The changelog

7 OCTOBER 2026

StackingWizard Beta 4.7.1 - Moonrise

• Moonrise is not cloud. I ran Mike A.'s three panels of IC 1396 through the Looking Glass on my MacBook and it set 88 of the 114 frames aside. The night was clear; the Moon was rising. The noise check compared each frame's noise with the night's quietest frames, and a brighter sky carries more noise of its own, so the later passes read as haze though their stars were as sharp as the first. The Looking Glass now reads a frame's grain against the brightness of its own sky, which is what moonlight changes and cloud does not, and a frame whose only mark against it is a brighter sky stays in the stack unless the sky has grown to three times the night's darkest (four times at Lenient; Ruthless keeps its old line). The weighted vote already gives a brighter frame less say. On Mike's strip that is 114 of 114 good, and measured against my own hand-grading of two of Volker H.'s nights it catches much the same cloudy frames as before. A frame that is plainly clouded is still caught by the other witnesses.
• The same night, stacked twice, is the same stack. A rule of mine for every stacking change is that a night stacked twice must give the same stack, and this round it caught something. My own drizzled Soul Nebula night, stacked in the morning and again in the evening at the same settings, differed in 22 million pixel values. The vote works in tiles whose size follows the memory the machine has free, and the fine grid the frames' warps are read on was anchored to each tile's corner, so a different tile size moved every sample by a fraction of a pixel. The grid now belongs to the canvas. A drizzled night stacks to the very same picture at any tile size, and a plain one to within a rounding step. The same fault most likely made the straight-edged blocks Karel M. saw once in M31's dust lanes and could not make happen again; I could not test his night, so tell me if they come back.
• A clean bias was read as 8-bit. Gary P.'s DWARF mini night had its master flat made without its bias, with a note that the calibration files did not match. The wizard guessed a file's bit depth from its brightest pixel, and a clean 16-bit bias whose every pixel sits under 260 counts was taken for an 8-bit file and read 257 times too bright. The scale now comes from the file's own type. My own Dwarf 3 bias, which read as nine tenths of full scale, reads as the near-black it is, and the night stacks with its bias in. Gary, yours should too.
• The Handoff sends the mosaic. After a mosaic of named panels, the Handoff sent AstroWizard the panel on screen, not the assembled mosaic, which I found on my own MacBook. It now sends the mosaic when one was assembled, a panel goes by Save, and the note under the buttons says which is going.
• Every page fits a laptop. On a 14-inch MacBook the Pile page ran off the bottom of the window and nothing scrolled. A page that wants more height than the window has now scrolls, and the Pile says the same in fewer lines, about a third shorter. James S. and Hubert asked for scrolling pages back in September.
• A 'failed' frame is judged like any other. A Seestar's or a Dwarf's frame named 'failed' was set aside on its name alone (Mitja). It is now measured like every other frame, and the note counts them. On a Dwarf mosaic of mine two such frames turned out sound and are in the stack. Such a frame is still never the night's reference.
• The note after a curved fit. When a frame's stars needed a curved fit, the note blamed 'the far side of a meridian flip' when there had been no flip (Martin R., an FLT91 with a flattener). It now says what it found, a field that shifted or turned across optics with a little distortion, with a drift, a dither or a flip among the causes, and that there is nothing to fix.
• On a Mac, the Dock brings the window back. A minimised StackingWizard did not come back when its Dock icon was clicked. It should now - tell me if it does not.

Thank you Mike A., Karel M., Gary P., Mitja, Martin R., James S. and Hubert - and the wall now carries 1341 wizards. wizard@lukomatico.com

7 OCTOBER 2026

AstroWizard: Clear Eyes

• Sharper on screen. The picture on the canvas is the real one shrunk to fit, and the shrinking used a quick filter that softened every star, which is why the saved picture always looked a touch crisper than the preview. The draw that stays on screen now uses a finer filter, on the canvas and in the Stretch Wizard's panes; drags and zooms still use the quick one, so nothing feels slower. On my test picture a star's peak stands a sixth higher against the sky, and the fit-to-window draw takes no longer than before. On my MacBook's Retina screen the finished picture in Preview still looks a little sharper than the canvas; that is a different cause and still on my list.
• Star-True Colours on a stack with burnt-out stars. A star that is clipped in a stack has a flat or a noisy top, and the star finder took dozens of 'brightest stars' a pixel apart from one such top, so the matching gave up and Star-True Colours declined a picture full of stars. Every peak on one burnt-out top is now one star. Mike A.'s third IC 1396 panel, which the last build declined, matches 1,152 stars; my own mosaic of his strip was declined until BlurX had tightened the stars, and it should now match before BlurX too, which I am checking as this goes out. When it does decline, a line now suggests sharpening on step 5 first and coming back, which is what got my mosaic through.
• An older drizzled stack. Gary P.'s M74 and Brent F.'s NGC 281, both drizzled, were refused by Star-True Colours. A stack drizzled by StackingWizard before the last build carries its frames' pixel size unchanged, twice the truth after a 2x drizzle, so the search for Gaia's stars was sixteen times too wide. The last build writes the drizzle into the header; for the stacks made before it, AstroWizard now reads 'drizzle2x' from the file's own name. Proven on Gary's own header; tell me if yours still refuses.
• No moat round bright stars, and a Custom mode for the Native High-Pass. Xavier H.'s 'funny stars' after the Native High-Pass sharpening were a dark ring round every bright star, dug by the sharpening's dark side. That side is now held to the sky's own grain, so faint detail still sharpens both ways and a bright star cannot dig a moat. And a Custom button beside the three strengths lets you set the amount and the radius yourself, with the guard on or off.
• A second press while an engine runs. A double-click on BlurX while it was already working threw away the first run's half-written result and said 'execution failed' (my own MacBook again). A second press on BlurX, NoiseX, StarX, StarNet or GraXpert now does nothing, and says the engine is still working.
• Keep this look. 'Keep this look' in the Stretch Wizard applied the stretch, but the step's banner and 'Done here - Next Step' stayed dark until you came back to the page. They light at once.
• On a Mac, the Dock brings the window back. A minimised AstroWizard did not come back when its Dock icon was clicked. It should now - tell me if it does not.

Thank you Mike A., Gary P., Brent F. and Xavier H. - and the wall now carries 1341 wizards. wizard@lukomatico.com

5 OCTOBER 2026

StackingWizard Beta 4.7 - Vanishing Act

• Bright plane and satellite trails, found and left out. Phil's Skies showed me a frame graded good with a plane's lights straight across M81 and M82, and asked whether the wizard looks for trails. It does now. Every frame is read for the bright, straight trail of a plane or a satellite. In the Looking Glass such a frame carries a gold stroke after its name and the trail is drawn on the bird's eye view; the frame stays in the stack, and only the trail's own pixels sit out of it. A bright star's spikes and a galaxy seen edge-on are safe: they sit at the same place in frame after frame, and a trail does not. I read 2,857 of my own frames to prove it: fourteen carry a satellite's trail, all fourteen were found, and no other frame was called one. On a real night of mine the faint mark two trails left in the stack went from about a fifth and a third of the stack's noise to under a twentieth. It is on by default, and the lever is on the Workbench: 'Plane and satellite trails'. It will not catch every trail: a faint one can slip past, and those are left to the stack's own rejection as before. I have no real plane on my shelf, only satellites, so if a plane's trail still shows in your stack, tell me.
• Big walks: the frames that 'would not settle'. Andrew H. had 46 of 2,193 frames laid, and Derge lost 3,576 of 9,217, each with the words 'its stars would not settle on the canvas'. The cause, measured: a short exposure's list of bright points is mostly noise, every unmatched point was added to the walk's star catalogue, and after a few hundred frames the real stars were drowned. Pairing now asks the stars that two or more frames have seen first, the catalogue is cleaned as it grows, and a frame that will not settle is held and tried again at the end. On five nights of Volker H.'s (518 frames) every frame is laid, and a rebuild of Andrew's symptom that dropped nine good frames now lays all nine. I could not test Andrew's, Derge's or Philipp K.'s own nights, so this should cure it - and if frames are still left out, the receipts now end with each one's reason in plain words and what to try. Send me those.
• A mosaic where the telescope moves on after every frame or two. I shot one myself, two frames at each place, and the Looking Glass set 73 of my 76 frames aside. Its checks for cloud compare a frame with the others at the same place, and with two frames a place there is nothing to compare with, so every frame was measured against other star fields. On a walk like that the Looking Glass now sets a frame aside only for its own stars, soft or trailed, and leaves cloud to the stack: on my mosaic that is 19 frames instead of 73. Where the frames it doubts are the only ones at their place they stay ticked, 9 of my 19, because trailed stars are better than a hole in the picture; the note says so and you can untick them. Stacking it showed one more thing: a good frame at the walk's edge was left out, because a frame held for a second try borrowed its rotation from whichever frame was laid last, five degrees off on a mosaic shot near the zenith. It now asks its nearest neighbour on the sky, and all 66 of my ticked frames are laid. Every other kind of night is graded exactly as before.
• A doubted frame is better than a hole. Tom's Andromeda mosaic came out with a block missing over the middle of the galaxy. The Looking Glass had taken the galaxy's own glow for cloud and set aside every frame at that place, so the stack had nothing to lay there. On any walk, a frame the Looking Glass doubts now stays ticked when no good frame through the same filter covers its place; the note says how many there are and you can untick them. On a walk of Volker H.'s, six doubted frames were the only ones over a twentieth of the walk's sky, and that sky is now in the picture. Telling a bright galaxy from cloud is the real cure and it is still to do. I could not test Tom's own nights, so this should close the hole; tell me if it does not.
• The Handoff opens your newest AstroWizard. After an update Christine J. found the Handoff still opening an older copy of AstroWizard, the one that asks for the update. On Windows the Handoff looked for AstroWizard in the usual Desktop, Documents and Downloads folders, and when OneDrive keeps your Desktop that is the wrong place: an older copy left in Downloads was taken. It now asks Windows where those folders really are, and follows a shortcut on your Desktop to its program. On every system it also looks in the folder StackingWizard itself sits in, so two wizards kept together always find each other. And if you once pointed the Handoff at AstroWizard by hand, a newer download now takes over from that older pick; pick again with 'use a different one...' and your choice holds. I could not test on Christine's own computer, so tell me if the Handoff still opens the wrong one.
• A box on the sketch before the vote. Volker H.'s ask: on a big walk the vote is the long part, and it spends much of it on corners you will crop away. A new lever on the Workbench, 'A box on the sketch before the vote': set it to Yes and the stack stops when every frame is down, shows you the sketch and waits for you to drag a box; the vote then works on that box alone. A stack never stops to ask unless you asked it to: the lever is for the next stack only, goes back to No by itself and is never remembered. With several filters you are asked once, and the box is found on the other canvases by their stars.
• Two filters on a colour camera are two stacks. A colour camera's filter name was ignored, so a folder holding nights through two different filters became one stack with both filters' light mixed. Each filter is now stacked on its own, with its own flats, and AstroWizard is told when the two stacks are a pair.
• How much of each exposure went in. A stack built from more than one exposure length says so: '53/53 frames (20x60s, 33x30s)' in the receipts, in the saved file and in AstroWizard when it is loaded (Irwin).
• NINA's compressed FITS. Frames saved as .fits.fz are found and stack to the very same picture as the plain files (Joe K.). On the way a file's header came to be read without reading the whole frame, fifteen times quicker on a Seestar frame, so every big night sorts faster.
• Filed in the right pile. Unistellar's 'StackInput' frames are lights, not finished stacks (Christos K.), and a tile pointed at a folder no longer takes the wizard's own finished stack for one of the night's frames (Alex O.).
• A drizzled stack knows its own pixels. A stack drizzled 2x carried its frames' pixel scale in its header, twice too coarse, and AstroWizard's plate solve suffered for it (Armin J.). The header now names the stack's own scale. On my own drizzled stack the solve went from 365 matched stars to 1,297; I could not test Armin's own stack.
• Two for the Mac. The Handoff searched your folders for AstroWizard while the window waited, which I believe was Alan M.'s beach ball: the search now runs in the background. And your frames are read without memory-mapping, the likely cause of Mike M.'s crash. Neither is proven on a Mac yet, so tell me if they are still there.

Thank you Phil's Skies, Andrew H., Derge, Philipp K., Volker H., Tom, Christine J., Irwin, Joe K., Christos K., Alex O., Armin J., Alan M. and Mike M. - and the wall now carries 1305 wizards. wizard@lukomatico.com

5 OCTOBER 2026

AstroWizard: Signature Move

• Your signature on your picture. Your own words, and a picture of your own if you like: a logo with a see-through background keeps it. Choose its font from six styles or use a font file of your own, put it left, centre or right at the top or the bottom, and set how far it stands from the edge and how strongly it shows, from solid down to a faint watermark. Set it up once from step 9 or from the Save all window; from then on the finished picture is signed when it is saved as a JPEG, a PNG or a TIFF, and so is its labelled copy. A tick beside the save switches it off for a picture you want bare. A FITS is never signed, nor are the starless and stars-only layers. When the stack's own file says how much light went into it, that can stand as a line beneath your name: '429 x 10 s, 1 h 12 min'.
• The Scrying Orb as a video. Instagram refuses a GIF (Stardust and magus424). The Orb's window now offers a GIF, an MP4 video or both: the same steps and the same holds, 30 frames a second, and your signature on it by a tick. AstroWizard writes the video itself, so nothing was added to the download for it. The video is larger than you might expect, about 1.3 MB a step at 1080 wide, because each step is stored exactly as it is.
• How much light, said at the door. Load a stack made by StackingWizard and AstroWizard says what went into it: 'This stack: 429 frames × 10 s = 1 h 12 min of light · gain 80' (SaltyPogue's ask).
• North, and the mirror. After a plate solve AstroWizard says which way north lies against the top of your picture and whether the picture is mirrored (Gregg S.). Proving it turned up a fault nobody had reported: on a mirrored picture the names in 'What's in my picture' and the saved plate solution landed in the wrong place, 2,572 pixels from the star on my test. They now land within a pixel.
• A Bin 2x2 after sharpening, for a drizzled stack. A stack drizzled 2x is four times the size, and once BlurX has done its work the extra pixels have given what they had. The sharpening step now offers a Bin 2x2 card for a drizzled stack (Richard D.); other stacks see nothing new.
• AstroWizard knows its own finished picture. A picture AstroWizard had already finished, loaded again, was shown through the wizard's stretch a second time, brighter than it was saved (Hitesh M.). A finished FITS, a finished TIFF and the wizard's own working files are now each shown as they were saved.
• Leaving the Stretch Wizard. Leaving the Stretch Wizard without a pick could leave a grey picture on the canvas (skaiii). The page is no longer redrawn underneath the wizard's first pick, which is what did it.
• The Stretch Wizard's first page. The two colour panes, Linked and Unlinked, were drawn at one fixed strength, which left them washed out on some pictures and too dark on others: on five test pictures the sky sat anywhere from a seventh of full brightness to four fifths. They now work the strength out for your picture, so the sky sits at the same brightness every time, before and after noise reduction, and the colours can be judged. The second page and the finished stretch are exactly as they were.
• Star-True Colours on narrowband data. It used to refuse a dual-band or an SHO/HOO picture outright. It now says what it would do to such a picture and lets you run it anyway; Undo brings the picture straight back.
• Gaia's stars: wide fields, and the nights the archive is down. On a wide mosaic of the Milky Way, 'What's in my picture' and Star-True Colours could wait a long while and then say Gaia could not be reached. The catalogue's main archive cannot answer such a field in time, and the fallback behind it had a fault that meant it had never worked. Both are put right: a wide field asks for its bright stars only, the fallback answers, and my own Seestar mosaic of five degrees is placed on the sky in half a minute. The same fallback now covers the nights when the main archive does not answer at all. It was down for most of the night I tested this build: after a twenty second wait for it, five different fields each had their stars from the fallback within a few seconds, and Star-True Colours ran on my stack. The build before this one had no stars at all whenever the main archive did not answer. While it works, the line on screen counts its seconds and says what it is waiting for, and if the archive really is slow AstroWizard says that, and no longer blames your internet.
• A picture that was never stretched is not saved dark without a word. Gordon B., new to all this, cropped a fresh stack and saved it. AstroWizard's screen showed the picture; the JPEG and the TIFF were almost black. A fresh stack is linear, what you see is the wizard's screen stretch, and the save wrote the data as it was and said nothing. Before a never-stretched picture is saved as a JPEG, a PNG or a TIFF, AstroWizard now stops and says so, with one button that takes you to the Big Stretch and one that saves it as it is. The question at Final Polish is asked whichever step you come from, in plainer words. A FITS is saved as before, with no question.
• The manual dial's pick, on every pass. 'Pick SP from the picture' worked once: after the first Apply a click on the picture did nothing (Martin K.). The page behind the dial is built again at every Apply and the pick was tied to the first one. It now goes to whichever picture is on show, pass after pass.
• Small tools, small files. StarNet2 refuses a picture under 512 pixels on a side; AstroWizard no longer sends it one, and says why (Anne W.). The dual-band pair's S, H and O files are kept after Combine & Load, and carry their stack's place on the sky (Chuck W.). Two colour stacks through two different filters arrive from StackingWizard as a pair. And a drizzled stack's plate solve finds its stars (Armin J.).
• One for the Mac. Your files are read without memory-mapping, the likely cause of Mike M.'s crash. It is not proven on a Mac yet, so tell me if it is still there.

Thank you Stardust, magus424, SaltyPogue, Gregg S., Richard D., Hitesh M., skaiii, Gordon B., Martin K., Anne W., Chuck W., Armin J. and Mike M. - and the wall now carries 1305 wizards. wizard@lukomatico.com

30 SEPTEMBER 2026

StackingWizard Beta 4.6 - The Magical Mender

• Hot pixels, mended at the root. On a night without darks the sensor's hot pixels ride into every frame. On my own Soul Nebula night (three narrowband filters, no darks, a dither every few frames) they left the stack peppered with single pixels far darker than the sky, and those went on to spoil the noise reduction in AstroWizard. The wizard now reads the hot pixels from the night itself: a pixel is called hot only when it stands out in every frame the wizard looks at while the sky has moved underneath it. Each one is repaired in every frame before the frame is laid. It is on by default, it stands aside by itself when you bring darks (a dark does the job properly), and the lever is on the Workbench: 'Hot pixels on a night without darks'. It needs the sky to have moved a little across the sensor during the night, by a dither or a drift; without that a hot pixel cannot be told from a star, and the wizard leaves well alone. Proven on my own night and on a Dwarf mosaic - smart telescope nights should gain from it too, so tell me what you see.
• The gradient stays in. On walks, mosaics and drizzled nights the wizard used to take the canvas's large gradient off in the stack. On my Soul Nebula through a 3 nm filter there was next to no gradient to take, and what went was a good part of the nebula's own glow: a curve fitted to the whole canvas cannot tell a large nebula from a gradient. From this build the gradient is left in the data for AstroWizard's Background step, which knows where the nebula is. The Workbench's lever 'The canvas's large gradient' takes it off as before. The lever moves no star: measured on two nights, both ways, the stars sit where they sat.
• Frames and masters that do not match. Alex O.'s two nights arrived at two bit-scales with a bias and flats that fit only one, so half his frames came out of calibration empty, the anchor frame had no stars and nothing stacked. Each frame now takes its calibration at its own scale, the masters are tried on a frame before they are trusted with it, and the anchor must prove itself before the night is built on it: another frame has to align to it. I rebuilt his night in small and got his receipts back line for line; this build stacks it whole.
• Mine to own. Since Beta 4.5.1, when the anchor of a plain night was not the night's reference, the reference was left out of the stack and the anchor was counted twice. The anchor now leads its own cohort and every frame is counted once.
• Camera RAW, further. A camera's RAW frames can be drizzled, walked and laid on one canvas (Shamik G.'s Nikon night stopped at 'stack failed'). A RAW in a folder called Darks, Flats or Bias is filed by its folder, as a FITS file is, and a burst of short flats is no longer taken for copies of one frame: the clock in the file counts whole seconds, so the file's own content now decides (Ninbe's Nikon Z7). A first swing at all three - if yours still lands wrong, the receipts will tell me why.
• Nothing read twice. Pointing a tile at a folder sorted the whole night again from nothing, and before the first frame of a big night the wizard read every header a second time without a word (Jeff P.; Neal G.'s 2,762 frames). Each file is now read once, each master is brewed once, and the status line says what the wizard is doing.
• The Pile. A frame brought in through a tile is read like one brought in by its folder: the 'Colours wrong, or should this be colour?' button is there, and a mono night keeps its filters. And four seconds reads as four seconds, not three (Christos K.).
• Linux. Links open through your system's own opener, the wheel scrolls the list and zooms the stack, the working files never land in a /tmp that lives in memory, and no worker thread calls into the window toolkit any more, the classic cause of a stall or a crash (Rowald N., Lonny S., Jean-Pascal O., Klaas B.). The app uses your system's own X11 and font libraries ahead of the ones it carries. One trade you should know of: when an input method server (ibus, fcitx) is running, the input method is switched off inside the app, so CJK input and dead-key accents do not work in its text boxes for now. In my tests the build started on Ubuntu, Debian, Fedora and Arch and stacked a small night, but it has never run on a real desktop - so tell me what is still wrong.
• Small mercies. The Workbench's button wears gold, so it can be found. A hover tip stays inside its own window (Klaas B.).

Thank you Alex O., Shamik G., Ninbe, Jeff P., Neal G., Christos K., Rowald N., Lonny S., Jean-Pascal O. and Klaas B. - and the wall now carries 1243 wizards. wizard@lukomatico.com

30 SEPTEMBER 2026

AstroWizard: Fine Tooth Comb

• NoiseXTerminator and the single dark pixel. A stack can carry single pixels far darker than the sky around them; a night without darks leaves them beside its hot pixels. They are too small to see, and they spoil noise reduction: on my own Soul Nebula NoiseXTerminator came out weak and patchy, worst right beside the stars. AstroWizard now goes through the picture with a fine tooth comb before NoiseX runs and mends every such pixel from its neighbours, and the console says how many it found. Proven on my own data; StackingWizard Beta 4.6 cures the same thing at the root.
• NoiseX's own Iterations. The presets used to run the whole tool two or three times over. They now hand NoiseXTerminator its own Iterations setting in a single run, as they were always meant to (Russ M.'s catch). It is quicker, and it is lighter than before, most of all on noisy data; Heavy now lands close to Normal. If you liked the old strength, press the preset again: that is what the old one did. The console says what ran, for example 'NoiseXTerminator (Denoise 0.90, Iterations 2)'.
• GraXpert's half-downloaded model. If GraXpert's AI model only half downloaded once, GraXpert failed on every run with 'NO_SUCHFILE ... model.onnx' and never fetched it again (Daryl M. on Windows, Glen W. on a Mac). AstroWizard now clears the broken model so that GraXpert downloads it afresh, and says so rather than reporting a crash. This should cure it - if yours still fails, send me the console's words.
• The combiner. A combined picture keeps its masters' header, so it knows its place on the sky and the steps that need it can find it. And a fourth choice beside the palettes, 'My own mix': any masters, any order, no palette assumed.
• Correct Magenta Stars keeps each star's light. It used to hand stars back brighter than it found them, by about a seventh on a typical star and by up to a half on some, which read as bloat. A star now keeps its own brightness and loses only the magenta.
• Crop, Undo and the star layers. The plate solution follows a crop, and after a turn, a flip or a bin it is dropped rather than kept wrong. The star layers belong to the picture they were made from: Undo and Redo bring them back with it, and a second star removal on a picture that is already starless asks before it replaces them (Doug H.).
• What's in my picture. A name no longer lands inside another object's circle, and a short line joins a name to its circle when the two stand apart (Benedict H.'s own suggestion). When the stars have been taken down, the picture is placed on the sky by an earlier step that still had them. The window fits the app's own window and holds its size.
• Tools. A tool that is not where AstroWizard expects it is looked for through the PATH, and the Locate dialog opens there (Klaas B.). A tool's words are read past a character the Windows code page has no letter for: that stopped BlurXTerminator four times on one picture (Amer A.). This should cure it - if BlurX still 'fails' for you, send me the console's words.
• Linux. StackingWizard's list, and more: the wheel zooms the picture, the popup when going back a step is no longer blank (twelve other popups had the same fault), the labels window fits the app's own window, and buttons that were only a glyph are drawn (Rowald N., Lonny S., Klaas B., Jean-Pascal O.). The same trade on the input method applies: with ibus or fcitx running, CJK input and dead-key accents do not work in the app's text boxes for now.

Thank you Russ M., Daryl M., Glen W., Doug H., Benedict H., Klaas B., Amer A., Rowald N., Lonny S. and Jean-Pascal O. - and the wall now carries 1243 wizards. wizard@lukomatico.com

25 SEPTEMBER 2026

StackingWizard Beta 4.5.1 - The Night's Name

• The night's name. Malcolm picked his 'M31 ANDROMEDA_90s' folder and the Pile called the night 'M 51 WHIRLPOOL GALAXY'; Mitja stacked NGC 7380 and the save dialog offered a name from a master of his last session. Both had one root: a finished stack found among the frames could name the night, and the first target name seen won, whatever file carried it. Now the lights name the night by a vote of their own headers, a finished stack never does, a night whose headers are mute takes its folder's name, and when the folder says M31 and the headers say M 51 the folder wins and the Pile tells you why. The save dialog offers the autosave's own name.
• The anchor. On a plain night the canvas is one anchor frame's frame, and the anchor was chosen by grade; on Stephan's Veil that was the session's first frame, pointed a little off, so the nebula's edge was cut and a black band ran down one side. The anchor is now the frame nearest the middle of the night's pointings when the frames carry sky coordinates, the night's reference when they all say the same, and the receipts say which rule chose it.
• Small mercies. The green update banner opens the StackingWizard page rather than the site's front door (Jeff F.). A save with no folder lands in a real one. 'Support this project' now sits beside the Workbench gear on every page, with the usual doors. The Pile's notes finally admit that camera RAW is in.

Thank you Malcolm B., Mitja S., Stephan L. and Jeff F. - and the wall now carries 1165 wizards. wizard@lukomatico.com

25 SEPTEMBER 2026

AstroWizard: The Landing

• The save lands in your folder. On the newest macOS the final save could fail with 'Read-only file system' (Robin): a Mac app opened from Finder sits at the very top of the disk, the save dialog was given no folder to open in, and the picture was written where it could not be. Every save now opens in the folder your picture came from (else Pictures, else home), and a name that comes back without a folder, or at the root, is landed there. Save Final Image, the Scrying Orb and the labelled copy all follow the rule.

Thank you Robin - and the wall now carries 1165 wizards. wizard@lukomatico.com

24 SEPTEMBER 2026

StackingWizard Beta 4.5 - The Night Chart

• The night chart. On the Looking Glass, once the pile is graded, the night draws itself: sky glow, star size and star count as three lines in the order the frames were shot, with the verdicts as a band beneath. A passing cloud is a dip, a focus drift is a slope, and a frame that failed the grade sits under a mark you can see from across the room. Hover a point and the frame is named with its grade and the reason; click it and the Magnifier opens it. The grades CSV now carries the grade and the reason beside every frame, and the receipts are saved beside every master as a text file, so a report writes itself.
• One canvas for every filter. A mosaic shot through several filters came out as one master per filter, each on its own canvas of its own size, and AstroWizard could not combine channels of different sizes (Mike A.'s five-filter WR 134 set, then Ambroise A.'s NGC 7000 in SHO). The richest filter's canvas is now the grid, and every other filter is voted straight onto it by its stars, so the masters come out the same size on the same pixels. A second filter's master carries the pointing of the grid it sits on, not its own first frame's.
• Camera RAW. Canon, Nikon, Sony, DNG, Olympus, Panasonic, Pentax and the rest come through the same door as FITS, as lights, darks, flats and bias, read by LibRaw, with a header made from the camera's own facts: the Bayer pattern, the exposure, the ISO and the clock. Proven on a Canon 700D's frames; every other make takes the same road unproven, so if yours does not land right, send one frame. Fujifilm's X-Trans pattern is not read yet.
• Drizzle 2x on the Workbench. Off by default. It rebuilds an ordinary night on a grid twice as fine, dropping each frame's pixels as small squares instead of blending them, for small stars (two or three pixels across) on frames that drifted a little between exposures, as every smart telescope's do. The master comes out four times the size and takes longer to vote, and its thin outer margin is filled from its neighbours. Twenty frames or more; leave it off for a mosaic of finished panels or a night with few frames.
• The vote weighted by quality. An ordinary night's final vote now counts a clear, steady frame for more than a hazy one, as mosaics and the Shelf already did. And the silent stage before the vote is gone: 2,500 frames (Bruce G., Gedaliah W., Neal G.) went from about an hour of nothing on screen to a minute and a half.
• The hot-pixel lock on the Dwarf mini. A sensor with many warm pixels on a still field (John D.'s M16, Maxim S.'s North America) had its hot pixels mistaken for stars; the defect map now tells the two apart. John's and Maxim's own nights are the test - tell me.
• Small mercies. The name labels that stuck to the screen on Windows (Scot B., Steve R.) are one window now, shown and hidden, never orphaned. While a canvas grows, a caption says what the right pane is (Steve R. took the live sketch for the result). When the room check says the drive is short, 'Stack anyway' is your word to go on (Max C.'s Mac showed 79 GB free while the wizard saw 19: a Mac does not count its purgeable space). The unticked frames can be moved aside into a suspect_subs folder on your own press, after a warning, and never deleted (keepsake19938, Sabine S., Oleg, Darran P.). A Mac wears its own Dock icon (Greg R.). A panel trimmed before the laying keeps its bearings (Ambroise A.). The Workbench opens no taller than the screen. One finished stack is counted as one.

Thank you Mike A., Ambroise A., Steve R., Scot B., Max C., Greg R., John D., Maxim S., Bruce G., Gedaliah W., Neal G., Sabine S., Oleg, Darran P. and keepsake19938 - and the wall now carries 1156 wizards. wizard@lukomatico.com

24 SEPTEMBER 2026

AstroWizard: What's in my picture

• Name what's in my picture. On step 9, one click places your picture on the sky by its stars, the same match Star-True Colours makes against Gaia, and names every catalogued galaxy, nebula and cluster inside it on a copy for sharing, drawn on the picture exactly as you see it. A catalogue of 12,163 objects rides inside the app (OpenNGC by Mattia Verga, CC BY-SA 4.0), so nothing is downloaded and nothing is asked of you. A FITS export now carries the plate solution too, so other programs know where your picture points. It needs sky coordinates in the file's header: a smart telescope's stack or a plate-solved capture has them.
• The one online switch tells the whole truth. The switch in Settings now covers the three things AstroWizard ever goes online for: the version check, the Fellowship roll and the Gaia star lookup. With it off, a load fetches nothing; the button on step 4 still asks Gaia for that one field, because that is you asking. Only the sky position and the field's size are ever sent, never your image, and certificate checks are always on. A correction to the 3 September notes: they said every AI engine honours the compute-device choice, and StarNet2 has no such switch.
• The star finder's ladder. Star-True Colours tries several settings in turn, so soft and oversampled stars that fell off the old cliff are found and matched.
• Save all project files. One button on step 9 (Harald B.'s ask) saves the final image, the starless version and the stars-only layer in the formats you tick, into one folder, named from your original file; nothing is ever overwritten. PNG saves are eleven times faster than they were.
• The header on narrow screens. The Fellowship pill, the file's name and the progress bar are seated from the bar's real geometry (Greg R., rsmath1): the pill keeps its full name, slides right to clear the zoom buttons, and on the narrowest windows sits a little over Settings & Help rather than hiding; the file's name is fitted to its seat. The hover tips on Windows are one window now, shown and hidden, so the black and grey boxes some of you saw left on screen are gone. Show Tips is the switch beside Reset View.
• Smaller mercies. Once the stars have set the balance on step 4, Auto Colour Balance steps out of the gold trail on step 8, with a line saying why. A Mac wears its own Dock icon (Greg R.). The Scrying Orb's card says what it does in plain words. Start Over and the mono combiner forget the previous image's replay frames and star data.

Thank you Harald B., Greg R., rsmath1 and Mattia Verga for OpenNGC - and the wall now carries 1156 wizards. wizard@lukomatico.com

18 SEPTEMBER 2026

StackingWizard Beta 4.4 - The Workbench

• The grid, measured properly. Since the walk arrived a hard stretch of some stacks has shown a grid of squares, and this build is the honest account of it. On a real Seestar walk of 120 frames the finished stack has no lattice at all, at the cell size or the tile size; what the eye finds is the cloud verdict's own cells and the thin outer pointings, where two frames cannot look like thirty. So the cloud and haze verdict is judged at cloud scale now: each frame's disagreement with the rest is smoothed over a hundred pixels, judged over plain sky only (never a halo, a bright star's zone or structured nebula), counted only as a patch at least 160 px across, and faded out with a soft edge - a lone cell can never draw a square. A walk's outer pointings drop their single highest and lowest value at each pixel, so no hot pixel or trail rides the thin edges. The twin sample of 4.3 stays for turned panels. By Luke's eye each of these took the pattern down again on the Veil and on an Elephant's Trunk mosaic, and it is not claimed gone: a stack that still shows it is exactly what I want to see.
• The Workbench. A gear beside UI Scale, on every page, opens the levers that measurably do something, and only those: the cloud verdict (sky only at cloud scale, every cell, off), how hard the final vote rejects an outlier, each frame's own sky shape, the canvas's large gradient, vignetting measured from the walk itself, the twin sample on panels, and memory in the final vote. The wizard's choice sits first on every row with one plain sentence, and Help and examples, on by default, says under each lever when a person would move it and when to leave it: a thin cloud the stack still shows, a laptop with 8 GB, a night shot with real flats. The choices last for this session or forever, as you say; one press returns everything to the wizard's defaults; and every choice that is not the wizard's is named in the receipts, so a report tells me what was on.
• Sorted right. Frames converted by Siril carry a stack count of one and no image type, and every one of them was filed as a finished stack (Nils, Guillermo, Hubert): a count card now speaks only above one. A finished stack loaded among the subs by accident, a 4294 second frame among ten-second ones (Christos), moves to the finished stacks pile and is named. A mosaic's autosave carried the anchor frame's pointing, one edge of the canvas; it now names its own centre, so a plate solve or AstroWizard's star lookup covers the whole canvas. A panel of one finished frame (Steve W.'s Dwarf 3 megastacks, one file per panel) is laid as it is instead of standing aside for want of three.
• Small mercies. Play resumes where it paused (Marcel). The name tooltip on the Looking Glass never outlives its page (Steve R.). The Handoff's list scrolls past eight stacks so the Combine button stays in reach (James, Hubert's 369). The grades export to a CSV (Roger). The receipt names the canvas's depth range. The star finder asks a peak to light its neighbours a little more before it is a star, so a warm pixel with a little bleed anchors no registration (Rick).

Thank you Marcel T., Nils S., Guillermo, Hubert M., Steve R., James S., Rick M., Roger K., Christos K., Steve W. and cbskousen - and everyone whose night still shows a square. wizard@lukomatico.com

18 SEPTEMBER 2026

AstroWizard: True Colours

• Star-True Colours. The stars in your picture set the colour balance. When you load a FITS with sky coordinates in its header, AstroWizard fetches the field's stars from ESA's Gaia catalogue in the background, a few hundred of them, while you get on with the early steps. On step 4, one click finds the stars in your pixels, matches them to the catalogue (scale, turn and mirror all free, no plate-solving program needed), measures each in red, green and blue, and fits the camera's response to the stars' true colours; the balance then puts a Sun-like star at white and sets the sky neutral. It is photometric colour calibration of the kind the big packages charge a setup evening for, with nothing to install and no catalogue to download. It declines honestly: on narrowband palettes (a palette is a choice, not a star's colour), on stretched data, without sky coordinates, or with fewer than thirty clean stars. Barry's blueprint from August, built at last. This work has made use of data from the European Space Agency (ESA) mission Gaia, processed by the Gaia Data Processing and Analysis Consortium (DPAC).
• The Scrying Orb. On step 9, your whole edit replayed as an animated GIF, one frame per step from the snapshots the app keeps as you work, each frame named by the step that made it, the first and last held longer, 720 to 1440 pixels wide, ready to post beside the finished picture with a small mark of where it was made, yours to leave off.
• The manual dial. For hands that know what they want: the generalised hyperbolic stretch on step 7 in the shape PixInsight users know. The histogram and the transfer curve share a graph you can zoom, the types are generalised hyperbolic and linear, the stretch factor and the local intensity (negative too) have their dials, the symmetry point is typed, slid, or picked from the picture with one click, shadow and highlight protection and an invert are there. Every Apply stacks on the image as it stands, the GHS way: a strong first pass, then smaller ones at other symmetry points, with the main window previewing each move; Undo steps back and Reset Step returns to the start. Step 7 itself is reshaped around it: the Stretch Wizard alone in its card, Keep this look in its own box, the linked switch out in the open, the dial in its own box marked advanced, and the classic and deep stretches under Legacy options. The dial follows the transformation family published by Dave Payne, whose PixInsight process with Mike Cranfield defined the workflow; AstroWizard's implementation is independent, written from the published mathematics.
• Two dual-band stacks, one Hubble palette. A colour camera's Ha/OIII stack and its SII/OIII stack, on step 1: the second is registered onto the first by its stars, Ha, OIII and SII are split out (the two OIII averaged), and Combine Mono Channels opens ready as SHO. Four of you asked (Tomas, Andy, Joao, Tim).
• Smaller mercies. The Fellowship fetches on your own click when the startup check is off, so the switch stays honest and the hall stays reachable. Show Tips is remembered between sessions. GraXpert falls back to its newest model when the pinned one is missing or not kept (Dmitry, Glen), and a model that will not compile on a Mac retries on the CPU (Chris). The RC-Astro licence button no longer vanishes for a session after a CLI update on Linux (Jerry). A Settings switch hides the free tools where an RC-Astro tool does the job (Matthias); GraXpert's background extraction always stays. The StarNet2 failure note names the executable bit with the exact command (John).

Thank you Barry S., Chris S., Dmitry, Glen W., Jerry R., Matthias, John L., Tomas, Andy, Joao and Tim - and the wall now carries 1037 wizards. wizard@lukomatico.com

12 SEPTEMBER 2026

StackingWizard Beta 4.3 - The Thin Strip

• The thin strip. Four of you wrote in with mosaics whose panels share only a sliver: a 3x2 from an ASIAIR at ten percent overlap, a row of three, a Dwarf 3 2x2 and a Dwarf Mini pair, and each laid a panel or two and left the rest out. The edge correlation asked for a share of coincident stars a ten percent strip can never give, and the sky's map wanted three placed panels spread two ways before it would predict, which a row of three never reaches. Now, when the raw match fails, every shift is judged by the stars that could land on placed canvas there, and two placed panels make the map. A three-panel California Nebula lays three of three in every shape I could give it, a synthetic 3x2 at ten percent six of six. Your own thin nights are the proof, so send the receipts either way.
• The levels of a row. The three panels came out white at both ends: the canvas's trend, a quartic fitted through two narrow seams. The trend is now taken only where the frames' footprints vouch for it, a panel meeting the canvas along one narrow seam keeps its level, and each panel's soft edge is cut before the laying, so no dark seam lines.
• The grid of squares. A hard stretch of a 3x3 showed a lattice of faint squares across the whole mosaic. Six of the nine panels sat two degrees from the canvas, and a turned panel's pixel grid beats against the canvas's every 1/sin of the turn, 27 px here; one cubic sample per canvas pixel carries the interpolation's own grain at that beat. Each pixel is now sampled twice along the panel's diagonal and averaged: the pattern's swing halves on the test mosaic, a star gets 4 percent wider, and walks are untouched. What remains under an extreme stretch is the mosaic's depth itself: the corners gathered one panel's light where the middle gathered nine.
• The star zone. A second-magnitude star on a Seestar walk left an arc of dark cells four hundred pixels out; the cloud verdict's quiet zone now grows with the star's light.
• The colour door, always. An Odyssey Pro night whose header names one pattern and reads right as another showed no colour door; it is on every manifest now, and frames stored behind an empty primary header are read whole.
• Also: StarNet2 no longer crashes on Linux when AstroWizard was opened by the handoff (our bundle's library path rode along), and the receipts describe a mosaic's geometry panel by panel: size, pointing, how each was placed, and the overlap it shares with each neighbour.

Thank you to everyone whose mosaic wrote in. wizard@lukomatico.com

11 SEPTEMBER 2026

StackingWizard Beta 4.2 - The Dwarf's Panels

• The Dwarf's mosaics. Eight of you wrote in a day: a Dwarf 3 or Dwarf Mini mosaic stacked one panel and rejected the rest. The Dwarf writes the same RA and Dec on every panel and names them 'M 31(1)', 'M 31(2)'; the panel grammar read NINA's _PANEL_2 and its cousins only, so the Pile saw four targets and stacked the reference's. The grammar now reads a bracketed number at the end of the object name as the panel. Then the laying: with one pointing for every panel, every guess about where a panel belongs landed on the anchor's footprint, so a panel that only touches a far neighbour was held and left out. The canvas now tries a place beside every panel already placed, nearest first, until the stars agree. A Dwarf-shaped three by three, one pointing for all nine, places nine of nine.
• The mosaic toggle is always there. It used to appear only when the Pile had found something; it now sits under the Pile whenever there are lights, saying what Auto would do. Panels, on your word, takes named panels, a spread of pointings, or one folder per panel - the Dwarf keeps each panel in its own folder - so a mosaic the Pile could not see can still be stacked as one.
• The bilinear demosaic is the default again. VNG came in with Beta 4 on a Seestar measurement, and a night later put a fine diagonal maze over the whole sky of a tracked Hyperstar night shot without dither: eight times the bilinear's noise on a sub, and diamond-shaped stars. VNG and Malvar-He-Cutler stay behind the door for a chosen night; the receipt names whichever runs.
• The screen stretch is the auto-stretch of PixInsight, Siril and ASIFitsViewer. Per channel the shadows sit at the median less 2.8 sigma and the midtones put the median at a quarter of the screen, so every channel's sky lands at the same brightness whatever the file - the sky no longer casts blue on one master and brown on another. One ruler per run still: the growing stack keeps the first frame's window, the finish and the zoom render through it, and the frame pane shows each frame on its own terms. A walk's finish had been rendering red and green a touch darker than the in-flight view; it renders as the in-flight view did.
• More frames in flight on a big camera. The alignment counts its workers in gigabytes, and its guard budgeted eight frame-sizes for each frame being aligned where a measurement of the process's own high-water mark shows two and a half. Four are budgeted now, with a margin: a 16 GB machine with a 26 MP camera aligns four to six frames at a time instead of two or three, and looks less like a stack running on one core.
• Dwarf owners, use Panels. The Dwarf writes one pointing on every panel, so its mosaics stack panel by panel: leave the Pile's toggle on Auto when the panel numbers are in the names, or set it to Panels when each panel sits in its own folder. The live single canvas of the video is the Seestar's walk, which reads the pointing from every frame's header, and the Dwarf's headers don't carry it.

Thank you to everyone whose Dwarf night wrote in. wizard@lukomatico.com

10 SEPTEMBER 2026

StackingWizard Beta 4.1 - The Bright Star

• The dark box on the brightest star. A walk shot at more than one exposure - Volker's night mixes ten-second and sixty-second frames - came out of Beta 4 with a dark rectangle on its brightest stars. Two things stacked up. A short frame's gain against a long anchor was capped at four where the exposures ask for six, so every short frame's stars and halos sat a third low while its sky was matched; and the canvas's eye for cloud then judged the bright star's cell, where the few long frames stood above the many short ones for reasons that are not cloud, and the cell was voted by the short frames alone. The gain now follows the exposure ratio the headers give, within a factor of two, and cloud is judged on sky cells only, and only where a minority of a cell's frames disagree. Verified on the night that showed it.
• Everything else is Beta 4, an evening older: the live canvas, the exact level solve, VNG, the grading that learned from an eye.

Thank you Volker H., whose night found it within the hour. wizard@lukomatico.com

9 SEPTEMBER 2026

StackingWizard Beta 4 - The Live Canvas

The biggest change since the first beta: the mosaic and walk road is rebuilt from the ground up, and every colour stack now takes the new demosaic. A night that stacked fine on 3.2 should stack better here; if one comes out worse, send it - that is what a beta is for.

• One canvas, start to finish. A walk - a Seestar or Dwarf night shot in framing mode - no longer stacks pointing by pointing and joins the pieces at the end. Every frame lands on one growing canvas as the night was shot, placed against the canvas's own stars, and the wizard votes on every pixel at the end at full size, straight from the frames on disk, with no working files. You watch it happen: each frame on the left with its matched stars, its outline in gold where it landed on the right, an honest line saying how deep the canvas is where it is deepest, and a sketch of the canvas saved the moment the frames are all down, before the vote. Volker's Veil, four nights of Seestar frames at four exposures mixed together, comes out as one canvas.
• The sky levels, solved together. Every frame's sky level is now solved at once against every neighbour rather than frame by frame, and the canvas keeps the anchor frame's own brightness. On the way here a clear night showed the canvas striped in columns, whole pointings sitting a few percent above or below their neighbours; the joint solve was the cause and the exact one is the cure, measured on that same night.
• The canvas's own eye for cloud. Where a frame disagrees with everything its neighbours put on the same patch of sky, that patch is cloud for that frame and the vote gives it no weight there; a frame cloudy over two fifths of its area stands aside. It speaks only where three or more frames saw the patch, since two cannot say which of them is wrong, and it can never leave a hole - Mike's 3x3 taught it that on the Garnet Star.
• Panels and the rig's optics. Named or pointing panels go through the same canvas, with the rig's field distortion measured from the stars the panels share and undone before they meet. Mike's 3x3 lands nine of nine with seams as clean as a single panel's own stars.
• VNG demosaic. Colour frames are demosaiced by the variable-number-of-gradients method for the finished pixels: cleaner star cores and less colour fringing than the bilinear. It costs a little time on a plain night, well under a minute on a typical Seestar session, and the star searches and the live view still take the fast road.
• The grading learned from an eye, twice. On a walk, a frame is judged against its own pointing for the things that change across the sky (the star count, the brightest stars, the patchiness) and against the night for the rest. And the trail and softness lines sit a step tighter at every mood: on a blind sample of thirty frames the grading agreed with my own eye on twenty-nine, where the old lines let twelve of my sixteen throws through. A clean Seestar night loses two frames to it.

Thank you Volker H. and Mike A., whose nights carried this whole round - and everyone who sends a night that does not work. wizard@lukomatico.com

7 SEPTEMBER 2026

StackingWizard Beta 3.2 - The Gradient Gamble

• The grading sees the cloud. Two things at once. The grade now judges every frame as it was shot, and calibration is kept for the stack: Robert's night kept 106, 110 and 73 frames of the same data depending on which masters were aboard, because the sky test is a percentage over the best frame and a bias pedestal or the vignetting spread quietly dilutes it. And thin cloud is finally caught: a frame whose star count, or the light of its brightest stars, falls well under the night's median is a suspect (the mood sets how far - lenient, standard, ruthless). On Volker's NGC 6888 night five frames that used to pass are suspects now, on his IC 1318A night fourteen, and every one of them is cloud when you look. Grading also runs one star search per frame instead of two.
• The walk's tiles and colour patches. Volker's Seestar walk came out of 3.1 as a staircase of rectangles at different levels with green and magenta patches, and the PixInsight forum tells the same story about restacked Seestar mosaics: each pointing carries its own vignetting and gradient in, and no sky offset can flatten a bowl. Now, on walks and pointing mosaics only, each pointing's sky is fitted as a gentle surface anchored on its darkest blocks and that surface's shape is subtracted before the pointing lands, and one colour gain per channel per pointing is solved from the stars every overlapping pair shares, all pointings at once. On synthetic walks with an eighteen percent vignette and a drifting gradient the tile spread fell from ninety-four percent to just over one, and with per-night colour casts the star-light disagreement fell from twenty percent to four. A real walk has not been through it yet - that is the gamble in the name, and Volker's Veil is the test. Named panels keep their calibration and are untouched.
• Dark flats load as bias. Ray's NINA night held ninety one-second darks that are dark flats at his flat exposure, which the wizard has no slot for. The tile is now Bias / dark flats, a frame typed or named dark-flat files itself there, and the note says so instead of "no lights share their exposure".
• The final vote never dies for want of one band. Randy's Seestar S30 Pro night died at the final vote for want of two hundred megabytes. When an allocation fails the vote takes that band in halves, and halves again if it must - bit-identical, a little slower, with a line in the receipts.
• The memory guard, measured. Beta 3 added a guard that runs fewer frames at once when memory is short, and its estimate was thin: a 26 megapixel frame in flight was measured at two and a half gigabytes, not two. The guard now uses the measured figure, so big sensors with modest RAM run fewer frames at once and finish. This is also the honest answer to "it got slower": on small frames the pace is unchanged, on big ones the guard trades speed for a machine that survives, and lifting the ceiling of ten workers bought nothing here.
• The pile, and the lanes. The pile tiles take a folder again, subfolders included, and loose files have their own door. A frame stamped LIGHT by mistake (Frank's NINA flats) is refused with a button that takes it anyway, so the sweep that protects a real night is kept. The temperature gap between darks and lights is said out loud past five degrees (Alan). The receipts carry the clock on every line (Volker), and the mosaic reading's note only speaks for Auto.
• The Handoff on a mono night. Mike's WR 134 shelf, one camera in six filters, was called six different rigs; it now gets the filter line. Panels that never met their mosaic no longer travel to AstroWizard as a set, and the failure line says what they can and cannot do. RGB and narrowband stacks never travel together, because AstroWizard combines one family at a time, and a filter that stacked in one panel rides along with the mosaics instead of being left behind.

Thank you Volker H., Mike A., Randy M., Robert H., Ray, Alan, Frank and Alec - and everyone who sent a night that did not work. wizard@lukomatico.com

7 SEPTEMBER 2026

AstroWizard: The Steady Hand

• The hang is gone. John's report carried the whole story in its stacks: the interface thread inside a button press, waiting for the maths workers to hand back their bands; eleven workers idle; one worker inside an interface call, waiting for the interface thread. Each waited for the other for ever. Python's garbage collector runs on whichever thread is allocating, and when it ran on a worker and found a dead interface object it disposed of it there, through the interface thread that was waiting on that very worker. The banded maths now holds the collector while its workers run. Reproduced on demand and cured, and it is very likely the beach ball Dean, Neal and Marco met after a palette or a spell - if one ever visits you again, the console is the clue.
• Spells off the interface thread. Mike's copper spell on a nine-panel canvas left the window not responding for ten hours. The colour spells now run off the interface with a counter, a second cast queues behind the first instead of being refused (John's Cool Down then Add Warmth twice), and the bands in flight are counted against free memory, so that canvas takes a few seconds within the memory it has instead of ten hours in the swap. Bit-identical.
• Flatten the Sky, built in. A card on the Background step beside GraXpert: a gentle sky surface fitted to the darkest parts of the frame and subtracted - the tilt of light pollution, the bowl of vignetting - nothing to install, a second to run, with Hold to view and Reset Step as for GraXpert. Two guards keep it honest: little clear sky in the frame, or a removed surface as tall as the sky itself, and it says so and points at GraXpert's AI, which stays the stronger tool for curved gradients and nebula-filled fields.
• Linux and the Mac. The UI size change on Linux wedged the window (Kevin, Alec); the setting is now saved first and applied at the next launch, with a line saying so. StarNet2's failure note reads the engine's own words: a missing weights file gets the one line that fixes it - StarNet2_weights.pt belongs beside the StarNet2 program - and the suspects it lists are your platform's own (Alessandro).
• The set door. When StackingWizard hands over several stacks, a filter that arrives more than once is named rather than silently reduced to its first file, and a set carrying both RGB and narrowband says which family it left out.

Thank you John P., Dean T., Neal G., Marco M., Mike A., Kevin K., Alec and Alessandro - and the wall now carries 837 wizards. wizard@lukomatico.com

6 SEPTEMBER 2026

StackingWizard Beta 3.1 - The Walk

• The walk. Volker's Seestar night in framing mode - 411 subs stepping across NGC 6888 - came out of Beta 3 as thirty-five panels, thirty-five tabs and a staircase. A night whose pointings step across the sky by a fraction of a field is now read as one walk: the pointings stack on their own inside, meet on one growing canvas with the sky levels matched and the seams crossfaded, and the finish shows the canvas alone - one result, one save, one send, and the pointing stacks' safety copies in a folder of their own. Proven on a synthetic walk of twelve pointings; a real framing-mode night from a Seestar or Dwarf is the test, so send yours in.
• The Mosaic toggle. The Pile's "They're mosaic panels" button becomes a switch: Auto, One canvas, Panels or Off, with a line saying what Auto found. A heavily dithered night is never assembled in panels by accident, and a named-panel night can be told to grow one canvas instead. A change re-sorts the night on the spot.
• The corner that stood aside. Mike's 3x3 came back with its corner missing. The placement pairs raw star positions with a tight tolerance, and this rig's field distortion moves a corner star by ten pixels, so every placement scraped by on a handful of stars and the corner, whose overlaps lie on the most distorted ground, found too few and stood aside. Now, once the optics are measured from the panels that did place, every panel that stood aside is tried again with the distortion undone - nine of nine on Mike's masters - and a mosaic that still lacks a panel says so in the receipts and in its count instead of quietly showing a hole.
• Sky levels, solved together. The levels used to be matched along a chain of neighbours, which never closed around a ring of panels - part of the darker square in the middle of that 3x3. Now one level plane per panel per channel is solved from every overlapping pair at once, with the anchor panel as the gauge: smaller seam steps on the nine-panel test set. The rest of Mike's square came from the capture, not the blend.
• The deepest stack is the grid. Combining stacks of different exposure lengths (Volker's NGC 6888, four exposures across four nights) used the shown stack as the grid, and with the thinnest as the grid the deeper ones could not be placed and the combine died. Now the deepest stack - frames times seconds - is the grid, with a receipt, and a grid that can host nothing hands over to the next deepest. And the star finder no longer hunts at the edge of the data: a master laid on another cohort's grid carries a black band where its frames never reached, and the step from sky to nothing posed as a ruler line of false stars. Four of four on his masters, in every order.
• The Vespera declares itself. Vaonis writes an exposure, a Bayer pattern and an object name but never a frame type, so a Vespera frame with no type word anywhere is a light (Roger's Vespera Pro and Anton's original both left the pile empty).
• Field fixes: the Pile's notes wrap to the window (Volker's green text ran off both edges); a panel list past six folds to one line; the cohort tabs scroll sideways past eight; the working-files picker sits on the Looking Glass before the first frame moves (Mike had to start and stop a stack to reach it); the save button says "Save the combined stack" once a combine exists (Darran); the pile buttons take many files at once and a frame that declares itself a light is filed as one wherever it was picked from; a big night shows its reading progress instead of looking frozen; and the colour door fits a small Mac's screen and resizes (Marco).

Thank you Volker H., Mike A., Marco M., Darran, Roger, Anton, Shawn and Neal G. - and everyone who wrote in with a night that did not work. wizard@lukomatico.com

6 SEPTEMBER 2026

AstroWizard: The Fine Brush

• NoiseXTerminator's bands, untangled. The engine refuses its plain denoise switch beside any of its band switches, and the Custom popout could send both - Frank's "conflicting denoise options". The four cells (intensity and colour, fine and coarse) are now resolved from the strength and the Custom sliders and sent as one plain denoise when they agree, else as the four cell switches on their own.
• Engines that read from a network share. The RC-Astro engines cannot take a file on a share by its network path - one spelling they reject as an unexpected argument, the other they cannot match (Bill's Synology). An engine input on a share is now copied to a local working file and its result brought home to the share.
• The starless file follows the frame. Remove stars, crop, then Save Starless used to save the uncropped starless; the starless working file now takes every framing op like the star layers already did (Alex).
• A raw frame is named on load. A two-dimensional file carrying a Bayer pattern is a single unstacked frame, and the app now says so instead of showing a grey checkerboard - stack it first. And the dual-band palette warns when the guidance profile says broadband: there is no OIII and Ha to separate in such an image, so the palette can only drain it of colour. Skip it unless you shot through a dual-band filter (Phil's grey stretches).
• Previews, by area. The small previews inside tools - the Stretch Wizard grids and their tiles, the linked-against-unlinked compare - built their thumbnails by taking every fourth pixel, which crunches the noise and drops small stars between the strides. One area-averaging shrink now serves every tool preview, so they should look smoother on a dense screen; tell me if yours still look coarse.
• Small things: the working-file label keeps its row - a long path is shortened from the front so it never runs under Undo, Redo and Compare Original; a stale working copy of ours from an earlier session is cleared before an engine runs, so StarXTerminator no longer refuses with "output file already exists" after a restart (Marco); the window remembers its size and place (Stephen); the step dots name their tools on hover (Lloyd); the Settings note gains a mail door, the setup text names your own platform's folders, and the palette warning's window fits its message on the Mac.

Neal's Mac stall after the palette has not returned since a fresh download and a local copy of his file; Marco's is still being chased, and it has not shown itself on any machine here, the M5 MacBook on the same kind of data included - if a stall visits you, the console is the clue: start the app from Terminal and send what it prints. Thank you Frank L., Bill L., Alex, Phil W., Marco M., Stephen, Lloyd, Matthias, Neal G. and Eric - and the wall now carries 773 wizards. wizard@lukomatico.com

4 SEPTEMBER 2026

StackingWizard Beta 3 - Magic Mosaics

• Every desktop. The beta now ships for Windows, Apple Silicon Macs, Intel Macs and Linux - the same wizard and the same four steps, signed on Windows and notarised on the Mac, built from the AstroWizard recipes that already proved themselves.
• Mosaics, first light. Panels your capture software named (NINA's _PANEL_2 grammar and friends) are found on the Manifest, each stacks as its own cohort on its own anchor, and when the cohorts are done the panels are placed by their stars onto one canvas big enough for all of them - sky levels matched in the overlap, the seams a slow crossfade, a deeper panel weighing more. The mosaic is the stack you save and send; the panels stay behind their buttons. A night whose pointings sit far apart without names is asked about rather than guessed at, a wide-lens panorama is refused with a reason, and a lone frame caught mid-slew is set aside. The panels are placed by consensus (every star a panel shares with every neighbour pulls it into line, hundreds per panel) and the optics are measured and undone - a RASA puts a corner star four or five pixels from where a flat field would, and every seam is made of corners. Proven on Mike Ales's two-panel and nine-panel NINA nights: nine 26-megapixel panels assemble in about three and a half minutes, and against PixInsight's own mosaic of the two-panel set 67 stars agree to three quarters of a pixel. Tell me how yours turns out.
• The curved fit. Every frame used to be placed on the reference by a straight affine fitted to a dozen bright stars. On optics with field distortion that is not enough for a frame from the far side of a meridian flip (its star field sat two to six pixels off at the edges while the dozen fitted fine), and it was never as precise as it could be for any frame. Now each frame is also fitted on hundreds of stars with a second-order curve, used only when it clearly beats the straight fit, and resampled through it. Measured on the nine-panel night: post-flip frames go from about 1.4 px to 0.2 px, ordinary frames from about 1 px to a quarter of a pixel. Sharper edge stars on every night that crosses the meridian, RASA and Hyperstar owners especially. The honest cost: a big-sensor frame takes about half again as long to align as before, and the number of frames aligned at once now follows the memory actually free - a nine-panel night of 26-megapixel frames took 42 minutes on the workshop PC.
• The middle of the turn. On multi-night alt-az data the reference frame used to be the best frame wherever it sat in the night's rotation, and the canvas paid in wasted edges. The grading now measures the night's turn and anchors on the best frame from the middle of it, saying so on the Glass and in the receipts. The four-night Seestar set that started this turns close to 80 degrees end to end; measured on it, the part of the canvas holding at least half the frames grew from 69% to 95%, with the deep core unchanged.
• Darks per cohort. One master dark per exposure length and gain, swapped per frame like the flats; a ready-made dark library (a Dwarf's) is read by nearest temperature and never blended across temperatures; the size guard and the master votes trust the majority of your lights, not the first file in the folder, and say both sizes when they disagree.
• Think this should be colour? A colour camera whose files carry no colour pattern is read from the pixels - the green phases agree, red and blue sit apart - and the Manifest offers the two possible readings of one sub side by side. Your pick is remembered per camera, outranks a header that says otherwise, and "no, it's mono" is a pick too.
• Working files, the rules. They go on the subs' own drive when it is internal, and never spill silently onto a USB RAID or a network share (a force-quit that held a RAID for minutes taught this one). The room is checked before the night starts, with the ways out spelled out; a drive picker lets you choose; a slow drive is named by a timed write before you wait on it; the frames live in 2 GB pieces so a FAT32 stick's 4 GB ceiling cannot stop a night; a crashed run's leftovers are cleared at the next launch at once. And the cache lets go: it used to keep every aligned frame it had written resident in memory as well as on disk (Windows keeps a mapped file's pages in the process), so a big-sensor night quietly ate the machine; now a frame is written and released, and forty 26-megapixel frames later the stacker still sits under a gigabyte.
• Hot pixels, by their stillness. Six frames spread across the night tell a defect from a star by whether it moves, and defects never star again - no darks needed. The "matched too tightly" warning keeps its tripwire but now names its evidence (matched points sitting on whole-pixel centres) and speaks softly to a night that is simply steady.
• Field fixes: finished stacks you already have go straight to the Handoff (a step-1 button, the Manifest's door filling the shelf, and stacker-signed files recognised); same-colour stacks split only by exposure length offer a one-press combine; Send to AstroWizard never traps you behind a missing safety copy and sends the set only when the night holds more than one filter; the frame column sorts into time order; the pixel peeper zooms 1x to 4x and remembers it; a Dwarf 3's slightly different frame sizes fit; a remembered folder on an unplugged drive no longer stalls the Mac's picker; a night with no recognised lights (a Vespera's files, a darks-only folder, two finished stacks and nothing else, the wrong folder) no longer dies silently on the Pile - the Manifest draws and offers its doors, which was also why Alex's two masters only appeared after a trip to step 3 and back; the number of frames aligned at once now follows the memory actually free rather than the core count, so a big-sensor night no longer runs the machine dry mid-stack; and the beta notes no longer claim mono is missing.

Thank you Volker H., Neal G., Stefan H., DrD, Wayne W., Mike A., Alex, Mitja S., Trey M., SaltyPogue, Russ McN. and Bob C. - and everyone who wrote in with a night that did not work. wizard@lukomatico.com

3 SEPTEMBER 2026

AstroWizard: The Steady Hand

• The curves hang, found and fixed. On some Macs a tone change followed by Curves left the preview saying "rendering" for good and the app beach-balling. Three things were feeding each other under the hood: the window re-rendering itself in a loop every time its own preview label changed size, the live preview running the full image on the interface thread on wide screens, and the sharp full-resolution pass handing its result back from the wrong thread. All three are gone - the windows only re-render on a real resize, the live tier is capped, and results come back through a queue on the interface thread. Verified on the bench and on a Mac in the field; if yours still stalls, tell me.
• The free crop grows up: a Lock row (1:1, 3:2, 4:3, 16:9, 9:16) keeps the box's shape while you resize, move and rotate it, a Tilt box sets the angle exactly and follows a hand rotation, the fine rotation dial gets a degrees box and a reset, and the shade outside the box is a proper dimmed image on every platform. Page 2 loses the old Shave buttons - the free crop does that job now - and page 9 gains a "Last-minute crop?" for the spot you notice right before saving.
• The star layers follow the frame. Remove stars, then crop, shave, flip, rotate or bin - the separated stars take the same cut, so Replace Stars keeps working; Reset: Cropping hands them back too. Save Stars Only returns beside Save Starless on the export step, and both are always there, greyed until stars have been separated.
• Two new dials in Settings. A compute-device choice (Auto, GPU or CPU) that every AI engine honours, and a model version per tool - BlurX, NoiseX and StarX take a version number, GraXpert's background and denoise models too. The Custom NoiseXTerminator popout grows the CLI's full dial set: intensity against colour, fine grain against coarse blotch, the scale between them, and the passes.
• The range law. Float TIFFs of any scale, and tool results that come back on another scale, are normalised to the working range and the view is re-solved, so a result can no longer render white or black. A 0-1 image with a few hot pixels above 1 is clipped, not divided into darkness.
• Field fixes: GraXpert's watchdog no longer mistakes a graphics-provider warning for a crash, and never puts down an engine that is still computing (Clive's twenty-minute CPU denoise on Linux); StarNet2 starts from its own folder and is handed its weights outright, so the Mac build finds them (Stefan), and older StarNet2 builds skip the doomed progress flag; Combine Mono Channels' Browse no longer opens behind the window; leaving "See both, pick one" without a pick no longer blanks later previews (Mike); the Custom BlurXTerminator popout sizes itself to its content (Mitja); going back now names the step it goes back to, and the step numbers show their names on hover (Mitja); a low-disk guard warns at 2 GB and refuses an engine the drive cannot hold the result of (Chris).
• Small things: tips off gives the tools that room back; a "Say the word" button opens a mail to the wizard from the export step and Settings; the Community Guide has a pointer in Settings; on macOS a file dragged onto the app icon opens; exported TIFFs carry an sRGB profile; macOS trackpad scrolling of the panels is smoothed.

Thank you Neal G., Nathan S., Clive B., Stefan H., Chuck W., Mitja S., Mike A., Alex, Chris - and the wall now carries 725 wizards. wizard@lukomatico.com

2 SEPTEMBER 2026

Windows, signed

• Both Windows downloads carry a code signature now - AstroWizard.exe and StackingWizard.exe are signed with a Microsoft-validated publisher certificate through Azure Artifact Signing, with a timestamp, so Windows can see who built them and the signature stays valid for good. They are the very same 31 August builds with the signature added and nothing else changed, so their checksums on the release pages changed with them; the Google Drive copy of AstroWizard was swapped in place. Mac and Linux are untouched.
• What you may still see, for a while - SmartScreen weighs a brand-new signature by how many people have run it, so “Windows protected your PC” may linger briefly: More info, then Run anyway. It fades as the signature earns its reputation, and because the reputation belongs to the publisher rather than to each file, it should not reset with every build the way the unsigned days did.

31 AUGUST 2026

StackingWizard Beta 2.4: the mono era

• Mono filter sets, in full. Drop a night of Ha, OIII and SII (or L, R, G, B) and the wizard reads every filter dialect your capture software writes - single letters, full words, or filters named only in the filename. The manifest declares the night ("an SHO night: Ha ×24 · OIII ×23 · SII ×21"), the frame list grows filter and exposure columns, and every frame is graded against its own kind: same filter, same exposure length, same night. A moonlit session no longer executes a clear one, and a 30s minority no longer dies against a 10s majority.
• One night in, several stacks out. Each filter stacks separately to one shared alignment, with its own honest display. Flip between the finished stacks with a click, save them all into one folder, and press Send once: AstroWizard opens with Combine Mono Channels already filled in - S to red, H to green, O to blue - with the final say still yours.
• Flats brew per filter - dust wears a different face through every filter, and now each light meets its own filter's flat. Calibration shot through a capture sequencer (type card says light, the name says bias) sorts honestly, and a finished stack hiding among subs is recognised by its own "105x1m" naming and set aside.
• Quality of the wizard's life: a finished night clears in one press (and the same escape hatch lives on page 1), autosaves tidy their own older copies, crash leftovers sweep themselves at startup, the window opens properly maximised at last, zoom steps gently and panning stays smooth at any depth, and the stacking maths now uses many more of your CPU cores.

Thank you Russ McN., Doug H., Bruce S., SaltyPogue, Climb Bike, Barry W. and the five-for-five tester - and the biggest thank you to the corpus of real nights that caught a dozen bugs before any of you could meet them. wizard@lukomatico.com

31 AUGUST 2026

AstroWizard: the receiving dock

• The mono handoff: several stacks arriving from StackingWizard are read by filter and Combine Mono Channels opens itself pre-filled - SHO, HOO and LRGB all recognised, the channel-type declaration still a visible human choice.
• The crop learns Lightroom's manners: draw your own box on the image, corners resize, inside moves, and dragging just outside rotates the picture live behind the still frame - auto-filling so no dead corner ever enters. A fine rotation dial and a second row of aspect ratios (3:2, 4:3, 2:1, 9:16) join it.
• Curves and the narrowband tune stop wading on big images - while your hand moves only the dialog preview renders, and the sharp full-resolution pass lands the moment you rest, computed off the interface so nothing ever freezes. The Custom BlurXTerminator popout grows the full dial set: stars, halos, nonstellar, and a manual PSF diameter for tight star fields.
• Faster where it counts: curves and narrowband commits now run across every CPU core - identical results, measured about twice as fast on large images.

The two wizards were built to meet in the middle. This is the release where they shake hands. wizard@lukomatico.com

27 AUGUST 2026

StackingWizard Beta 2.3: the list that cannot die

• The frame list is rebuilt from the ground up. The old list drew one widget per frame and simply gave up past a few hundred - rows went invisible, suspects vanished off the bottom, and four of you reported it four different ways. The new list draws only what is on screen, so two thousand frames scroll like twenty. Long filenames trim in the middle (the ends are what tell frames apart) and hovering shows the full name.
• A UI Scale dial, top left, 90 to 140 percent, remembered between launches - for 1440p screens, 4K screens, and eyes of every age. The previews now render at your screen’s true resolution at any scale, and the layout holds still while a stack builds.
• Headers that arrive broken get repaired on reading. Some cameras write illegal header cards - one tester’s QHY put its colour pattern down without the quote marks the FITS standard requires, and every build until now refused the whole night. His own frame proved the cure and now lives in the wizard’s test bench. And when a frame truly cannot be read, it is marked unreadable in red and set aside - never again graded 70 by a table of made-up numbers.
• The Dwarf family is spoken here. Dwarf 3 and Dwarf Mini never label their frames, so their nights landed as “unidentified”. The wizard now recognises them by their own naming, reads their nested dark libraries, survives their two cameras sharing one calibration folder, and takes exposure times from filenames that state them.
• The manifest says what it found, before you stack it - mixed targets, mixed filters, mosaic panels, duplicate copies, and where an exposure time came from. One test folder held two targets in three filters plus a finished stack and a rejected frame; every one now lands where it belongs, and the mix is declared up front.
• The finished stack looks like the night you watched. The live view and the final result now share one display ruler from first frame to last - same brightness, same colours, same smoothness, at every zoom and every scale. What you watched build is what you get.
• The hot-pixel warning wakes earlier - a night where only some frames lock onto sensor defects (stacked unmoved, showing as doubled stars) now gets named at three frames, where the old warning waited for most of the night to be lost.

Thank you Kevin C., Des F., Charlie M., Russ McN., Denis H., Morgan H. and Richard D. - and the test folders full of deliberate traps that found two more bugs before any of you could. Keep them coming: wizard@lukomatico.com

26 AUGUST 2026

AstroWizard: The Seeing Build

• Every tool preview is a real viewport - Curves and the step-8 tune window fit the whole image by default, scroll-zoom through the actual pixels, drag to pan, double-click to refit. At depth you are looking at real data, never an enlarged thumbnail - and the tools render live on the main canvas too, where a full-resolution pass lands the moment your hand rests, so zooming the big view stays sharp as well. Two of you asked for this by name: Bob wanted to see the whole picture while shaping a curve, David wanted sharpness when he zoomed. Both, done - this build was promised to you.
• ✨ Happy already? Keep this look - step 7’s preview is already a solved stretch, so a new button commits exactly what you see: same engine, same statistics, full resolution, one click. No more hunting for the manual profile that matches what your eye already approved.
• The two wizards now talk - StackingWizard Beta 2.2 ships stacks untrimmed and names their well-covered core in the headers. AstroWizard reads that note and solves its display stretch from the core, so the honest edges never steer the view. Load a fresh stack and it simply looks right.
• UI Scale - one dial in Settings, 90 to 140 percent, remembered at every launch. Bigger text and buttons everywhere for eyes of every age - and Undo and Redo have grown into proper targets while we were at it.
• Small kindnesses - a contact line in Settings so nobody has to hunt for wizard@lukomatico.com, and the Fellowship wall rebaked at 528 wizards with every name in its highest hall.

Thank you Bob A. and David D. - and everyone whose reports shaped tonight’s StackingWizard build below. Two apps, one night, all of it from your emails.

26 AUGUST 2026

StackingWizard Beta 2.2: the honesty build

First, a correction. Beta 2.1’s notes said hot pixels could no longer pose as stars. That was true for mono cameras and wrong for colour ones, where processing spreads a hot pixel into exactly the shape our filter trusts - one tester proved it within a day. The wizard now at least confesses when an alignment is impossibly perfect instead of celebrating it, and the real cure is being built for Beta 3. We will keep correcting ourselves out loud like this - it is a beta, and things breaking is rather the point.

• Every frame stacks exactly once - one tester’s folder had been through another program first, so a second copy of every sub sat in a sub-folder, and the wizard stacked the lot: 48 “frames” that were 24 exposures twice each, a result overstated by 41 percent, and double the time and disk to pay for it. Two frames from one camera can never share a timestamp, so now they cannot share a seat - and the manifest says how many copies were set aside.
• The auto-crop is gone - it existed to hide scruffy edges, and we patched it twice, but a meridian flip taught us there are nights no percentage can referee: it cropped a 5496-pixel sensor to a 434-pixel strip and handed that to AstroWizard without a word. Rather than destroy pictures quietly, the wizard now ships your full canvas, honest edges and all, and names a meridian flip when it sees one. Cropping is an editing decision, and it belongs in AstroWizard, where you can see what you are choosing.
• The Pile stops guessing - what the header says now outranks what the filename says, which outranks what the folder is called, so a file named dark can never be swallowed by a folder named lights. The old “longer than five seconds is probably a light” rule is deleted: frames with no identification are counted as unidentified and said so. And the four count tiles are now buttons - point any pile at its own folder and your word is law. Thank you Frank L. for the idea.
• The safety copy asks before outliving its purpose - every finished stack autosaves so nothing is lost to a crash or a forgotten click. But once your real save succeeds, that copy has done its job - four runs left one tester 380 MB of identical twins sitting among his subs. The wizard now asks, once: Remove or Keep, and remembers your answer.
• The finished view stops dimming at the finish line - the completed stack now renders with the same sky brightness as the live view you watched it build at, however deep the night, and zooming in no longer changes the stretch. Extra depth shows as smoothness, never darkness.
• Smaller honesties - the receipt’s “2.5% rejected” was pixels clipped by the voting, not frames thrown away, and now says so. Lights on a network drive get named as the slow part before you wait an hour to find out. The Handoff grew its own copy-the-receipts button. Shift-click ticks a whole run of frames at once, and the mood switch warns before re-grading over hand-made picks.
• For Mac users - handing off to an AstroWizard that is already open should no longer stall. We cannot test this on hardware we do not own, so this one is an attempt, not a promise - if yours still loses its way, tell us.

Thank you to Mike N., whose saved log held two bugs he had not even reported, and to Frank L., Volker H., RickJ, Mike A., Sigurd M. and Dawid G. - this build is made of your emails. Keep them coming: wizard@lukomatico.com

25 AUGUST 2026

StackingWizard Beta 2.1: what the testers found

• Hot pixels can no longer pose as stars - a sensor’s dead pixels sit in the same place on every frame, so on soft or defocused data the wizard could match them perfectly, compute an alignment that moved nothing, and stack the whole night unshifted while cheerfully reporting a fine result. A real star lights its neighbours; a dead pixel does not, and now only the former counts. An impossibly perfect alignment is reported as a warning rather than a triumph.
• Blank frames stop winning - our idea of “best” was really “least noisy”, so a frame with nothing in it beat every real one, became the reference, and left nothing able to match it. One tester’s observatory dome was shut for some frames and his night collapsed to five of forty-four. A finished stack that wandered into a lights folder did the same thing. Neither can lead now.
• Network drives work again - the working files created in Beta 2 arrived on a NAS with their permissions locked down and shut the wizard out of its own cache. They now live on a local disk, made the ordinary way. The space needed is shown on The Stack page before you begin, and the quiet drop to half precision when disk was tight is gone entirely - if there is no room, the wizard says how much it needs and where.
• Your stacks know where they were pointing - saved files now carry coordinates, date, focal length, pixel size, camera, filter and a proper frame count and integration time. They plate solve, and other programs can tell a colour stack is colour. And a colour camera whose filter pattern we cannot read is stacked as mono out loud instead of silently.
• One wandering frame no longer crops your picture in half - the crop asked that every single frame cover a pixel; it now asks nine in ten, for single nights and for combining.
• The column headings stay put while the list scrolls, which seven of you asked for - and it turns out those columns have been sortable all along, which nobody could discover while the headings slid away.
• The wizard explains itself - when a brightening sky sets frames aside it says so, and when one frame was exposed quite differently from the rest of the night it names it. Previews stop shrinking as the stack builds, the running tally is back beside the time remaining, and there is no more name.fits.fits.
• A known-issues list now sits on the StackingWizard page, so you can see what is confirmed and still being worked on. This is a beta and things breaking is rather the point - thank you to Aaron, Phil, Helen, Sigurd, Marcos, Charlie, Karl, Mitja, Frédéric, Mike, Craig, Alex, Daz, Joaquin, Volker, Mario, Elliott, Mitja, Bill, Tom and everyone else who took the trouble to write it down.

24 AUGUST 2026

StackingWizard on Mac: signed, and welcome

• Macs just open it now - the macOS download is signed with an Apple Developer ID and notarised by Apple, and the approval ticket travels inside the app, so it opens even on a Mac with no internet. No right-click, no Control-click, no detour through Privacy and Security. It is the same Beta 2 code as yesterday, simply signed, so the macOS checksum on the release page changed with it.
• AstroWizard's Macs, the same day - both Mac downloads, Apple Silicon and Intel, have been replaced with signed and notarised builds of the very same 21 August code, so they open the same easy way. Their checksums on the release page changed with them; Windows and Linux are untouched.
• Windows is in the queue - the same story on Microsoft's side is with them for review. Until it clears, Defender may still frown at a fresh unsigned build; the false-positive report for the current one is filed.

23 AUGUST 2026

StackingWizard Beta 2: big nights

• Big nights, fixed at the root - the first beta kept every aligned frame in memory, so a 500-frame Seestar night could eat 40 GB and crawl into swap on Windows and Mac alike. Aligned frames now stream through a disk cache that is freed afterwards, memory stays flat whatever the frame count, and the final vote is a new sort-once algorithm running on every core. measured on the workshop PC: 187 Seestar frames in 2 minutes 31 seconds end to end, where the first beta needed over 8 minutes for the final vote on 60 frames alone. Thank you Dean, Marco, Paul, Volker, Alex, and the Dwarf 3 M31 reporter - your receipts drew the map.
• Never silent again - grading, aligning and the final vote each show a measured time left. The "half a minute or so" line was a lie at 521 frames and is gone. A Stop button stands while stacking.
• Stricter alignment - six stars must agree before a frame is placed (four-star fits stacked extra galaxy cores on a Dwarf 3 M31 night); a frame gets a second, harder look before it is given up; frames a Dwarf itself marked "failed_" never lead the stack.
• Small things you asked for - the copy-the-receipts button sits in the card where no window width can hide it; previews fit the panel (portrait Seestar frames showed as a zoomed crop); saved names carry the frame count; "stacked", not "master", in the header; combining leaves out a stack that shares under half the field instead of shrinking your result to a strip; grading folds frames that arrive at a different bit-scale (the S30 Pro "300% sky" frames) before any verdict.
• Known, and next - the Looking Glass list struggles past about 500 frames (grading itself is fine; a lighter list is next), then the uncropped/union canvas and a calibration library. Defender still frowns at unsigned betas: the false-positive report for this build is filed and signing is in progress.

22 AUGUST 2026

StackingWizard: the public beta opens

• A new wizard joins the family - StackingWizard turns a night of telescope frames into a finished stack, simply: point it at the folder, the wizard sorts lights from darks and flats by their headers, grades every sub, aligns and stacks the keepers live, and hands the result to AstroWizard with one gold button. Free for everyone, forever - and very much a beta, with the rough edges to prove it.
• Four steps, in order - The Pile, The Looking Glass, The Stack, The Handoff. No settings to learn; the wizard picks, you look, and every decision leaves a receipt.
• Already got stacked files or masters? - several sessions of one target from the same rig combine into one deeper stack, smart-telescope nightly saves very much included. The wizard even notices when two stacks are secretly the same data, and politely declines.
• Windows and Apple Silicon Macs tonight - Intel Macs and Linux follow. Get it from the StackingWizard page.
• Found something odd? - press "copy the receipts" on The Stack page and email the lot to wizard@lukomatico.com. Every report makes the wizard wiser, and beta reporters are remembered.

21 AUGUST 2026

Badges and Bundles

• AstroWizard introduces itself to RC-Astro - every call to the XTerminators now carries the app's name and build (Russ Croman's own request), so if you ever write to RC-Astro support they can see exactly which integration your machine runs. Newer CLIs only - older ones are left undisturbed.
• Kinder words when RC-Astro is missing - the message now points to the stand-alone tool at rc-astro.com instead of just the Settings path. A field note from the Croman household, with thanks.
• Licence troubles say so - when an XTerminator declines to run because it isn't activated, the top line now says exactly that, and where the 🔑 lives. No more generic "execution failed" for a licence matter.
• No terminal required, ever - activate your RC-Astro licences and update the RC-Astro tools from inside AstroWizard. One button each, in ⚙ Settings & Help: 🔑 view or activate licences, ⬆ fetch the newest build with RC-Astro's own updater. Simple buttons, no command line.
• The Bundle wears gold - the RC-Astro buttons found their colours, and the best-value bundle now says it is one.
• The wall rides at 404 lines - forty-one wizards inducted since Tuesday. Welcome, every one of you.

18 AUGUST 2026 · REV B

Same day, sharper tools

• The colour switch minds its own business - on finished, stretched images the "image looks all one colour?" flip no longer softens the colours you committed on screen. View only - your files were always true - but the canvas now agrees with your Curves.
• Resizing mid-grid behaves - changing the window size while the depth grid is open now closes it cleanly; one click reopens it at the new size.
• Failed loads fail honestly - a TIFF or XISF that cannot be read no longer half-opens a session.
• The wall rides at 363 lines - tonight's inductions aboard the same day they were made.

18 AUGUST 2026

Doors and Dials

• Open with AstroWizard - drag a FITS onto the app, or right-click and Open with, and it loads itself. Groundwork for broader tests of StackingWizard - the wizard's own stacking app, built from the ground up, taking its first steps in the workshop now. More on that when it's ready.
• The depth grid, redesigned - two honest rows on the Big Stretch: the classic stretches with their dial above, the deep stretches with theirs below. Soft has retired after an unclicked career - Custom ×50 recreates its look exactly for anyone who misses it.
• Deep dialing, both moods, one slider - the new Custom deep pane carries its own sky dial and a small keep colours tick: flip it and the pane re-solves on the spot, holding your star colours or equalizing your channels - honouring the colour choice you made at the start, at last.
• Dials live with their panes - the grid's sliders now dock to the bottom edge of their own previews at any window size, instead of floating in the space between.
• The colour switch works on stretched images too - flip "image looks all one colour?" on an already-stretched picture and the channels equalize in the view - a pure cast fix now, no sneaky brightening.
• JPEG saves a touch richer - export quality rises a notch, still a fraction of the old file sizes, with more headroom for anyone who edits your image after you share it.
• The wall rides at 350 lines - 334 wizards honoured, each seated once in their highest hall.

15 AUGUST 2026

The Quiet Cures

• Blank previews heal themselves - if the screen renders pure white or black while your data underneath is fine, the wizard re-anchors the view from fresh statistics and says so in the console. Your image was never broken; now the preview knows it too.
• Silent crashes get caught - GraXpert failures that die without printing a word now trigger the automatic CPU retry too, and older StarNet2 builds that refuse the progress flag get a plain retry that still separates the stars.
• Star tools after Undo - running the star engines again after an Undo no longer trips over their own earlier files.
• Saving leads with JPEG - the save dialog now offers JPEG first at a fraction of the old file size, star colours untouched, because a finished image wants sharing. FITS masters stay one click away.
• Update RC-Astro from Settings - one button runs the official updater; no terminal needed.
• The Stretch Wizard stands alone - manual stretches fold away behind one click on the Big Stretch step, so the guided path is unmissable and the classic buttons remain for those who know their own way.
• Seestar stacks recognised on load - the right profile sets itself from the file’s own header.
• A dozen smaller kindnesses - Save Starless offered whenever stars were separated, copied consoles note when a newer build exists, the Fellowship banner steps aside on narrow screens, the two Hubble mix buttons explain themselves on hover, and pre-stretched data displays true on Final Polish.
• The wall rides at 309 lines - the biggest induction wave yet, every name aboard.

12 AUGUST 2026

Four Platforms

• Intel Macs join the family - a native build for pre-Apple-Silicon Macs (2020 and earlier), field-tested on the wizard’s own old MacBook. Same app, same steps, fourth platform.
• Graphics crashes cure themselves - if GraXpert’s hardware acceleration fails on your machine (older graphics cards, mostly), AstroWizard now retries on the CPU automatically and remembers your machine prefers it. Re-pick the GraXpert path in Settings if you ever upgrade and want acceleration back.
• The white-screen mystery, solved - if GraXpert was set to save in 16-bit, its results came back in a range that washed the whole preview white. Engine results are now range-checked like every loaded file, whatever your GraXpert settings.
• The wall rides at 256 lines - every name from the wave, aboard and remembered.

10 AUGUST 2026

The Library Opens

• The Wizard’s Library is open - free practice skies from the wizard’s own telescopes: twelve volumes across four shelves, from first-light galaxies to the Troublesome Tomes. And there’s a 📚 door right on Step 1, for anyone with a cloudy fortnight and nothing to load.
• Star Colour joins Replace Stars - a second slider that deepens the colour of your returned stars before they land: golds go golder, blues bluer, brightness stays put. A supporter’s “turbo charge button” ask, granted - live-previewed like everything on that page.
• StarNet2 speaks when it stumbles - its own error messages now appear in the console instead of a silent “failed”, and instant failures get honest first aid (driver, the free Microsoft C++ runtime, antivirus - the usual three suspects, named).
• Per-engine GPU cures - when a graphics chip gives up mid-run, the console now speaks the right cure for the engine that fell, instead of one-size-fits-none advice.
• Settings learns StarNet2 properly - it joined the numbered setup list with its own free-download button, beside GraXpert’s.
• The engines take a bow - the GraXpert team, Mikita Misiura (StarNet2) and Russ Croman (RC-Astro) join the Acknowledgements, links and all. The humans made the wizard; these three make much of the magic it conducts.
• The friendly fine print - the app now carries its © line and plain-English terms: free for personal use forever, genuine copies from this page, no re-selling or re-packaging. Kindly meant, kindly worded.
• The wall grew mightily - this build carries every name inducted through the launch wave. Welcome, all of you.

7 AUGUST 2026 · REV B

Same day, small truths

• A text clip fixed on the Image-Saved window - the kindness lines now read in full at default size.
• Downloaded the Mac or Linux build in the first hour of release? Those early files were missing the compressed-image codecs StarNet2’s output needs - this revision makes certain everyone has them. (Windows was never affected, and later Mac/Linux downloads already had the fix - if that’s you, this update is just the text fix and you’re free to skip it.)

7 AUGUST 2026

Stars for Everyone

• ⭐ Star removal is now FREE for everyone. StarNet2 - Mikita Misiura’s free star-removal neural net - joins AstroWizard as a fully supported engine beside StarXTerminator. Install the free StarNet2 command-line tool (starnetastro.com), and AstroWizard finds it automatically in the standard spot. The honest hierarchy is printed right on the card: SXT strongly preferred if you have it - StarNet2 is the free path, and it does a fine job. You kept asking for this one, loudly and often - here it is.
• AstroWizard now does the star maths itself. For every engine, the app computes both star layers from your own image - so Replace Stars is mathematically exact whichever remover ran, and it gains a choice of blends: Exact (returns your stars precisely, to the last decimal) and Soft (the classic PixInsight-style screen blend - gentler over bright areas, never clips). Both preview live. The brightness dial also learned its safe floor: below 0.6 the returned stars could sink beneath a bright nebula entirely, so the dial now starts where every star survives.
• 📂 AstroWizard reads XISF. PixInsight’s native format walks straight in the front door - Load Image or the mono combiner, compressed or not, metadata intact - and converts safely to the usual working format, so the rest of the wizard never changes. Long asked-for (thank you Michael, and Derek whose real masters road-tested it) - welcome, PixInsight stackers.
• 💾 Save Stars Only. Beside Save Starless on the export page: just the separated stars, for your own blending recipes elsewhere (asked for by Mo Moore - whose request arrived attached to a tip; what a way to file a feature).
• The palette mixes now name their data type - “OSC Dual-Band to Hubble” and “MONO SHO to Hubble” - so nobody reaches for the mono mix with a colour camera again.
• The header grew up (and the Fellowship took the throne). A tidy two-row top bar on every screen size: zoom and undo/redo on the left, settings and compare on the right, and The Wizard’s Fellowship at true centre stage with the working-file status and progress bar tucked neatly beneath it. The hall itself now greets the curious with how names get up there - AstroWizard is founded on, and maintained by, kindness - and closes on gratitude.
• Mac kindnesses: the quarantine cure line now includes the crucial sudo (thank you Jody, whose terminal wall of “Permission denied” taught us), and the Locate picker explains that apps look greyed out in a folder chooser - that’s macOS, not you (thank you Michael R.).
• One honest note: the download grew by about 20 MB - that’s a complete image-codec library, so compressed TIFFs from any tool (StarNet2’s included) simply work everywhere.

5 AUGUST 2026

The Steady Eyepiece

• Your screen no longer “goes crazy” after noise reduction. The preview aims its screen-stretch using the image’s own statistics - so when NXT or GraXpert wiped away the noise, the view would suddenly dig far deeper and light up like a different image. Nothing was ever wrong with your data, but the jump looked like damage and has worried people for months. Denoising and sharpening change noise, not brightness - so the preview now simply holds its aim through them: same view, suddenly cleaner. (Your real stretch at step 7 is untouched - this is purely the preview being steadier.)
• The “already applied” note now tells the truth through Undo. Undo a step and the note vanishes along with the work; Redo and it honestly returns. It used to linger after an Undo, claiming work your image no longer contained - which could send you hunting for a problem that didn’t exist.
• Final Polish greets you properly. Its buttons stack by design - so the scary “doubles the effect” warning is gone, replaced by a friendly note that blesses the layering.
• All three caught by the wizard himself in a single night of field processing - and shipped before the sun came up, because holding back a fix this kind felt wrong.

4 AUGUST 2026

The Immortality Update

• 📜 The Wizard’s Fellowship - supporters’ names up in lights, inside the app, for all time. A rolling hall of every kind soul who keeps AstroWizard free: legendary supporters, the whole fellowship, and the newest members right at the top. Scroll it by hand or let it roll - and the top-bar button raises a toast to a random member now and then. Tips to the coinpurse join the wall (here and in the app) forever; Patreon and channel members are celebrated regularly in video credits - and welcome on the wall too, just say the word. Bought an RC-Astro tool through the app’s links? A mighty tip in itself - you belong there too, no receipt needed.
• Replace Stars now restores your stars exactly - and shows you first. A live preview with full zoom and pan (asked for by rendenwow in the comments - great call) lets you dial star strength against your real image before committing. And an honest confession found during the build: ever since the feature was born, replaced stars came back very slightly dimmer than the originals. The wizard’s own before/after test caught it, we measured it, and the maths is now exact - 1.0 means a perfect restore, to the last decimal.
• Ctrl+Z / Ctrl+Y hotkeys for Undo and Redo, everywhere in the app (asked for by Greg - who also correctly diagnosed, from his own screenshots, why an image cropped after stretching was treated as unstretched; both shipped together).
• Saving a JPEG saves a JPEG. The save dialog used to force a .fits extension whatever format you chose - so “save as JPEG” could try to overwrite your FITS master (caught by Chris in the comments, receipts and all). Whichever format you pick now supplies its own extension.
• The framing page joined the flow: crop, shave, bin or rotate and the friendly “done here” banner appears instantly, with honest wording - framing ops are meant to stack. No more wondering why page 2 behaved differently (Chris again - good eyes).
• When an engine can’t start, the app now tells you the true reason. Pointed at the PixInsight plug-in instead of the standalone CLI? A shortcut instead of the program? macOS quarantine? A laptop GPU giving up mid-run? Each case now prints its own plain-English cure right in the console - built from real support cases (thank you Joseph, and everyone who shared their logs).
• Polish throughout: live previews survive “Hold to Compare”, settings survive upgrades from the very oldest builds, warmer words on every credits surface - and every public sentence now punctuates the way the wizard actually writes.

1 AUGUST 2026

The Stretch Wizard

• 🪄 The Stretch Wizard - the Big Stretch becomes two clicks, all by eye. Pick 1: which colours look right (two bright side-by-side previews). Pick 2: how deep - a grid of six live previews of your actual stretch: Soft, Normal, Aggressive, Deep, Deep keep-colours, and a Custom pane with its own dial. Click the look you love and that exact stretch is applied. Nobody leaves the wizard unstretched.
• Synced zoom across every pane. Scroll on any preview and all of them dive into the same region, pivoting on your cursor; drag to wander. Compare how five different stretches treat one dust lane, live - the anchors are solved once and every zoomed view renders through the exact maths that clicking commits, so the previews cannot lie.
• The Custom dial went live. Drag it in the grid and watch your factor render in real time against the presets; the “more…” button opens the full dialog as a remote - including Deep’s depth dial - while the grid stays on screen. The standalone ⚙ Custom button previews live on the main canvas too.
• ◀ Undo / Redo ▶ moved into the top bar - browser-style, one consistent home, scoped to the step you’re on (crossing steps uses the step numbers, which explain themselves). The red “Undo Last” buttons retired from every page: added functionality, removed clutter.
• Curves grew two powers: an S channel - a true saturation curve (aim it at just the blues via the colour-family menu, brightness stays put) - and a Hold: Compare Original button on its preview.
• Final Polish decluttered: the Tune and Custom-Strength dialogs now ride as small ⚙ buttons on their main buttons (the spellbook pattern), and the recommended mix wears a ✦ star no button colour can hide.
• One honest question, as a last resort: heading into Final Polish with an image that looks unstretched now asks “stay & stretch, or continue?” - once per session, because bringing already-stretched data is a perfectly legal move.
• Forged in five same-day field rounds by the wizard himself - every pane, caption, dial and warning in this release was road-tested on real dust before it shipped.

30 JULY 2026

The one where saving doesn’t end the session

• Save as many times, in as many formats, as you like. Saving used to quietly wrap up your session behind the scenes - so saving a JPG and then trying a PNG could fail with a bare “Export failed” (reported in the comments, diagnosed by the wizard himself). Saving is now just saving: FITS for the archive, TIFF for printing, JPG for the group chat - then keep tweaking and save again. Undo even works after a save now.
• 💾 Save Starless Version. Asked for by Cyyberpunk in the comments barely a day ago: if you removed and replaced stars, the save page now also offers your image exactly as it was the moment before the stars returned - every mix, spell and curve you cast on the starless canvas included.
• The Mac download is now a single zip. One unzip and there’s the app - no more zip-inside-a-zip surprise (thanks ascaniosobrero for flagging the confusion; the double-wrap was a packaging quirk on our side, and both layers always contained the same working app).
• Small honesty upgrades: if an export fails, the message now tells you why; cancelling a save says so; and the post-save thank-you note appears once per session, not after every save.

23 JULY 2026 · REV I

The guidance era

• Let the Wizard Guide You. Tell the app what you shot - one optional question on step 1 (colour camera? dualband filter? mono narrowband?) - and a gold trail of suggested clicks lights the way through every step, tuned to your data type. Follow the gold and you can’t get lost; ignore it and nothing nags. The mono combiner asks its own version of the question and even relabels its slots (“S → red, H → green, O → blue”) - the eternal which-file-goes-where question, answered where it’s asked.
• The Wizard’s Way: see both, pick one. The gold-bordered button on the Big Stretch (and step 1) fills the screen with two live previews - linked and unlinked, side by side, neutrally labelled - and you simply click the one that looks right. On the stretch page the panes show your actual stretch both ways, so what you pick is exactly what you get; changing your mind afterwards re-applies instantly.
• Wizard’s Curves - Colour Transmutation Spells. Five of the wizard’s own finishing curves as one-click stackable spells: Yellow to Copper & Gold, Blue to Teal, Dynamic Pop, Add Warmth, Cool Down. And every spell carries a 📖 button that opens the recipe inside the Curves window, fully editable - see the move, learn the move, make it yours.
• Deep learned a second stance: “keep colours”. Classic Deep balances each channel itself (perfect for palettes and neutral skies); the new sibling digs just as deep while preserving your field’s true colour balance - golden Milky Way fields stay golden.
• Normal is stronger now (150, the community-classic value). If you loved the gentler look, Custom 50 recreates it exactly - the Custom dial even marks the spot, and it now travels logarithmically so equal drag means equal visual change.
• Auto Colour Balance learned some humility. On fields whose brightest stars are genuinely warm, it used to “correct” honest star colour into a full green cast (found the hard way, twice, on the same stubborn dust field). Corrections are now bounded to ±12% - a field’s true colour is not a calibration error - and a lurking maths bug that could blank a channel died in the same surgery.
• Flow, everywhere: the step numbers up top are clickable; going back asks whether to rewind or keep your image as-is; finished steps offer a gold “Done here - Next Step”; already-applied steps warn before you double-run an AI tool; and the live console is on by default (it’s the app’s black box recorder - off-switch in Settings).
• Every feature above was forged in marathon field-testing by the wizard himself - including debugging one of his own spells with the 📖 button an hour after it was built.

20 JULY 2026 · REV D

The one that digs deeper - and learns spells

• The Deep Stretch. Last release we promised “round one” on very faint data - this is round two, and it’s a new engine. Instead of stretching harder (which, we measured, just brightens the sky), Deep anchors to your data: it solves its own white point, black point and strength from the image itself. Faint dust fields open fully, M51-class galaxies reveal their tidal halos, and bright cores stay unblown. Thanks to Lee Polikoff, whose unstretchable stack started this whole hunt. One honest note, printed right on the button: Deep digs up everything - a blotchy background means flaws in the data (flats, gradients), not a bug.
• Wizard’s Curves - Colour Transmutation Spells. The wizard’s own finishing curves, straight from years of tutorials, now one-click and stackable: Yellow to Copper & Gold (faint yellows go copper, bright stay gold), Blue to Teal (cores pull to OIII teal), and Dynamic Pop (the gentle finishing S-curve). More spells will join the book over time.
• Stretch wisdom, updated by experiment: with data that came through the earlier steps, stretch linked first - whatever your data type. The wizard A/B-tested his own long-taught “dualband wants unlinked” rule and the linked path won: deeper faint extensions, subtler colour. The tip lives right next to the switch now.
• The stretch panel got simpler and smarter - Gentle retired (Custom covers it), the Custom dial now travels logarithmically with Normal and Aggressive marked on it in plain English, and if a stretch comes out genuinely flat, the app quietly points at Deep instead of leaving you guessing.
• Navigate like you own the place: the step numbers up top are now clickable; going back asks whether to rewind your image or keep it as-is (crop a spot off a finished image without redoing five steps); and revisiting a step whose work is already in your image gets a clear “already applied” warning - no more accidental double-sharpening.
• The live console is now on by default - it’s become the app’s black box recorder and bug-report machine (off-switch in Settings, if you must).

18 JULY 2026 · REV K

The all-night Mac stakeout

• The real Mac bug, caught red-handed. Last release we woke the app from its nap - a true fix, but not the whole story. This time, armed with his own MacBook and Apple's diagnostic tools at 3am, Luke caught the deeper culprit in the act: on modern macOS, the app's “refresh the screen” call could get trapped re-drawing the window forever, so it never noticed GraXpert had finished - minutes ago. The interface looked alive; the part that watches for results wasn't. Fixed for good. Field numbers from the stakeout: background extraction 3 seconds, denoise 34 seconds - steps that used to “hang” for 20+ minutes. Mike Frazier's zoom-trick observation was the clue that cracked both chapters of this saga - poking the window was literally breaking the spell.
• Denoising switched to GraXpert's 2.0.0 AI model on every platform - dramatically faster (about 40× on Apple Silicon), with output quality we couldn't tell apart from the newer model in side-by-side tests. Same model everywhere, so tutorials match your results whatever you run.
• The Big Stretch reaches deeper into very faint stacks - data scaled into the bottom sliver of the numeric range (thanks to Lee Polikoff's file) used to come out as flat grey mud at any strength; the white-point logic now recognises real signal and opens it up. Honest note: this is round one. On the deepest data there's still more in the tank than a 1000-strength stretch pulls out - deeper reach is actively in the works.
• One engine at a time, and the app says so - starting a second AI tool or changing steps mid-run now gets a friendly “hang tight” instead of a silent gamble with your working image. The console also narrates honestly: when the result file starts landing on disk, how big it's grown, and exactly when it's collected.
• Quieter and tougher - the app no longer burns a full CPU core while waiting on an engine, never mistakes an old output file for a fresh one, and if its own plumbing ever fails, the full error prints right in the console ready to copy-paste to us.

17 JULY 2026 · REV B

The one where the app was asleep

• The great Mac mystery, solved - the app was literally napping. macOS puts idle-looking apps into “App Nap”, and waiting for GraXpert is exactly when AstroWizard looks idle. GraXpert finished its work fine; AstroWizard dozed until you clicked something - which is why zooming or pressing Next Step made runs “suddenly complete”. The app now opts out of App Nap entirely. Full credit to Mike Frazier, whose zoom-trick observation cracked the case, and Rick, who called it exactly: “AstroWizard is not seeing the processing has completed.”
• Stackers with messy headers can't break saves anymore - malformed header cards (like ASTAP's broken CALSTAT) are now repaired on the fly instead of aborting your crop. Thanks shreddybear.
• macOS quarantine check: if GraXpert.app still carries the downloaded-app flag, the console now says so before running, with the exact one-line cure. Thanks Greg.
• Windows shortcut trap: pointing Settings at a desktop shortcut (.lnk) instead of the real program now gets an orange warning and plain instructions, instead of a cryptic “not a valid win32 application”.

17 JULY 2026

The night-shift build

• Linux previews work now. A library the image viewer needed wasn’t bundled, so Linux showed “No module named PIL._tkinter_finder” instead of your image. Fixed - with thanks to David Huff and johnsteed for the pinpoint reports.
• The Linux file picker shows extensionless tools - no more renaming GraXpert to .sh just to select it.
• RC-Astro no longer fails on decorated filenames. Files named like “NGC 6992, 2025-08-23, (CV)…” - which is most stacker exports - broke BlurXTerminator with a cryptic “cfitsio status 366”. AstroWizard now keeps its working filenames free of the characters that trigger it. Thanks Pawel!
• Macs can finally pick GraXpert properly. The old file dialog often refused to select .app bundles - one tester’s path ended up pointing at macOS’s own “Apps” launcher, which AstroWizard then dutifully ran instead of GraXpert (thanks Brian - your report solved it, and you were never the problem). The Mac picker is now a folder chooser that accepts GraXpert.app or its folder, and any path that doesn’t look like the right tool shows orange in Settings with an explanation.

16 JULY 2026

The comment-section build

• GraXpert on Macs now talks. Its progress and error messages were being silently swallowed on macOS - they now stream into the console live, and AstroWizard collects its finished result wherever and however it saves it (including Seestar-style .fit files). Built directly from console evidence sent in by launch-day testers - thank you Michael, Greg, Rob, and Brian.
• The Settings window no longer multiplies when clicked repeatedly - which also fixes chosen paths appearing not to save.
• One-click bug reports: the console gained a Copy button. If something misbehaves, Show Console → Copy → paste it in a comment.
• Final noise touch-ups grew to a full set: Light and Normal strengths for both GraXpert and NXT (thanks JustJoe).
• The Step-1 tip now matches the renamed “Image looks all one colour?” switch.
• Version numbers finally agree everywhere - Finder, Settings, and this website all show the same build date.
• Install guide covers macOS Sequoia (Apple removed the right-click → Open trick; it’s System Settings → Privacy & Security → Open Anyway now).

15 JULY 2026

The Mac fix, properly this time

• The real cure for GraXpert hanging on Macs: it turns out GraXpert was finishing its work and saving the result all along - its process just never exited afterwards, so the progress bar bounced forever over a completed job. AstroWizard now watches for the finished file, collects it, and tidies up the leftover process itself. (Thanks to Greg for the detective work - his spot of a double file extension cracked the case.)
• Output files no longer end up named .fits.fits.
• Right-click now removes Curves points on Macs (Macs report right-clicks differently; Control-click works too).
• The green-cast toggle is now “Image looks all one colour?” - it fixes red and blue casts just as well.
• The preview no longer resizes when processing starts and ends - the progress bar has a permanent home.
• Multi-pass noise reduction now shows “pass 1 of 2” as it works.
• The Mac app’s version number now shows the build date instead of an eternal 1.0.0.

14 JULY 2026

The Apple Silicon fix

• GraXpert no longer hangs on Apple Silicon Macs (all M-series chips). AstroWizard was accidentally passing its own internal library settings to GraXpert on launch; GraXpert loaded the wrong libraries and froze. Standalone GraXpert was never the problem - your setups were fine all along.
• The window now fits smaller screens. On 13″ laptops the default window was wider than the display, hiding the Settings button off the right edge.

13 JULY 2026

Zero-click setup

• AstroWizard now finds GraXpert and the RC-Astro tools by itself on standard installs - checked at every startup, so most people never need the Locate buttons at all.
• Selected a folder instead of the program? It now works that out and finds the right file inside.
• Clearer wording throughout setup: rc-astro is one file that runs all three tools - no need to hunt for three programs.

12 JULY 2026

The app that updates itself (politely)

• Update notice: at startup AstroWizard checks this website for a newer build and shows a small banner if one exists. There's an off switch in Settings, and a promise printed next to it: nothing about you or your images is ever sent.
• Your build date now shows in Settings - no more guessing which version you have.
• ASTAP-stacked files no longer load as a pure white screen (different stacking programs use different pixel ranges; AstroWizard now handles them all).
• Masked Saturation and Make it Pop are gentler per feedback, with a new Custom Strength dial for both (1.0× = the classic look).
• GraXpert problems now explain themselves: its own messages appear in the console, crashes are detected and reported instead of bouncing the progress bar forever, and first-time-setup advice is offered when its AI models are missing.
• PixInsight-stacked files no longer slow the RC-Astro tools to a crawl - huge file headers were choking them; AstroWizard now tidies these automatically (and the underlying bug was fixed by RC-Astro too, from our bug report).

9 JULY 2026

The stretch overhaul

• New stretch engine: the linear preview adapts itself to your data, and the Gentle / Normal / Aggressive buttons were rebalanced to be meaningfully different (with honest labels).
• Star-rich targets like the Pleiades now stretch properly instead of vanishing into black.
• The green-cast toggle resets between images so it can't quietly discolour your next photo.
• Tool windows (Curves, Tune, Mask Reach) resize with the window.
• First solid Apple Silicon build (native, no Rosetta) with the Settings window fixed.