
• In-memory light calibration avoids intermediate I/O. Disk mode costs roughly one calibrated FITS per light. The preparation path can fall back to disk calibration when available memory is below 2 GiB.
• Registration retains metrics in SQLite so unchanged groups can be reused. Non-drizzle alignment writes one registered image per retained light; drizzle writes one small descriptor per light instead, plus dense coordinates only when checkpoints are kept.
• A drizzle scale of (s) approaches (s^2) output pixels. Mosaic alignment similarly expands canvas area beyond a single frame.
• Local normalization, per-pixel weighting, rejection/quality maps, and report preview generation add memory, CPU, and disk use.
• Limit memory usage bounds what the stack holds. The registered and comet-aligned images the run writes are read a band of rows at a time from their files, so the stack holds a few rows of every light and reads each light once whatever the cap. Lights calibrated in memory, warped as they are read or read from camera files are read whole, and under a cap the stack reads them again for each tile, so a lower cap costs those runs more I/O and time. Lights saved as compressed XISF are decompressed in pieces that span many rows, so the stack holds those rows of every light where the cap allows and warns when it does not, since a smaller hold decompresses each piece again. The run log's Frame reads line says which way each group was read, and the Stack tiles line how many rows the stack held at a time.
• The Summary disk estimate is advisory. Verify real free space before a long run.

For a safe prototype, run registration for one representative group, inspect its metrics, then use Run This Group with maps enabled. Do not extrapolate a measured runtime from the synthetic unit/integration fixtures to production data.

Worked example

• Evidence mode: Source-and-test-derived from src/AfterNightTests/Test_ImageIntegration/integration/test_full_workflow.cpp. The fixture was executed within the complete deployed MSVC Release and Debug Test_ImageIntegration suites; no approved astrophotography dataset was executed for this page.
• Input condition/provenance and settings: the fixture creates two synthetic 64×64 FITS files for each of Bias, Dark, Flat, and Light; uses one group per calibration type, a fixed first-light registration reference, normalization with the first light, and drizzle artifacts in the full manifest.
• Expected checkpoint: the test asserts successful creation of master_bias.fit, master_dark.fit, master_flat.fit, both calibrated lights in disk mode, registration_metrics.db, normalization coefficients.json plus metadata, the drizzle group's descriptor sidecar, stacked_TestObject_L.fit, and a history archive containing a schema-3 manifest and report path. A later identical run exercises reuse; a separate ten-light/five-group test changes one local-normalization group and asserts that the registration database and four unaffected groups retain their modification times.
• Inspect at 100%/statistics/history: no visual 100% or statistics result is claimed because the inputs are synthetic test patterns. The verified inspection point is file/database/history structure and deterministic incremental timestamps. With approved real data, inspect calibrated residuals, registered edges/stars, rejection/quality maps, and the master at 100% before accepting settings.
• Limitations: the fixture proves orchestration and artifact contracts, not real-world calibration quality, photometric fidelity, color accuracy, drizzle resolution gain, or production performance.

Warnings and limitations

• Generate 64-bit images is currently inert. The checkbox is present and tested for visibility only; execution does not read it and current stage parameters request non-64-bit output.
• A floating-point light must hold fractions of full scale. Its samples are calibrated as they are, while an integer master's codes and the pedestal are divided to fractions, so a floating-point FITS, TIFF or XISF light written in ADU (samples up to 65,535) is calibrated against masters and a pedestal 65,535 times smaller, and nothing warns of it. See Calibration under Model and equations.
• Multiplication Factor does not control an optimized dark. With dark optimization on, a single-channel light of at least 512×512 pixels gets the engine's own 0–2 factor and the entered value is ignored; a colour light, a smaller one, or one whose dark does not vary over that square is scaled by the entered value instead, and the run's warnings say so once per calibration. With optimization off the entered value is applied as a fixed scale.
• The Registration UI tooltip describes a group as sharing dimensions, channels, and binning, but the current key is only resolution + effective object. The manifest checks multi-frame dimensions, not channel count. Split incompatible data yourself.
• Import Project... reads only a project exported as JSON; any other file is refused with a message and nothing is loaded.
• No sequence, active-image mask, real-time process preview, or Apply action is supported. The Lights Calibrated control is a static source-preview inspection path, not an RT preview.
• Calibration warnings are non-blocking. A ready status can still represent an intentionally uncalibrated or fallback-calibrated group.
• Source preview stretch/demosaic/calibration is for inspection and is not the saved processing result.
• Changing grouping, master assignment, reference, calibration settings, registration settings, normalization references, or stacking settings invalidates related work and can propagate downstream.
• Reprocess existing stages current outputs for replacement. Run history is bounded; it is not a permanent backup.
• A cancelled or failed run may retain and promote already completed groups as the current partial baseline. Read the archive/result before rerunning.
• Low-disk status is an estimate, not a hard validation gate. Disk-full/locked-file failures can still occur mid-run.
• Online plate solving may transmit solve data according to the configured solver; see the Plate Solving page and Settings before enabling it.
• Drizzle and high-order transforms cannot create information absent from sampling/matches. Use them only with suitable dithered data and inspect maps/edges.
• Experimental Stars+Comet output exists internally but is deliberately hidden; the visible Comet Tracking output is Comet only.
• Refine in Track Comet is currently disabled. Find Moving Object creates diagnostics until a candidate is explicitly accepted.
• The five requested screenshots remain deferred; no image on this page should be treated as evidence until capture/provenance is recorded.

Troubleshooting

Symptom or exact error	Verified cause/validation path	Recovery	Evidence source
Fix Issues First / Execute disabled	Empty lights, empty/invalid/unwritable output directory, or another model error	Use each Go to action; add lights and select/create a writable directory	IntegrationDataModel::validate, SummaryTab
Import Project... refuses the file and loads nothing	The file is not a project exported as JSON	Export/import the project as JSON	ImageIntegrationProcess::onOpenRecentProject
Import says “Unsupported file format”	No codec the engine registers reads the file's format (its extension is not one of the supported still formats)	Convert/extract to FITS/TIFF/PNG/JPEG/RAW/XISF; do not import a sequence container	FileImport, CodecRegistry
Import warns “Metadata read failed”	The format is read but the file's header could not be probed	The frame is still imported - its type taken from its folder or the source type you chose, its size unknown and its file modification time as capture time; check the file, or remove the row and add it again	FileImport
Calibrated is disabled	No Light is selected; the group is Pre-Calibrated; no usable master remains after matching/manual unassignments; a selected generated/assigned master file does not yet exist; or CCD mode has a Dark but no Bias	Select a Light, review the checkbox tooltip, build/select the required master files, correct matching/overrides, or disable Pre-Calibrated/CCD mode only when that reflects the data	ImageIntegrationProcess::updateLightCalibrationPreview, ImageViewControl
Light shows missing calibration	No exact/fallback/manual compatible master, or it was explicitly unassigned	Inspect Coverage, grouping metadata, matching policy, and master source; use Pre-Calibrated only when true	CalibrationMatcher, model validation
Dark fallback warning	Exact exposure match absent and fallback is within tolerance	Add matching darks, narrow tolerance, disable fallback, or accept after inspection	CalibrationMatcher
Registration metrics unavailable	Group has not run for the current settings version, database/schema is missing/stale, or frames failed. A database in a format this version does not read is refused and never changed	Run Registration; inspect Process Output; clear and rerun stale registration if necessary	RegistrationTab, ExecutionPlanner
Registration has too few matches	Detection thresholds/reference/method do not produce required pairs	Choose a stronger reference, lower threshold cautiously, raise Maximum Stars per Frame, use phase/manual points when appropriate	RegistrationSettingsDialog, LightsRegistrationStage
The run does not start: Unknown <setting> '<text>' in group '<group>'.	A stored stacking method, rejection, weighting or star-count scaling (for a calibration group, its method or rejection) that this version does not know - a project edited by hand or written by another version	Open the group's Stacking Settings..., choose the setting again and accept the dialog	WorkflowManifest::validate
The reference frame has N stars, and a star match needs at least 5 pairs; its lights cannot be matched by their stars.	The registration reference holds four stars or fewer, so no light can be matched to it by its stars	Choose a reference with more stars, lower the detection threshold, or place the lights by their world coordinates under Mosaic	LightsRegistrationStage
A light is left unregistered: <light> has no world-coordinate solution and no plate solver is configured	Star matching could not place the light, and it could only be placed by a solve, which needs a configured solver	Configure a solver in Settings > Plate Solving, or improve the star match	LightsRegistrationStage, PlateSolveService
Manual Points cannot be accepted	Reference missing or a target has fewer than three pairs	Select a fixed reference and add at least three well-distributed correct pairs per target	ManualRegistrationDialog, manifest validation
Comet workflow rejected	Fewer than two enabled accepted observations, observation outside group, or fixed reference absent	Mark/accept two time-separated points in group and choose a reference	CometTrackingDialog, WorkflowManifest::validate
Drizzle falls back or fails	Too few usable transforms/drizzle descriptors or incomplete cache	Run registration, retain at least two transformed frames, clear stale registration/drizzle artifacts, rerun	LightsDrizzleStage, LightsStackingStage
Black/low-coverage transformed edges or a zero-valued sample differs between standard and drizzle stacks	Standard rejection treats exact zero as missing to suppress null/alignment-border samples; drizzle uses transform coverage and keeps finite zero data, while drizzled output no footprint reaches carries no data and a master pixel no frame reaches is written as zero	Inspect the coverage map and registered/drizzle geometry; crop low-coverage edges when appropriate and compare like-for-like standard/drizzle settings	StackFrameSources, DrizzledStackFrameSource, StackingStageTest
Non-drizzle stack says no registered outputs	Alignment products are absent for current transforms	Run Registration/Alignment; do not delete registered_lights between plan and stack	LightsAlignmentStage, LightsStackingStage
Photometric weighting unavailable	Registration metrics lack usable zero-point coverage	Configure Gaia/WCS as needed and rerun registration, or choose another weighting mode	FrameQualityDashboard, photometric weighting utilities
The window starts empty and names a stored value	The saved session holds a value no dialog writes	Open the project from its JSON export after correcting the value, or choose the settings again; the stored session is left as it was	IntegrationDatabase, SessionDocument
A preset is refused	The preset holds a value no dialog writes, or is in a format this version does not read	Save the preset again from the current settings; the file is left as it was	PresetManager
“Nothing to process”	Manifest and required artifacts match current state	Use existing outputs, or select Reprocess existing only when a full rebuild is intended	ExecutionPlanner
A deleted master/stack is unexpectedly rebuilt	Unchanged-manifest artifact scan found a required file/database row missing	Allow targeted regeneration or restore the exact valid artifact	ExecutionPlanner::populateMissingArtifactsPlan
Rerun recomputes downstream groups	A master/light assignment, registration setting/reference, or integration setting changed and dirty state propagated	Inspect Execution Plan Reason/Dependencies; revert unintended setting changes	ManifestDiff, DirtyStateTracker, ExecutionPlanner
64-bit checkbox changes nothing	Current known inert control	Treat output as the engine's default precision; do not claim 64-bit output	StackingTabUiBuilder, stage parameter sources
Dark Multiplication Factor changes nothing	With dark optimization on, a single-channel light of at least 512×512 pixels ignores the entered factor and the engine solves its own value	Leave at default; use matching darks and read the run's warnings, which name each calibration whose lights the entered factor scaled	LightCalibrationSettingsDialog, ImageCalibration.h (prepareImageCalibration), CalibrationMasterSet
Report missing after a run	No completed integration result to report, archive write failed, or run stopped before report generation	Review Process Output/log; use a completed archive row; verify output/history write access	WorkflowArchiver, WorkflowProgressDialogV2
Old history disappeared	20-entry or 5-GiB retention limit removed oldest runs	Copy important archives elsewhere; run history is not backup	WorkflowArchiver::cleanupOldArchives
Cancelled run left new masters	Earlier groups completed and were committed before cancellation	Inspect the cancelled archive and committed-subset baseline; rerun/reprocess intentionally	WorkflowExecutor, WorkflowArchiver

Related workflow steps

• Before integration, organize/correct source data with Sequences, Extract Sequence, or Debayer only when that separate handoff is required. Image Integration itself expects still files and normally calibrates RAW/CFA before demosaic.
• Use Image Calibration when you need only a calibration batch without the persistent full workflow.
• Use Image Registration for a focused registration/alignment job or diagnosis.
• Use Plate Solving to prepare/diagnose WCS and solver configuration.
• Use Image Stacking when inputs are already calibrated and registered and you only need combination/rejection/drizzle controls.
• Use Image Integration Lite for the simplified front end; it delegates shared workflow behavior with fewer exposed decisions.
• After a linear master is accepted, continue with Auto Gradient Removal, Spectrophotometric Color Calibration, and a controlled stretch such as Histogram Transformation.
• See Saving And Exporting for delivering final images; workflow archives are reproducibility records, not final publication exports.

References and provenance

• AfterNight production and test sources are the primary behavior source.
• FITS and SIP terminology follows the FITS WCS/SIP convention implemented by the engine's SipDistortion.h; no external behavior beyond the implementation was inferred for this page.
• Drizzle is identified in the UI as Variable-Pixel Linear Reconstruction; the documented scale/kernel/pixel-fraction behavior is based on the engine's drizzle (DrizzleExecutor.h, DrizzleCanvas.h, DrizzleMapDescriptor.h).
• Qt standard-path and settings locations are described symbolically because their concrete directories are platform/user dependent.
• No external catalog, solver, or approved astrophotography dataset was executed.