Star Correction

Correct star shape -- roundness and elongation, including the FWHM inflation denoise introduces -- without sharpening or deblurring the background or nebulosity.

schema 1 · revision 2

Recognize it

Stars are slightly elongated or irregular after denoise, especially on dense fields (globular clusters, open clusters).
Star FWHM measurably increased relative to the pre-denoise image.

Intended output

Rounder, tighter stars, with background and nebulosity that receive no intentional sharpening -- though RC-Astro documents that PSF correction itself can still visibly affect nonstellar structure near well-supported stars, so 'unchanged background' is checked per image, not assumed by construction.

Required input state

Linear, unstretched data, after denoise -- this step exists specifically to correct what denoise did to star shapes.

What NOVA does

NOVA runs bxt_star_correct: RC-Astro's BlurXTerminator, correct-only mode, dispatched via the RC-Astro CLI/RunPod GPU path (rcastro_gpu.run_rcastro) -- not PixInsight -- with sharpen_stars, sharpen_nonstellar, and adjust_halos all forced to zero and automatic PSF detection, most consequential on dense star fields where denoise inflates FWHM. Unlike bxt_deconvolve and denoise_nxt, this function does not call the shared _detect_and_fix_flip orientation check. cc_stellar (Cosmic Clarity stellar-only sharpening) is a separate ontology experiment variant, forced by the public_free_core workflow's own default for its free-tool-only path -- a deliberate per-workflow choice, not a runtime fallback triggered when BXT or PixInsight is unavailable; the ontology's own 'Fallback if PI unavailable' description does not describe current dispatch behavior. The M66 execution record cited below has no isolated step_verdicts entry for star_sharpen, so it is evidence for that run overall, not standalone validation evidence for this step.

Use when

Denoise measurably inflated star FWHM or introduced visible elongation on a dense star field.

Skip when

Stars are already round and FWHM-stable after denoise; there is nothing to correct.
Elongation is severe trailing from an acquisition-time tracking/guiding failure rather than the mild FWHM inflation denoise introduces -- Correct Only is not a substitute for fixing tracking/guiding at the acquisition stage, even though RC-Astro documents guiding error as one of the aberration classes the tool is trained on.

Limits

This is pure geometric correction, not sharpening or denoising -- 'Star Sharpen' is a misleading name for a step whose sharpening amounts are deliberately zero.
This step happens before the star/starless split (remove_stars_linear); it is not the same operation and does not create or use a separate star layer.
RC-Astro documents BlurXTerminator as trained on common acquisition-time aberrations including guiding error, correctable in limited amounts per the vendor's own technical manual, and recommends Correct Only as a first pass on images affected by coma or guiding error before a second, manual-PSF pass -- so a categorical 'cannot fix guiding-error elongation' claim overstates the tool's documented limits. NOVA has not independently measured this step's effectiveness on guiding-error elongation specifically, and Correct Only is not a substitute for fixing tracking/guiding at the acquisition stage.
RC-Astro documents that correction is applied to nonstellar features as well as stars wherever stellar support exists nearby -- Correct Only disables NOVA's sharpening amounts (bxt_stars/bxt_nonstellar forced to 0.0), but does not confine PSF correction to star pixels alone, and in some cases the correction itself can visibly resemble sharpening.

Scientific and aesthetic notes

Jeff's own manual verification renamed this step from 'Star Sharpen' to 'Star Correction -- BlurXTerminator Correct Only' precisely because the old name implied sharpening that does not happen here; the taxonomy's family name reflects that correction.
Do not equate BlurXTerminator Correct Only with Cosmic Clarity stellar-only sharpening -- both improve star shape but are different engines and need their own before/after check.
A BXT-Correct-Only-vs-Cosmic-Clarity-stellar comparison is a strategy comparison -- two different engines with different internals -- not a clean causal test of one shared algorithm factor; a stronger experiment adds a true no-op (uncorrected) baseline alongside both engines, not just the two engines against each other.
Star population counts must be matched between before/after measurements: the same detection threshold, crop, and plate scale. A drop in detected star count can come from a stricter threshold rather than from worse star shape, and the two must not be conflated.
RC-Astro's guiding-error and nonstellar-correction capability is vendor-documented and NOVA's dispatch path is source-confirmed; neither establishes that one correction strategy is universally superior across all dense star fields -- that requires its own controlled comparison.

Evidence and measurements

Measurements	Acceptance criteria
Star FWHM and eccentricity from matched before/after luminance extractions. Star support population (count of stars measured), noted when it differs between before/after. Noise evaluation as a non-regression check, not a success metric.	Eccentricity decreases without material FWHM growth. Star support population is comparable before and after, using the same detection threshold, crop, and plate scale -- a fewer-stars result from a stricter threshold is not evidence of worse shape. No new halos, ringing, lost detail, or meaningful noise regression. Background and nebulosity show no intentional sharpening; any visible change to nonstellar structure is attributable to PSF correction near well-supported stars, not an enhancement amount, consistent with RC-Astro's documented behavior.

Measurements

Acceptance criteria

Star FWHM and eccentricity from matched before/after luminance extractions.
Star support population (count of stars measured), noted when it differs between before/after.
Noise evaluation as a non-regression check, not a success metric.

Eccentricity decreases without material FWHM growth.
Star support population is comparable before and after, using the same detection threshold, crop, and plate scale -- a fewer-stars result from a stricter threshold is not evidence of worse shape.
No new halos, ringing, lost detail, or meaningful noise regression.
Background and nebulosity show no intentional sharpening; any visible change to nonstellar structure is attributable to PSF correction near well-supported stars, not an enhancement amount, consistent with RC-Astro's documented behavior.

Cross-tool matrix

Tool	Equivalence	Expected result	Evidence
nova	Exact replay	Lower eccentricity, stable-or-improved FWHM, comparable star support; background and nebulosity receive no intentional sharpening, though visible PSF-correction effects on nonstellar structure near well-supported stars are possible and should be checked, not assumed absent.	nova-star-correction-source, nova-m66-run-1.24.7
pixinsight	Exact replay	Lower eccentricity without material FWHM growth, halos, ringing, lost detail, or noise regression.	m66-manual-verification
siril	Same engine, adapted host	Matches the PixInsight/NOVA result closely; same engine, different host.	m66-manual-verification
saspro	Functional alternative	Improved star roundness with the background and nebulosity unaffected.	saspro-source-1.18.0, m66-manual-verification

NOVA Python pipeline workflow 1.24.7

Run immediately after denoise, on the active linear image -- the star/starless split has not happened yet at this point in the pipeline.
Use automatic PSF detection; leave all sharpen/halo amounts at zero.
Compare matched luminance extractions (identical FWHM/eccentricity settings) before and after.
Dispatch is via RC-Astro CLI/RunPod GPU (rcastro_gpu.run_rcastro), not PixInsight -- there is no PixInsight step in this function's own execution path.

Controls and starting ranges

mode: Correct Only (--correct-only)
PSF: automatic detection
sharpen_stars / sharpen_nonstellar / adjust_halos: 0.0 / 0.0 / 0.0 -- forced, not tunable
dispatch: RunPod GPU endpoint first, local RC-Astro CLI (CPU) second -- no further fallback engine

Mask behavior: Not applicable; BXT Correct Only operates on the full image.

Failure modes

Confusing this step's output with a separate star layer -- none exists yet at this point.
Treating cc_stellar (the public_free_core workflow's forced alternative) as an interchangeable substitute rather than a different engine needing its own comparison.

Recovery

Revert to the pre-correction linear image and re-check the denoise step that preceded it.

PixInsight BlurXTerminator, Correct Only Core 1.9.3 Lockhart host; BlurXTerminator version not independently confirmed in this repo

Run BlurXTerminator on the active linear image with Correct Only enabled and Sharpen Stars, Adjust Star Halos, and Sharpen Nonstellar all at 0.
Extract identical before/after luminance copies with the same FWHMEccentricity sensitivity, upper limit, and PSF model.
Use noise evaluation as a non-regression check only; Correct Only is not a denoiser.

Controls and starting ranges

M66 verified: Correct Only on; Sharpen Stars 0; Adjust Star Halos 0; Sharpen Nonstellar 0; automatic PSF

Mask behavior: Not applicable; full-image operation.

Failure modes

Measuring before/after populations with differing star-support counts and treating the delta as precise.
Retaining an older window/file name after later operations and mistaking it for a different input stage.

Recovery

Undo and re-run with matched extraction settings for a valid comparison.

Siril via RC-Astro > BlurXTerminator.py, Correct Only (licensed external integration) RC Astro BlurXTerminator wrapper model v4, verified in Siril 1.4.4

Open Scripts > Python Scripts > RC-Astro > BlurXTerminator.py with Correct Only enabled.
Leave Sharpen Stars, Adjust Star Halos, and Sharpen Nonstellar all at 0.
Automatic PSF is disabled/irrelevant in this mode; do not try to tune it.

Controls and starting ranges

M66 verified: Correct Only on; Sharpen Stars 0.00; Adjust Star Halos 0.00; Sharpen Nonstellar 0.00; Tile Overlap 0.20

Mask behavior: Not demonstrated; full-image operation.

Failure modes

Siril has no native equivalent -- attempting a native star-shape-only correction without this integration.

Recovery

Revert to the pre-correction linear image.

Seti Astro Suite Pro Cosmic Clarity Sharpen, stellar-only 1.18.0 source-inspected; procedure discussed but not independently tested at this exact stage

Apply Cosmic Clarity Sharpen in Stellar Only mode to the active linear image -- not 'the star layer', which does not exist yet at this point in the pipeline.
Keep non-stellar sharpening disabled so this stays a star-shape-only pass.
Do not treat this as the same operation as BlurXTerminator Correct Only; compare before/after star profiles independently, since they are different engines.

Controls and starting ranges

mode: Stellar Only; non-stellar sharpening disabled

Mask behavior: Not demonstrated; full-image operation.

Failure modes

Assuming stellar-only Cosmic Clarity Sharpen and BXT Correct Only are interchangeable without their own comparison.

Recovery

Revert to the pre-correction linear image, or defer star-shape correction to PixInsight BXT Correct Only.

Validation status

Not yet Jeff-validated. Use this as sourced guidance, not a certification.

Sources

M66 manual recipe verification feedback M66_Manual_Recipe_Verification_Feedback.md — Jeff validated
NOVA BlurXTerminator correct-only star-shape implementation (dispatched via RC-Astro CLI/RunPod GPU, not PixInsight) and ontology nas_server/seti_astro.py:2783; nas_server/processing_ontology.json:961 — NOVA source-confirmed
M66 workflow 1.24.7 execution record critiques/20260802_045251_seestar_galaxy.json — NOVA execution record
Installed Seti Astro Suite Pro 1.18.0 processing source saspro/pedestal.py; saspro/numba_utils.py; saspro/stacking_suite.py; saspro/abe.py; saspro/sssc.py — Artifact-confirmed
RC-Astro BlurXTerminator Technical Manual — Vendor-documented

Related evidence

The M66 recipe and Jeff-validation ledger will link here when their static exports are available. Until then, this article remains explicitly unvalidated.