ASI676MC Purple Frame Repair Details - aaronwmorris/indi-allsky GitHub Wiki

ASI676MC Purple Frame Repair Details

For setup and everyday use, start with ASI676MC Known Issues. This reference explains the detection and repair process, calibration evidence, advanced settings, and implementation details.

Overview

Some ZWO ASI676MC cameras occasionally deliver a RAW16 RGGB mosaic with a strong purple cast. This feature can detect that specific failure, exclude the affected frame without changing its pixels, or repair it before the normal indi-allsky image pipeline continues.

Leave Enable ASI676MC purple-frame handling off unless the camera actually produces this failure. New configurations start in Detect and exclude only mode so purple frames are preserved while calibration evidence is collected. Repair begins only after that option is deliberately turned off.

Safety boundary

Handling is limited to captures that meet every one of these conditions:

  • the master option is enabled;
  • the camera name identifies an ASI676MC;
  • the input is a two-dimensional, even-sized RAW16 mosaic;
  • the Bayer pattern is RGGB;
  • binning is 1x1; and
  • XBAYROFF and YBAYROFF are both zero.

An unsupported ASI676MC RAW layout is not changed. The reason is recorded in image metadata and written to the log. Other camera models stay outside this feature and are mentioned only at debug log level; they do not receive ASI676MC repair metadata.

The step-by-step guide covers collecting frames, running calibration, and enabling repair.

Runtime behavior

ASI676MC handling runs immediately after indi-allsky opens the camera FITS and before dark-frame calibration, debayering, stacking, or standard FITS saving. The detector and repair therefore operate on the original RGGB parity layout.

The image worker follows this sequence:

  1. Confirm that the feature, camera, and RAW layout are eligible.
  2. Sample the four RGGB parities and calculate the overall purple, red-side, and blue-side ratios.
  3. Classify the frame using all three configured thresholds.
  4. In Detect and exclude only mode, keep a purple frame unchanged and mark it excluded.
  5. In repair mode, repair a temporary copy of a purple frame.
  6. Recalculate the signature on the temporary copy.
  7. Commit the copy only if it no longer matches the failure signature and passes the signal sanity check, which rejects collapsed, blank, or wildly rescaled output.

The last step is atomic from the rest of the image pipeline's perspective. If validation fails, the original mosaic is retained and the image is marked validation_failed.

Status, outputs, and notifications

The status stored under asi676mc_repair_status is one of:

Status Meaning
normal The frame did not match the purple-frame signature.
excluded A purple frame was detected in Detect and exclude only mode; its pixels were retained and it was excluded from standard timelapses.
repaired A purple frame was repaired and passed post-repair validation.
validation_failed The repair did not pass validation; the original frame was retained and excluded from standard timelapses.
skipped The camera or RAW layout did not meet the safety boundary, or the configuration was invalid.

Frames with excluded or validation_failed status are also omitted from stacking history and automatic exposure control, so they cannot affect the next normal image.

Metadata also retains the available before/after ratios, timing, and reason text. Gallery decoration is optional and does not affect processing.

A skipped check or failed post-repair validation creates a Camera notification. The first occurrence appears immediately; identical notification types are suppressed for two hours. Normal frames, expected exclusion-only detections, and successful repairs remain visible through their logs and gallery metadata without generating notifications.

Detection and repair model

The detector sparsely samples the four Bayer parities while preserving RGGB alignment. A naturally magenta scene is less likely to be mistaken for the camera failure because all three conditions must be met:

  • PURPLE_RATIO_THRESHOLD measures the combined red/blue dominance over the two green parities;
  • RED_SIDE_RATIO_THRESHOLD requires the red parity to dominate green; and
  • BLUE_SIDE_RATIO_THRESHOLD requires the blue parity to dominate green.

The repair runs in bounded row chunks to limit peak memory use. It:

  1. removes the observed one-row phase error;
  2. applies a calibrated lookup table to each RGGB parity;
  3. records prematurely clipped green locations before gain correction;
  4. reconstructs those green highlights from the repaired red/blue context;
  5. blends strongly coloured and near-neutral highlight estimates across the configured highlight boundaries; and
  6. copies a stable neighbouring row into the final row, whose source is lost by the phase correction.

The runtime and calibration engine share the detector, repair function, highlight-boundary conversion, and defaults. There is no second correction algorithm that can drift out of sync.

Settings reference

All keys live below IMAGE_ASI676MC_REPAIR. In the web interface they appear in the ASI676MC purple-frame handling card under Config > Image.

Purple-frame options

Web control Key Default Purpose
Enable ASI676MC purple-frame handling ENABLE False Master safety switch. Leave off for unaffected cameras.
Detect and exclude only EXCLUDE_ONLY True Detect, flag, and exclude purple frames without changing pixels. Disable only after reviewing calibration.
Log every ASI676MC frame LOG_EVERY_FRAME False Log normal-frame checks at info level instead of debug level. Purple frames and failures are always logged prominently.
Show purple-frame status in gallery GALLERY_ENABLE False Show repaired, excluded, and failed states and their filters. The excluded filter does not include manually excluded frames. Turning off the master feature also turns this option off.
Save purple and following normal FITS for calibration SAVE_DIAGNOSTIC_FITS False Save the untouched purple frame and its immediately following compatible frame as the preferred low-disk evidence. A successor with different capture/layout metadata breaks the group and is not retained as its reference.
Also save the preceding normal FITS SAVE_PRECEDING_FITS False Cache the immediately preceding normal FITS and save it when the next compatible frame is purple. Requires the parent diagnostic option.

Advanced options

These numerical controls should normally be set by the calibration tool; manual tuning is not recommended.

Web control Key Default Purpose
Overall purple-frame threshold PURPLE_RATIO_THRESHOLD 1.5 Combined purple signature threshold.
Red-side purple-frame threshold RED_SIDE_RATIO_THRESHOLD 1.15 Red-side signature threshold.
Blue-side purple-frame threshold BLUE_SIDE_RATIO_THRESHOLD 1.75 Blue-side signature threshold.
Detection sample step SAMPLE_STEP 32 Even Bayer-preserving interval used by live detection.
Clipped-highlight brightness level SOURCE_SATURATION_THRESHOLD 65000 Faulty-stream level treated as clipped for green reconstruction.
Red repair gain GAIN_R 0.91004 Repair gain for the red parity.
First green repair gain GAIN_G1 1.68652 Repair gain for the first green parity.
Second green repair gain GAIN_G2 1.09238 Repair gain for the second green parity.
Blue repair gain GAIN_B 0.59537 Repair gain for the blue parity.
Highlight correction start HIGHLIGHT_BLEND_START_RATIO 0.55 Red/blue balance at which highlight blending begins.
Highlight correction end HIGHLIGHT_BLEND_END_RATIO 0.75 Red/blue balance at which the near-neutral estimate takes over.
Rows processed at once CHUNK_ROWS 128 Even number of rows processed at a time during repair.

A full calibration derives only the four parity gains, the clipped-highlight brightness level, and the two highlight blend ratios. A preliminary analysis may suggest detection thresholds, but it never changes them automatically. Sample step, chunk size, and operational switches are never derived.

Collecting calibration FITS

Calibration needs the untouched purple mosaic and at least one compatible normal reference for each purple frame. A normal/purple/normal triplet is preferred because the two normal frames can be averaged; a normal/purple pair is also valid. The final evidence must include at least seven purple frames, seven distinct normal references, and two meaningfully different exposure settings.

Recommended: low-disk diagnostic capture

Enable Save purple and following normal FITS for calibration. For each detected purple frame, indi-allsky copies the original camera FITS before a temporary repaired array can affect later processing, then saves the immediately following compatible frame. This option adds no persistent full-frame memory cache.

Optionally enable Also save the preceding normal FITS. The image worker keeps up to one untouched normal FITS byte string per active camera and writes it only if the next compatible frame is purple. If that normal frame was already saved as the following member of an earlier group, indi-allsky reuses the database FITS rather than retaining a duplicate byte string. The cache advances after each normal frame and is discarded after a purple, skipped, or incompatible frame.

Each diagnostic FITS is a normal database-managed FITS asset with role metadata:

  • preceding - normal frame immediately before the purple frame;
  • bad - untouched purple frame; or
  • following - frame immediately after the purple frame.

One physical FITS can carry more than one role when consecutive groups share a frame. Diagnostic files use the normal FITS upload destinations, retention setting, and expiration task. They can be downloaded from the standard Image Viewer when the relevant download controls are available.

Diagnostic and standard FITS saving remain independent. If both paths select the same exposure, both files are retained: the diagnostic asset preserves the untouched input, while the standard asset follows the normal processing settings.

Fallback: standard FITS capture

If diagnostic saving does not catch the purple frame, temporarily set standard FITS saving to Every Image. A periodic interval cannot reliably capture a random failure.

In Detect and exclude only mode, standard FITS retain the original mosaic and can provide calibration evidence. When repair is active, standard FITS are written from the already repaired image and may no longer contain the failure; use diagnostic saving when untouched evidence is required.

The same distinction applies when Save FITS Pre-Calibration is enabled. ASI676MC handling runs before the optional pre-dark save point, so "pre-calibration" refers to dark-frame calibration, not purple-frame handling. In Detect and exclude only mode, both standard save paths retain the original mosaic.

The settings and calibration pages show a combined capture outlook based on the current repair and FITS options. That message describes future captures; saved-FITS discovery judges existing files by their actual metadata and availability.

Calibration tool

The tool appears under Tools > Fix ASI676MC purple frames only when:

  • the ASI676MC master switch is enabled;
  • a visible local camera is positively identified as an ASI676MC; and
  • the current user can save settings on the Config page.

Direct calibration URLs enforce the same conditions. With normal authentication, the user must be an administrator. With LOGIN_DISABLED, the tool follows the anonymous access policy used by Config while still separating private calibration ownership with a browser-specific token.

Multiple ASI676MC units

The camera selector binds evidence, session ownership, and save-time identity checks to one physical camera record. It does not create a per-camera repair profile. IMAGE_ASI676MC_REPAIR remains one installation-wide configuration block, so every ASI676MC on that installation uses the same applied thresholds and repair constants.

That design suits the usual one-camera installation. An installation that regularly swaps between or processes data from multiple physical ASI676MC units must keep one shared profile or recalibrate when the active unit changes. A per-camera repair profile would require configuration and runtime changes, not only a calibration-page change.

Choose an evidence source

The page provides two input paths. Use saved FITS is the preferred path for database-managed evidence; Upload a FITS collection accepts a collection that the user selected manually.

Use saved FITS

Choose a target of 7 to 30 purple-frame groups. The background job enumerates the complete configured FITS retention period, inspects every eligible row, and stages up to the requested number of usable groups. If fewer than seven groups exist, it reports that the retained archive was exhausted. Progress covers catalog enumeration, current-detector checks, missed-purple population discovery, fitting, and validation. Cancel search remains available during inspection and staging.

Existing diagnostic roles and purple-frame database flags provide context but do not decide admission or inferred populations. This allows discovery to find failures outside the configured detector thresholds. Untouched diagnostic RAW FITS and standard FITS from Detect and exclude only remain eligible and can be mixed as pairs or triplets. A standard FITS associated with a successfully repaired frame is excluded because it contains the corrected mosaic.

New FITS rows store the three measured detector ratios as small, threshold-independent metadata. Discovery can reclassify those captures without decoding their full images. Older rows without cached ratios remain eligible and are inspected directly; successful measurements are backfilled in batches for later searches. A lightweight header check still rejects a repaired ASI676FX FITS. Database discovery can also open indi-allsky's gzip-compressed FITS.

The database row's validated camera binding supplies the ASI676MC identity when a file's camera headers are absent or generic. An explicit conflicting camera identity remains authoritative and is rejected. Missing, corrupt, repaired, or incompatible files are skipped and summarized rather than silently shortening the search horizon.

The 200-FITS and 2.5-GiB limits apply to the final purple/reference evidence set, not to catalog discovery. Only selected evidence enters the private session. The saved-FITS path stages up to three groups beyond the requested count as validation reserves. If a group passes the main repair checks, but the complete repair is less than ten percent better than a simple colour-only correction, that inconclusive group is set aside, a reserve is promoted, and the fit is repeated. The complete repair must still outperform colour-only correction; otherwise calibration fails. The group count is reduced only when no reserve remains, and never below the seven-group minimum.

A hard link keeps staged evidence stable without a second copy; where hard links are unavailable, the tool makes a private copy. It never uses a symbolic link whose target could change after selection. If a selected source disappears during staging, that file is skipped and the remaining evidence is still evaluated.

Cancellation removes only private links, copies, or partial copies. Source FITS pixels and headers are never changed. The only durable discovery update is the threshold-independent ratio cache on successfully inspected legacy rows.

Upload a FITS collection

Select 14 to 80 uncompressed .fit, .fits, or .fts files. The browser uploads them sequentially so a Raspberry Pi does not need to buffer the full collection in one request. Browser and server enforce these limits:

  • 80 files per session;
  • 256 MiB per file; and
  • 2 GiB total per session.

Fourteen files are enough to start analysis because the tool cannot know the populations before reading them; the count does not guarantee a result. Calibration or threshold discovery still needs seven likely purple frames and distinct compatible normal evidence. Extra unmatched files are ignored only when the remaining collection meets all evidence rules.

Uploads are bound to the selected, currently available ASI676MC by database ID, UUID, and detected device name. This also provides a narrow camera-bound legacy compatibility path for standard FITS written with indi-allsky's default INSTRUME=indi-allsky header. If all camera-bearing headers are absent or generic, the bound identity may be used after every RAW-layout check passes. An explicitly different camera is always rejected. Standalone and non-camera-bound uses remain strict, and the result and text report state how many uploaded files used this legacy identity path.

The page checks the browser features required for multi-file selection, sequential requests, cancellation, progress polling, and result restoration. It displays a blocking explanation when those features are unavailable. An upload can be cancelled while it is active, and a server-side marker prevents a late request from reviving the cancelled session.

Progress, cancellation, and restoration

After either source passes its input checks, the setup area is replaced by a compact progress view with the current stage, progress bar, status detail, and cancel action. The page scrolls to that view once; polling updates do not keep moving the page. Cancellation stops discovery or private staging at its next safe checkpoint, cancels a queued job, or asks a running worker to stop cooperatively, then restores the input controls.

A failed analysis stays in the compact progress view with its full safe error message and a Try again action. Try again removes the finished failed session and returns to setup. Reloading while a retained calibration is queued, running, complete, failed, or waiting to be reset restores its current view.

How analysis works

Every usable frame must pass the same core checks before it can contribute to either outcome. The engine requires:

  • explicit ASI676MC identity, or the selected camera's identity through the camera-bound legacy path described above;
  • BAYERPAT=RGGB, an even-sized RAW16 mosaic, 1x1 binning, and zero Bayer offsets;
  • finite positive exposure, finite non-negative gain, valid timestamps, and decoded-size limits; and
  • exact frame-shape matching within each purple/reference group.

Primary and image-extension headers are merged. A repaired FITS marked ASI676FX or a file with conflicting camera identity is rejected. Every accepted frame is classified by the same signature implementation used during live capture.

The decision path then proceeds in this order:

  1. Apply the configured detector to every supplied frame. The tool never silently replaces configured thresholds.
  2. If at least seven detected purple frames have compatible adjacent normal evidence, continue to normal calibration.
  3. If fewer than seven are detected, or one configured threshold sits outside an otherwise clean normal/purple gap, check whether all three ratios form two strong, consistently ordered populations.
  4. Return either a full calibration, a preliminary detection-threshold suggestion, or a safe failure. Inferred labels are never used to fit repair constants in the same run.

Database flags and filenames do not decide inferred populations. A threshold suggestion is allowed only when both populations contain at least seven FITS, every ratio has a clean gap of at least ten percent, likely purple frames have compatible adjacent normal references, and at least two exposure settings are represented.

For a full calibration, every used purple frame needs at least one compatible normal reference. The complete evidence must also provide:

  • at least seven distinct normal references;
  • clean normal/purple signature separation;
  • sufficient unsaturated samples for each Bayer parity;
  • sufficient clipped-highlight evidence for both blend boundaries;
  • plausible gains with low between-pair median absolute deviation; and
  • a repair materially closer to nearby normal evidence than both the original frame and a colour-only correction.

Normal references that change too much across a two-sided triplet are rejected at the affected sample locations, limiting the influence of cloud, aircraft, or other transient content. Robust medians and median-absolute-deviation filtering reduce the effect of remaining outliers.

The engine then estimates the four parity gains, finds the premature clipping plateau, searches the supported highlight-boundary grid, and validates the rounded values by running every matched purple frame through the live repair path. Calibration fails if any repaired frame still matches the failure signature.

Possible outcomes

Full calibration

A successful result shows the derived and currently configured values, matched purple and normal counts, pair/triplet and exposure coverage, signature ranges, unused or rejected evidence, warnings, and whether the result effectively matches the configuration within a small tolerance.

Two-sided evidence is considered complete enough when at least 90 percent of matched purple frames belong to normal/purple/normal triplets. Below that guideline, the result recommends gathering more complete groups. Reused normal references remain a separate confidence warning.

A calibration remains successful when a configured detection threshold sits within fifteen percent of either edge of its observed safe gap. A dedicated table shows the current value, normal/purple range, and midpoint, while the guidance recommends collecting more varied evidence before changing that threshold. Repair constants remain valid.

Detection settings need adjustment

This is a preliminary result, not a calibration. It contains no repair constants and changes no settings during analysis. The repair-value table is hidden. Instead, the page shows current and recommended detection thresholds, their observed safe intervals, population and adjacency evidence, and an instruction to review the files and rerun.

When representative FITS can be rendered, the page shows up to two likely purple frames and two likely normal frames for both uploaded and saved collections. The images come from the original, unrepaired FITS and use one brightness stretch across all three colour channels so the colour difference remains visible. Previews are best-effort visual aids. If one cannot be created, the other previews and the authoritative filename, capture-time, and ratio table remain available; the valid threshold result is not discarded.

The user must confirm, after checking the previews when available plus the filenames and capture times, that the files listed as likely purple show the actual camera failure. Saving stays blocked until the confirmation is checked. A user who can save Config may then choose Save detection settings. Only fields marked Change recommended are saved; a configured threshold already inside its safe gap is retained instead of being replaced by a cosmetically different midpoint. Select Start over and rerun calibration after saving.

Analysis cannot establish a safe result

Overlapping, insufficient, or inconsistently ordered populations produce a plain-language explanation rather than an unsafe suggestion. The same applies when evidence validation or repair fitting cannot meet its safety checks. No confirmation is offered and no settings are changed. The retained failure view shows what to check before selecting Try again.

Report and applying values

Download details produces a text report for both full and preliminary results. It records the evidence source, quality assessment, values, configuration comparison from the start of the run, warnings, cleanup outcome, and recommended next actions. It never exposes private staging paths. The filename begins with local completion time in YYYY-MM-DD_HH-MM-SS_asi676mc_calibration_report.txt format.

Known rejection and safety-check reasons are translated into concise explanations with corrective actions. If every selected file is rejected, the failure view groups reasons and affected file counts. Unknown exceptions stay private and direct the user to the indi-allsky log. The background task list uses the same safe wording; if its short status cannot contain every reason, it directs the user to the retained failure view.

For a full result, Save calibration values writes only the seven derived repair keys through the normal configuration-save mechanism and queues an application reload. It refuses the update if the configuration or selected ASI676MC identity changed after the run began. Operational switches and detection thresholds remain unchanged.

For a preliminary result, Save detection settings can write only the explicitly recommended threshold keys after population confirmation. It cannot write repair constants.

Results are retained under unguessable session IDs so the user can leave and return. Per-tab state plus a shared list prevents tabs from replacing each other's run; after one run is cleared, the next retained run is restored. Transient polling failures retry without discarding the handle. The browser stores no FITS data or calibration values, and every result, report, discard, and apply request rechecks ownership and camera binding. Start over discards the retained result.

Session storage and cleanup

Private sessions live below the non-public Flask instance directory, or below ASI676MC_CALIBRATION_FOLDER when explicitly configured. Manifest and result files are written atomically because web and worker processes may access them at the same time.

CPU-heavy fitting and full-resolution validation run on a dedicated serial calibration thread in the low-priority video-worker process. The normal video queue remains available for timelapse and keogram work. Cross-process locks protect session changes, and a queued job can be claimed only once. Each owner may have two active sessions; the installation may have four.

Saved-FITS searches, queued jobs, population analysis, pairing, and fitting can all be cancelled at bounded checkpoints. Uploads and queued jobs become stale after 30 minutes; a running job is failed after two hours without a worker heartbeat. These safeguards prevent abandoned work from occupying a session quota indefinitely.

Cleanup depends on the source:

  • uploaded FITS are deleted immediately after success or failure;
  • private hard links or copies from database discovery are deleted immediately, while source FITS remain untouched; and
  • cancellation and reset remove the session's private inputs directly.

If a crash, power loss, or unavailable filesystem interrupts immediate cleanup, the regular expiration task removes sessions older than seven days. That fixed session retention is independent of database FITS retention; database FITS and diagnostic role links continue to follow normal indi-allsky expiration rules.

Troubleshooting

The tool is absent from the menu

Connect a visible local ASI676MC, enable Enable ASI676MC purple-frame handling, and use an account that can save Config. When login is disabled, the tool follows Config's open-access policy.

Fewer than 14 eligible saved FITS are found

Current settings describe future capture, not the retained archive. Existing purple frames may have expired or may never have been saved untouched. Prefer diagnostic FITS and collect more events. If diagnostics do not catch the failure, temporarily use Detect and exclude only with standard FITS saving set to every image.

Files upload but the collection is rejected

Confirm that the collection contains uncompressed ASI676MC RAW16 RGGB files, seven purple events with compatible adjacent normal references, two exposure levels, and visible clipped highlights. Extra files do not compensate for a missing minimum. Default indi-allsky standard FITS with only the generic INSTRUME=indi-allsky header are accepted through the camera-bound legacy path; an explicit different-camera header is not.

The retained error groups known rejection reasons and says what to correct. If it names a detection ratio, compare the likely-normal and likely-purple evidence. Clean populations produce a preliminary suggestion; unclear populations fail safely. Do not change a threshold unless the listed likely purple files show the actual camera failure.

No preview images appear with a threshold suggestion

Use the complete filename, capture-time, and ratio table to confirm the populations. A preview is optional and does not replace or weaken the evidence checks behind the suggestion.

The report time differs from UTC

Browser and report timestamps intentionally use local time, including timezone name and UTC offset where space permits. Private manifests retain UTC so expiration and communication between web and worker processes remain unambiguous.

Saving values is refused

The indi-allsky configuration changed after the run began, the bound camera is no longer the same local ASI676MC, or the user cannot save Config. Start a new calibration against the current configuration and camera.

Repair validation fails during capture

The original frame is preserved. Collect its untouched diagnostic group and recalibrate; do not loosen detection thresholds merely to suppress the warning.

Maintainer map

File Responsibility
indi_allsky/config.py Conservative new-install defaults, including disabled handling and Detect and exclude only mode.
indi_allsky/asi676mc.py Authoritative settings normalization, signature detection, repair, validation, metadata, and diagnostic-role helpers.
indi_allsky/processing.py Runtime eligibility checks, Detect and exclude only behavior, repair invocation, logging, and actionable notifications.
indi_allsky/image.py Untouched diagnostic FITS capture, optional preceding-frame cache, database records, and rendered-image metadata.
indi_allsky/asi676mc_calibration_engine.py FITS inspection, matching, evidence policy, numerical fitting, and validation through the live repair path.
indi_allsky/asi676mc_calibration.py Private sessions, uploads, database discovery/staging, population previews, cleanup, user guidance, result comparison, and text reports.
indi_allsky/video.py Dedicated serial calibration consumer, retained-FITS database access and ratio-cache backfill, plus seven-day session cleanup fallback.
indi_allsky/flask/forms.py Settings validators, calibration controls, saved-FITS lookup, viewer assets, and gallery filtering.
indi_allsky/flask/views.py Authenticated endpoints, task queueing, result/report access, and configuration apply/reload.
indi_allsky/flask/base_views.py and templates/base.html Context-aware Tools-menu visibility in the current navigation.
indi_allsky/flask/templates/config.html, config/image.html, and config/asi676mc.html Configuration save integration, responsive settings card, and contextual guidance.
indi_allsky/flask/templates/asi676mc_calibration.html Calibration setup/progress/result transitions, population previews, cancellation, reports, reset, and browser capability checks.
indi_allsky/flask/templates/gallery.html Optional repair/exclusion badges, outlines, and status-specific filtering.
indi_allsky/flask/templates/imageviewer.html Diagnostic preceding/purple/following FITS downloads.
testing/image/test_asi676mc_repair.py Detection, repair, validation, metadata, and diagnostic helper coverage.
testing/image/test_asi676mc_calibration_engine.py FITS inspection, matching, fitting, evidence policy, and shared-runtime coverage.
testing/image/test_asi676mc_web_calibration.py Sessions, cleanup, discovery, guidance, reports, and web workflow coverage.

Activate the project's Python environment, then run the focused tests from the repository root:

python -m pytest \
    testing/image/test_asi676mc_repair.py \
    testing/image/test_asi676mc_calibration_engine.py \
    testing/image/test_asi676mc_web_calibration.py

When changing correction math, update indi_allsky/asi676mc.py first. The calibration engine must continue to call that implementation for classification and final validation. When changing a setting or diagnostic role, trace it through defaults, form validation, view persistence, help text, metadata, report output, and tests so the complete feature remains coherent.