Capture, masking, visual QA, and asset production notes for many small shipped apps.
Several posts had placeholders where concept diagrams belonged, and posts with no video still rendered a video section. One commit added 17 diagrams and hid the empty video sections. One post still couldn't be fixed the cheap way.
The locale codes an Android app uses are not the codes Play listings use. And one language the app supports has no store listing at all — in any code.
Porting a web dashboard into an app, I matched every number byte for byte. The visible defect was not a number — it was a color, and the original had already avoided it.
Even with a viewBox, rasterizers honor the SVG's intrinsic size and pad the rest of the canvas. File size and resolution both look correct, so nothing caught it until I opened the file and looked.
The monochrome layer uses alpha only, not color. If the foreground has an opaque background square, the cells and pattern disappear into a solid block — and the presence of the tag guarantees nothing. A 17-app sweep.
Screens that need a session can't just be launched, so I extract them as value-only views and render PNGs from unit tests. I rendered a fully transparent screen and the previous test's drawing came back — 3,676 pixels of it.
A user told me the App Store preview video wasn't showing. The asset was flawless — state complete, zero errors, MD5 identical to my local master. The problem was coverage.
Shooting store screenshots per locale, the screen title was clipped only in the Arabic cut. It looked like a width problem, and there was evidence for it. Widening didn't fix it, and drawing the same string three times changing one variable showed the culprit wasn't the font, the weight, or the width — it was a negative letterSpacing. An improvement for Latin density was erasing letters in another script.
A blue-accent app shipped store screenshots where the selected category chip was lavender. I hadn't used the wrong color. I had left color slots undefined.
requestAnimationFrame is tied to tab visibility. Screenshot tools do not foreground the tab. So I finished a canvas animation and lost every way to judge whether it looked good.
One of five screenshot-caption A/B experiments came back rejected under 2.3.7. The images Apple attached were not the captioned shots I had made — they were paywall screenshots that had been live for months. Apple counts 'free trial' and 'save 41%' as price references. Here is the OCR audit of 267 live screenshots that found the same violation in 8 apps.
A paywall crept into screenshots from a 16-locale auto-capture pipeline. Which tab got contaminated differed per locale, so I couldn't pin the repro. Four causes overlapped, but the most important finding was in verification — the size gate checks 'is there a screen', not 'is it the right screen'.
Auto-capturing 16-locale screenshots via a launch argument quietly broke on iOS 26. The command succeeds and files appear, but the folder is named 'ja' and the contents are English. A pipeline that ran fine for years started producing wrong results one day — and the size gate can't even detect it.
My web-app promo shorts are built by screenshotting live pages. When a page returned 404 or 500, the script baked the error screen straight into the video. The fix was to check for HTTP 200 before capturing.
Even after switching to App Store Connect for screenshots, the pipeline only requested the Korean locale. 34 apps had already uploaded Japanese store screenshots, and the Japanese shorts were putting Japanese captions over Korean screens.
No Playwright dependency, no manual clicking. Just the headless Chrome CLI at the right viewport width — plus the one width value that stops two-column result cards from clipping.
Auto-picking app screenshots from a local folder quietly pulled in another app's permission popups and a login screen. The fix was to pull store-ordered screenshots from the App Store Connect API.