Statistical Stretch

Purpose and expected result

Statistical Stretch derives a black point and source median from the target, then uses one rational transfer to place that median near Target Median while keeping the computed black and normalized white anchors at 0 and 1. It can derive one pooled transform for RGB (Linked Stretch) or an independent transform for each RGB channel. Optional post-stages lift the upper range with a fixed piecewise-linear curve and divide by the output maximum.

The intended result is a nonlinear, same-size still image in the source pixel type. On ordinary finite normalized data, defaults usually provide a quick first display stretch. The target value is exact only for the transform's computed median before Apply Curves Adjustment, Normalize Image, mask blending, integer quantization, and output clamping. This is a one-pass mapping, not an iterative solver.

Every sample at or below the computed black point becomes black, and every statistic is the whole image's: the target's for Apply, the preview, and a Preview area, and each frame's own for a sequence, whose frames are each stretched from their own statistics.

At a glance

Item	Current behavior
Menu path	Process > Intensity Transformations > Statistical Stretch
Enabled when	An image or sequence is active, the same process is not already attached to it, and no attached process holds exclusive access
Inputs	One grayscale or three-channel RGB still image or sequence; any other channel count is refused with the error "a statistical stretch needs a mono or RGB source"; unsigned integer samples are read as fractions of their full scale, signed ones from their smallest value to their largest, and floating samples as they are; samples the image marks as holding no data are left out of the statistics; an optional active mask must match the target geometry
Outputs	Destructively replaces a still target after success; Take snapshot creates an auxiliary still without replacing the target; dimensions, channel count, pixel format, and pixel type are preserved; no file is written directly
Recommended image state	Linear, background-corrected, and color-calibrated data before the main display stretch
Preview support	Asynchronous RT preview that is the result Apply writes, plus a histogram of the mask-aware preview; disabling Preview returns the histogram to the source; a Preview area shows the whole target's stretch cropped to the area, exactly what Apply writes there
Sequence support	Supported: every frame is stretched from its own statistics, as the process stretches that frame alone, with progress, cooperative cancellation, and replacement of the sequence only after every frame succeeds
Mask support	Supported for preview, Apply, and sequences; white applies the result, black keeps the source, gray blends, and Invert Mask reverses the weights
External dependencies	None; execution uses the bundled engine and does not access a tool, model, catalog, or network service

Before you start

• Finish background correction and color calibration first. The estimator does not identify gradients, remove a pedestal selectively, or calibrate RGB.
• Use a one-channel or three-channel image. Any other channel count is refused by name when the process runs.
• Prefer floating-point data while stretching. Processing uses single-precision normalized buffers (double precision for a 64-bit floating or 32-bit integer image); integer results are rounded to the nearest code of the original type.
• Leave Black Point Sigma near its default until the clip readout and preview have been checked. Lower values raise the computed black point, and every sample at or below it becomes black.
• No Black Clipping keeps the threshold from rising above the image minimum. Despite its label, the minimum sample is still mapped to black and counted by the <= black point diagnostic.
• A Preview area is a view of the target: it shows the whole target's stretch cropped to the area.
• Duplicate the target or be prepared to use Image History. Apply replaces the current image after processing succeeds.

Procedure

1. Select a calibrated linear still and open Process > Intensity Transformations > Statistical Stretch. Confirm Target Image and leave Preview selected.
2. Start with Target Median: 0.25, Black Point Sigma: 5.0, and all four checkboxes clear. Read the separate black-point/clipped-pixel line below the curve controls; it is not the histogram's endpoint-bin overlay.
3. On RGB, compare Linked Stretch off and on. Off targets each channel's own median and can neutralize a color cast. On pools all RGB samples to derive one transfer, but still evaluates that nonlinear function per channel and does not preserve RGB ratios.
4. Adjust Target Median for overall midtone placement. Do not assume the final measured median must equal the control when curves, normalization, a mask, quantization, or clipping is active.
5. If the computed black point rises above the source minimum, the readout's clipped pixels become black. Increase Black Point Sigma or select No Black Clipping when that loses faint signal.
6. Use Apply Curves Adjustment only when extra upper-midtone/highlight lift is wanted; then increase Curves Boost cautiously. Use Normalize Image only when scaling the global post-curve maximum to 1 is intentional.
7. Inspect histogram endpoints, background, bright stars, channel balance, and mask boundaries at 100% zoom, then select Apply.

Complete parameter and action reference

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
Histogram plot	Graph, normalized intensity/count	Source or current preview; RGB channels overlaid	None	Binning comes from Plot resolution	Shows source or mask-aware result distribution and endpoint clipping; never changes samples	Always present; Preview chooses result versus source	Recalculation adds work; the endpoint counts do not depend on resolution	Display only; shown source/preview channels	Recomputed after source, preview, or resolution changes
Left-drag in histogram	Graph gesture	No readout	None	Hold left button and move in graph	Shows normalized coordinate readout	Histogram is always present	Diagnostic only; does not choose the statistical black point	Display only	Ends on button release
Middle-drag in histogram	Graph gesture	Fitted/unpanned	None	Hold middle button and drag	Pans the graph within its scroll area	Useful only when graph exceeds viewport	No output cost or effect	Display only	Until pan/zoom reset or window closes
Ctrl+mouse wheel in histogram	Graph gesture	Fitted size	None	Up 1.20x; down 0.8x; wheel without Ctrl scrolls normally	Zooms around the cursor	Interactive graph enabled	Larger graph uses more rendering area	Display only	Window lifetime; process Reset does not change it
Zoom in	Toolbar action	Fitted size	None	Each selection multiplies size by 1.30	Enlarges plot	Always available while idle	Plot rendering only	Display only	Window lifetime; process Reset does not change it
Zoom out	Toolbar action	Fitted size	None	Each selection multiplies size by 0.70	Shrinks plot	Always available while idle	Plot rendering only	Display only	Window lifetime; process Reset does not change it
Zoom reset	Toolbar action	—	None	One action	Fits graph and resets scrolling	Always available while idle	Plot rendering only	Display only	Immediate
Show grid	Toggle	Checked	None	Checked/unchecked	Shows or hides graph grid	Always available while idle	Display-only visual aid	Display only	Window lifetime; process Reset does not change it
Show clipping info	Toggle	Checked	None	Checked/unchecked	Shows or hides the share of samples at or below 0 and at or above 1 in the plotted image	Independent of the statistical clip-threshold text line	Counts exact endpoint samples, independent of Plot resolution; values just inside 0 or 1 are not counted	Display only; source or preview histogram	Window lifetime; process Reset does not change it
Absolute scales	Toggle	Unchecked	None	Checked/unchecked	Checked normalizes each visible channel to its own peak; unchecked preserves relative/global height	Always available while idle	Can conceal relative channel strength; never normalizes image data	Display only	Window lifetime; process Reset does not change it
Draw histogram areas	Toggle	Unchecked	None	Checked/unchecked	Selects filled-area or line rendering	Always available while idle	Area mode can obscure overlapping channel lines	Display only	Window lifetime; process Reset does not change it
Plot resolution	Combo box, bins	8-bit (256)	None	6-bit (64), 7-bit (128), 8-bit (256), 9-bit (512), 10-bit (1024), 12-bit (4096), 14-bit (16K), 16-bit (64K), 100 levels, 1000 levels, 10000 levels	More bins show finer distribution; the endpoint counts do not change	Always available while idle	Higher bin counts increase histogram memory/rendering; transform precision is unchanged	Display only	Window lifetime; process Reset does not change it
Target Median	Double value plus slider, normalized intensity	0.25	None	0.01..0.99; three decimals; spin step 0.01; slider is linear over 1,000 positions	Raises or lowers the rational transfer's median anchor before optional post-stages	Enabled while idle	High targets brighten background/noise and compress more upper range; low targets give a darker result	Preview/Apply; pooled or per-channel transform; mask post-blend; stills and every frame of a sequence	Captured per execution; Reset or a successful Apply returns 0.25
Linked Stretch	Checkbox	Unchecked	None	Checked/unchecked	Checked pools RGB samples and uses one black point/median/transfer; unchecked computes each channel independently	Enabled only when current target reports RGB; disabled for mono	Linked does not preserve RGB ratios; unlinked can neutralize or shift channel balance by forcing separate medians	Preview/Apply; RGB only; mask post-blend	Captured per execution; Reset or a successful Apply clears it
Black Point Sigma	Double value plus slider, robust-sigma multiplier	5.0	None	0.5..10.0; two decimals; spin step 0.1; slider is linear over 1,000 positions	Larger values lower the threshold toward the source minimum; smaller values raise it and increase the nominal <= threshold count	Disabled while No Black Clipping is selected	Low settings clip more shadows to black; the spread is estimated from a deterministic sample of at most 400,000 values per channel/pool	Preview/Apply; pooled or per-channel estimator; stills and every frame of a sequence	Captured per execution; Reset or a successful Apply returns 5.0
No Black Clipping	Checkbox	Unchecked	None	Checked/unchecked	Uses the image/channel minimum as black point instead of median - sigma × robust noise	Disables Black Point Sigma	Avoids any sample below the selected black point, but still maps/counts the minimum as black	Preview/Apply; pooled or per-channel; stills and every frame of a sequence	Captured per execution; Reset or a successful Apply clears it
Normalize Image	Checkbox	Unchecked	None	Checked/unchecked	After the stretch and optional curve, divides every processed sample by the single global maximum when that maximum is positive	Enabled while idle	Forces the brightest sample to 1; changes the requested median and can amplify noise; one global RGB maximum preserves inter-channel scaling only at this stage	Preview/Apply; whole mono/RGB output; stills and every frame of a sequence	Captured per execution; Reset or a successful Apply clears it
Apply Curves Adjustment	Checkbox	Unchecked	None	Checked/unchecked	Adds the fixed six-anchor piecewise-linear upper-range lift after the rational stretch	Enables Curves Boost	Can brighten upper midtones, flatten highlights, and move the measured median/maximum	Preview/Apply; same curve on every stored sample; stills and every frame of a sequence	Captured per execution; Reset or a successful Apply clears it
Curves Boost	Double value plus slider, unitless	0.20	None	0..1; two decimals; spin step 0.05; slider is linear over 1,000 positions	Raises two anchors above Target Median; 0 is identity and 1 drives both upper anchors to white	Enabled only with Apply Curves Adjustment	Strong values can create steep transitions and broad near-white highlights	Preview/Apply; all processed channels; stills and every frame of a sequence	Captured per execution even when ignored; Reset or a successful Apply returns 0.20
Black clipping readout	Selectable diagnostic text	Computed from current target and defaults	None	Threshold shown to six decimals; count/total and percentage to four decimals	Updates for source, Linked Stretch, sigma, and No Black Clipping	RGB-unlinked shows R/G/B thresholds; RGB count is a pixel count where any channel is <= its threshold	Count is exact over the current precomputed image, but threshold noise uses deterministic subsampling; the text counts exactly the valid pixels the stretch sends to black in at least one channel	Source diagnostic only; not mask-aware; one pixel counted at most once	Recomputed; not saved
Target Image	Image selector	Attached/active target	None	Open eligible image or sequence	Changes source statistics, histogram, preview, and Apply	Active image required; duplicate/exclusive process gates apply; a channel count other than one or three is refused by name when the process runs	Wrong target is destructively replaced; a sequence's frames are each stretched from their own statistics	Preview/Apply source and active mask	Follows attachment/selection lifecycle; Reset does not change it
Preview	Checkbox	Checked	None	Checked/unchecked	Starts/removes asynchronous full-image preview and switches histogram between preview/source	Process must be idle to change it	Full-image float conversion, statistics, transform, mask blend, and histogram consume time/memory	Preview only; mask-aware; current still/current displayed sequence frame	Window lifetime; Reset does not change it
Preview pane	Full-image viewer	Current transformed target	None	Standard viewer zoom/pan/display controls	Shows the mask-aware full-image transform	Exists only while Preview is checked	Holds an additional float/result image; frequent parameter changes schedule new work	Preview only; target remains unchanged	Window lifetime
Switch image	Preview-toolbar action	Transformed preview shown	None	Click; Space; right-click; Ctrl+Shift+O	Toggles the preview viewer between the current transformed result and original target	Exists only while Preview is checked	Display comparison only; does not recompute or change pixels	Preview display only; still/current displayed sequence frame	Preview-window lifetime
Preview area	Preview-toolbar checkbox	Unchecked	None	Checked/unchecked; move/resize the selection overlay	Shows the whole target's stretch cropped to the area, exactly what Apply writes there; the area's own statistics are never used	Exists only while Preview is checked	Stretches the whole target for each change, its statistics measured once and kept	Preview and Take snapshot only	Preview-window lifetime; Reset does not change it
Take snapshot	Preview-toolbar action	No snapshot	None	Click or Ctrl+Shift+K	With Preview area clear, re-executes the current settings at source depth and opens a new still named <target> - Statistical Stretch - SNAPSHOT	Exists only while Preview is checked	Allocates and processes an auxiliary image; active mask is honored; with Preview area selected, it is processed as the preview is	New still image; target is unchanged	Each action creates a separate image view
Apply	Button/action	Enabled after initialization	None	One action	Captures all parameter values and destructively processes the target	Target must lock; controls disable during work	Full-image and parallel; a sequence's frames are each stretched from their own statistics	Stills, sequences, and the active mask	A successful commit records history and resets stretch settings
Reset	Button/action	Enabled while idle	None	One action	Restores 0.25, 5.0, 0.20; clears all four checkboxes; recomputes statistics/readout	Does not change target, Preview, or histogram toolbar state/zoom	No target image change	Configuration only	Immediate
Close	Button/action	Enabled while idle	None	One action	Closes the process without applying current settings	Disabled with other controls during Apply; shared close handling applies to active workers	No output by itself	Window and active execution	No settings persist after destruction
Sequence progress / Cancel	Shared modal progress dialog and Cancel action	Created automatically when sequence Apply starts	None	One progress update per frame; cooperative Cancel	Processes into a temporary sequence, removes it on cancellation, and replaces the target only after success	Sequence target only	Protects the original sequence from partial replacement	Sequence framework; every frame stretched from its own statistics	Cancel discards the temporary sequence; success replaces the sequence, records one history step, and resets settings

How the algorithm works

Statistics and black point

The process converts the whole target to normalized single-precision float (double for a Float64 or 32-bit integer image) and measures it once for each change of the image, keeping the measurement for every preview and run. For mono and unlinked RGB it computes one median per channel. For linked RGB it computes one median across the pooled RGB samples.

The minimum x_min and the median M are exact: every sample the image's validity admits takes part, and the median of an even count is the lower middle value plus half the distance to the upper one. The lower-half spread is estimated from a deterministic sample of at most 400,000 values per stream. It keeps sampled values at or below the sample's own median (which is M while the sample holds every value), finds their median L, and estimates

\[ \sigma_{\mathrm{lower}}=1.4826\, \operatorname{median}_{x\le M}\left(|x-L|\right). \]

With Black Point Sigma k, the black point is

\[ B=\min\!\left(0.99, \max\!\left(x_{\min},M-k\sigma_{\mathrm{lower}}\right)\right). \]

No Black Clipping instead sets B=x_min. The control therefore prevents the threshold from rising above the minimum; it does not make the minimum pass through unchanged.

Rational target-median transfer

For a source sample x, define the black-rescaled sample and median

\[ z=\frac{x-B}{\max(1-B,10^{-12})},\qquad m=\frac{M-B}{\max(1-B,10^{-12})}, \]

and let t be Target Median. The engine evaluates

\[ f(x)= \frac{(m-1)t z} {m(t+z-1)-tz}, \]

replacing a denominator whose absolute magnitude is below 10^-12 with positive 10^-12. After optional post-stages, it clamps the image to [0,1]. When the denominator is well behaved, f(B)=0, f(M)=t, and f(1)=1. This establishes anchors in one pass; there is no convergence loop or tolerance. For an even-length sample set, transforming the two middle samples separately also need not make their average exactly t.

z is held to [0,1] before the transfer, so every sample at or below B becomes black - the population the clipping readout counts - and nothing above white exceeds it.

Linked and unlinked RGB

• Linked Stretch uses one pooled M, B, and f for all RGB samples, then applies the same nonlinear scalar function independently to R, G, and B. It is not luminance scaling and does not preserve channel ratios.
• Unlinked RGB derives M_c, B_c, and f_c separately for each channel. In the pre-post-stage transform, each channel median is directed toward the same target, which can reduce an existing background/color imbalance as well as stretch it.

Optional curve and normalization

With Apply Curves Adjustment, AfterNight builds six x anchors

\[ 0,\quad 0.5t,\quad t,\quad t+0.25(1-t),\quad t+0.75(1-t),\quad 1. \]

The first three and last y anchors equal x. For boost q, the other two are

\[ y_3=x_3^{1-q},\qquad y_4=x_4^{(1-q)^2}. \]

Linear interpolation between the anchors leaves the range through t unchanged and lifts the upper range when q>0. At q=0 it is identity; at q=1 both adjustable anchors are 1.

Normalize Image then finds one maximum over every processed sample (pooled across RGB) and divides all values by it when it is positive. Final clamping and conversion restore the source pixel type, rounding to the nearest code. The output is marked nonlinear with operation provenance statistical-stretch; RGB becomes display RGB and mono retains its monochrome basis. An active mask is blended afterward.

Parameter interactions and starting points

Combination	Conservative starting point	Too weak looks like	Too strong or defective looks like	Adjustment order
Target Median + Black Point Sigma	0.25 and 5.0	Background/faint structure remains too low	Washed background or crushed shadows	Establish a safe black point first, then set median
Linked + RGB balance	Compare unchecked and checked on calibrated RGB	Channels remain separated more than intended	Unlinked neutralizes a real color signal; linked nonlinear mapping shifts saturation	Confirm color calibration, compare histograms, then choose estimator mode
No Black Clipping + faint signal	Select it when a low sigma clips faint signal	Background pedestal remains more visible	Minimum still maps to black; it does not preserve every faint sample	Keep the faint signal first, then adjust Target Median
Curves Adjustment + Curves Boost	Leave off; if needed start near 0.05	Upper midtones remain flat	Highlight plateau/steep transition near adjustable anchors	Finish rational stretch, enable curve, raise boost slowly
Normalize + target accuracy	Leave off unless maximum scaling is required	Brightest sample remains below 1	Median and noise rise because the entire image is divided by a sub-unity maximum	Set stretch/curve first, inspect maximum, normalize last
Mask + global statistics	Derive settings on the whole source, then judge mask-aware preview	Protected region changes too little	Transition is visible or unmasked statistics poorly represent the applied region	Inspect mask, reduce transform strength, fix mask upstream

Inputs, outputs, and metadata

• One target produces one modified image or sequence. Dimensions, channel count, pixel format, and pixel type remain unchanged. The transform works in float and rounds back to the nearest code for integer sources.
• The target is replaced only after successful still processing. Take snapshot instead creates an auxiliary full-size still at source depth; no background model, clipping mask, statistics file, preset, or file is created.
• The source's header cards, acquisition record, and astrometric solution are kept. Processing state is updated to nonlinear with provenance statistical-stretch; RGB is display RGB and mono remains monochrome. The image states no transfer function, and its processing history records the stretch. Existing color-calibration provenance remains in the metadata.
• Image History records Statistical Stretch, target median, black-point sigma, No Black Clipping, Normalize, and whether curves were applied. It omits Linked Stretch and the Curves Boost value, so history cannot reproduce every execution setting.
• Mask blending is source × (1-mask) + transformed × mask, reversed when the mask is inverted. Statistics and the clipping-threshold readout are calculated from the unmasked source, not only the applied region.
• Each sequence frame is stretched from its own statistics. The shared framework creates a progress dialog, supports cooperative cancellation, writes a temporary sequence, and replaces the target only after success.

Performance

The whole target is measured once for each change of the image and kept, so moving a control or the preview area does not measure it again. The minimum and median count every sample; the spread reads at most 400,000 values per stream. The transform, optional curve, normalization, clamp, mask blend, and histogram traverse image-sized buffers, and a Preview area stretches the whole target before cropping it. Pixel loops and the maximum reduction run on the engine's threads. Linked RGB computes one statistics stream; unlinked computes three. Preview holds float/result buffers and refreshes its histogram after completed generations.

Prototype on a duplicate crop or disable Preview while making several rapid changes on a large image. Increasing histogram resolution changes only the diagnostic graph's memory/render cost, not transform precision; the histogram places a whole code on a bin boundary in its own bin, by exact interval arithmetic. No approved astronomical target was timed.

Worked example

• Evidence mode: Source-, focused-test-, and calculator-derived; not executed on project-owned astronomical data.
• Input condition and settings: A normalized mono row [0.02, 0.08, 0.15, 0.22, 0.30, 0.42, 0.62, 0.85], with Target Median 0.25, Black Point Sigma 0.5, and both optional post-stages off.
• Estimator checkpoint: M=0.26. The lower-half values are [0.02,0.08,0.15,0.22], their median is 0.115, the median absolute deviation is 0.065, and sigma_lower=0.096369. The resulting black point is B=0.2118155; the readout therefore counts the first three samples as <= B.
• Transfer checkpoint: The formula maps the conceptual median anchor to 0.25. It maps input 0.22 to about 0.05098; inputs 0.02, 0.08, and 0.15 lie at or below B and become black, the three samples the readout counts.
• Inspect in real use: If the threshold is above the minimum, compare the threshold readout with the darkest background at 100%. Prefer No Black Clipping or a higher sigma when faint signal is clipped.
• Limitations: No real-image stretch quality, median accuracy after masking, color behavior, or production timing was measured.

Warnings and limitations

Low-sigma black points clip faint signal

Every sample at or below the computed black point becomes black. Use No Black Clipping or raise sigma when the readout counts more pixels than the background should lose.

• Target Median is an internal anchor, not a guarantee about the final measured median after curves, normalization, masks, clamp, quantization, or an even-sample median definition.
• Linked RGB shares a function but does not use common luminance scaling. Any nonlinear per-sample transfer can change RGB ratios and saturation.
• Unlinked RGB pushes each channel's statistics toward a common target and can remove or create color balance changes. Use only after color calibration.
• No Black Clipping still maps the minimum to black and the readout counts it as clipped. It means “do not raise black above the minimum,” not “preserve every low sample.”
• Normalize Image uses one global maximum and changes the median target. One hot/saturated sample can determine the scale.
• The robust-noise estimator is deterministic but sampled on large images; it is not a full-resolution per-pixel noise model and is not mask-aware.
• A Preview area stretches the whole target before cropping it, so it costs a whole-image stretch per change on a large image.
• A non-finite sample in a pixel the image's validity admits makes the engine refuse the run, and nothing is published; pixels the validity rule rejects are left out of the statistics and left as they were. Supply finite normalized input.
• History omits Linked Stretch and Curves Boost, so it cannot reproduce the full settings.

Troubleshooting

Symptom or exact error	Verified cause/validation path	Recovery	Evidence source
Error: "a statistical stretch needs a mono or RGB source"	The image has other than one or three channels	Extract or combine to a mono or RGB image first	stretchImageStatistically input validation, reached from StatisticalStretch::executeTypedProcessImpl
Error: "no pixel participates in this statistical-stretch analysis"	Every sample of the image is marked as holding no data, so there is nothing to measure	Check the image's validity, for example a registration border covering the frame	analyzeImageStatisticalStretch
Darker background than the readout suggests	Every sample at or below a raised black point becomes black	Select No Black Clipping or raise Black Point Sigma	stretchImageStatistically; computeStatisticalStretchClipStatistics
Final median differs from Target Median	Curves, normalization, mask blending, quantization/clamp, or even-sample statistics change the final distribution	Disable post-stages, compare unmasked float output, and regard the control as the rational anchor	Statistical transfer and post-stage order
RGB color balance changes in Linked Stretch	One nonlinear function is applied independently to R/G/B values rather than as a common scale	Compare unlinked carefully or use a color-preserving stretch model; inspect star colors	analyzeImageStatisticalStretch (linked); stretchImageStatistically
RGB background becomes unexpectedly neutral/shifted with Linked off	Each channel receives its own median/black-point transfer toward the same target	Re-enable Linked Stretch or recalibrate/reassess channel intent	analyzeImageStatisticalStretch (unlinked); stretchImageStatistically
Curves Boost has no effect	Apply Curves Adjustment is clear, or boost is 0	Select the checkbox and increase boost cautiously	UI dependency; the highlight curve in stretchImageStatistically
Normalize changes brightness more than expected	Global post-curve maximum was below 1, so every sample was divided by it	Clear Normalize or judge it only after final curve settings	normalization in stretchImageStatistically
Histogram clipping and text clipping disagree	Histogram counts samples at or below 0 and at or above 1 in the source or mask-aware preview, over every channel; text counts source pixels at/below statistical thresholds	Use each diagnostic for its stated purpose and confirm output statistics after Apply	Histogram widget; statistical clip readout
History cannot reproduce the result	Step description omits Linked Stretch and Curves Boost	Record those values externally or preserve a duplicate result	StatisticalStretch::executeTypedProcessImpl
Apply resets the controls	Shared successful-commit path calls Statistical Stretch Reset	This is expected; re-enter settings for another still	resetSettingsAsync; StatisticalStretch::resetSettings

Related workflow steps

Complete Background Neutralization and Photometric Color Calibration or Spectrophotometric Color Calibration before nonlinear work. Compare Statistical Stretch with Histogram Transformation, Generalized Hyperbolic Transformation, and ArcSinh Stretch. Continue with Curves Transformation and deliberate nonlinear finishing while monitoring Image History and Image Statistics.

References and provenance

• AfterNight project source — authoritative for the custom estimator, rational transfer, curve anchors, normalization, UI, masks, state, history, the whole-image statistics rule, and the sequence and below-black-point behavior.