See shotfleet work
First, a few short clips of the commands. Below them is my list of what I have actually run shotfleet on, with dates, and what happened.
Clips
The clips are real terminal recordings from one session on the fixture apps in shotfleet's own repository; titles, subtitles and the sped-up labels were added afterwards.
Is this Mac ready?
Set up a project
Every language on iPhone
Every language on Android
Catching the wrong language
Checking screenshots you already have
Store rules
Export to fastlane
In CI
Evidence
What I ran, when, and what came out. I only list runs that are written down. The note under each line names the file where I keep the record (run logs, check.json), in the research folder of the public repository. If a record you want to see is missing, email hello@shotfleet.com.
Built to run on macOS 13 or later on Apple silicon (the build declares 13.0 as its minimum), but I haven't run it on macOS 13. My full runs, with simulators and emulators, were all on macOS 27. On macOS 14 and 15 I ran the installer, doctor and check for version 0.1.6, and on macOS 26.2 in a virtual machine for 0.1.2. Other macOS versions are not tested yet.
Releases
- 2026-10-11, release 0.1.9. The signed, notarized build passed 65 of 65 assertions on iPhone and iPad simulators and an Android emulator in 31 minutes 37 seconds, on Maestro 2.0.10 with iOS pinned to the 26.4 runtime. It took three tries. The first said it had passed, but 16 of its Android assertions never ran, because my script had Python 3.9 on its path and the helper that runs them needs 3.11 or newer. That was my mistake. The second failed: the four iOS simulators stopped answering while my Mac was short of memory, and 19 assertions failed. Neither one counts. The third passed, with the Android assertions included. The same fixture app also ran on Maestro 2.11.0 in the first job: Android passed 30 of 30 in 2 minutes and iOS 36 of 36 in 15 minutes 45 seconds. Record: docs/compat.md, research/019-gate-2026-10-11/
- 2026-10-11, the new 0.1.9 stop, on the shipped build. On a made-up React Native app with no translation files,
checkexited 2 and asked for--strings, andinitexited 2 where 0.1.8 had exited 0.runstopped at "Maestro isn't installed" on my Mac before it got to the strings check, so that part is proven by the unit tests only. No device was used. Record: research/019-gate-2026-10-11/proofs.out - 2026-10-08, release 0.1.8. The signed, notarized build passed 65 of 65 assertions on iPhone and iPad simulators and an Android emulator in 24 minutes 55 seconds, on Maestro 2.0.10 with iOS pinned to the 26.4 runtime. Then the same fixture app ran on Maestro 2.11.0: Android passed 30 of 30 in 1 minute 13 seconds and iOS 36 of 36 in 16 minutes 40 seconds. My Mac's swap was over 90 percent full the whole time, so these are slow-Mac timings. Record: docs/compat.md, research/018-gate-2026-10-08/
- 2026-10-08, the new 0.1.8 changes, from source. On the fixture app with Maestro 2.0.10 and iOS 26.4: a plain-text
scrollUntilVisiblewas translated, every flow had the settle step in front of each screenshot on iOS and Android,zh-Hanswas used aszh-CNon an Android build that shipszh-CN, andcheck --appon the two wger Flutter builds asked for--stringsand then checked 12 and 18 screenshots with 0 problems. A missing iPad set in a parent folder showed asmissingwith exit code 1, and[expect]withkeypassed on the right screen and failed on the wrong one. The regex warning, the port 7001 message, the splash message and the kept failure screenshot were tested with fakes only, not on a device. The Mac was busy, so the busy-Mac warning showed up. Record: research/018-device-check-2026-10-08/ - 2026-10-08, release 0.1.7. The signed, notarized build passed 65 of 65 assertions on iPhone and iPad simulators and an Android emulator in 35 minutes 50 seconds, on Maestro 2.0.10 with iOS pinned to the 26.4 runtime. Record: docs/compat.md
- 2026-10-07 and 2026-10-08, the new 0.1.7 commands. On the fixture app, German and Japanese, iPhone and iPad:
runskipped the iOS 27.0 simulators and said so, did not retry a flow that failed, andrun --retry-failedcaptured only the failed language.checkprinted the coverage table and--compare, andexport --print-uploadprinted the fastlane command. The Mac was busy, so the busy-Mac warning showed up. Maestro 2.0.10 only. Record: research/017-device-check-2026-10-07/ - 2026-10-08, the live 0.1.6. On GitHub's hosted Apple silicon runners with macOS 14.8.9 and 15.7.9, the installer,
shotfleet --version,doctorandcheckran again, this time on the build customers get now. No simulator or emulator was run there. Record: research/macos-14-15-live-0.1.6-2026-10-08.md - 2026-10-06, release 0.1.4. The same signed build passed 29 of 29 assertions on an Android emulator and 36 of 36 on iPhone and iPad simulators. The first iOS try ran out of memory on my Mac, so iOS passed in a second run a few hours later. Record: docs/handoff.md section 17, STATUS.md
- 2026-10-06, release 0.1.3. The same signed build passed 36 of 36 assertions on iPhone and iPad simulators and 29 of 29 on an Android emulator, in two separate runs. My Mac didn't have enough memory to get one full run with all three through. Record: docs/handoff.md section 17, docs/compat.md, STATUS.md
- 2026-10-05, release 0.1.3. On Android, shotfleet reads the screen text from Maestro's UI hierarchy instead of from pixels. On the Android fixture app it passed 30 of 30 assertions with Maestro 2.11.0, and 29 of 29 with Maestro 2.0.10, where it still reads pixels. Record: docs/compat.md and the Android fixture-app run log
- 2026-10-05, release 0.1.2. The signed build ran the fixture app on iPhone and iPad simulators and an Android emulator and passed 65 of 65 assertions in 37 minutes. Record: research/0.1.2-plan.md (Outcome), docs/handoff.md section 15, docs/compat.md
- 2026-10-05, same run. In the iPad part the Mac was heavily loaded and 3 of 4 simulators stopped answering; shotfleet finished all 4 languages on the 1 simulator left, with 0 problems. Record: docs/handoff.md section 15, research/0.1.2-plan.md
- 2026-10-05, release 0.1.1. On GitHub's hosted Apple silicon runners with macOS 14.8.9 and 15.7.9, the installer, shotfleet --version, doctor and check all ran; no simulator or emulator was run there. Record: research/limits-platform-reach-2026-10-05.md, docs/compat.md
Real apps
- 2026-10-01 and 2026-10-04, my app, WOD Tracker. 18 languages on iOS and Android, 0 problems, on Maestro 2.0.10. Then 19 languages each on iOS, an Android phone and an Android tablet, 0 problems, on Maestro 2.11.0; 9 iOS languages needed shotfleet's retry. Record: PLAN.md, docs/compat.md, research/maestro-2-11.md
- 2026-10-05, timing. WOD Tracker in 19 languages on 4 simulators and 4 emulators with Maestro 2.0.10: 404 s of capture on iOS and 415 s on Android phones, 0 problems, no retries. The Mac was not idle: its load went from 86 to 497 during the iOS part. Record: docs/compat.md, research/workarounds-maestro-resources-2026-10-05.md
- 2026-10-03, wger (Flutter). iOS and Android, German and Japanese: German on every screenshot. wger's Japanese is about 30% translated, so those screens show English, and check said so. Record: research/frameworks.md section 15, research/real-apps-2026-10-03/
- 2026-10-03, the obytes template (Expo). iOS and Android, English and Arabic, with Arabic right to left on the Settings screen. Its onboarding and login screens are hard-coded English, and check flagged them. Record: research/frameworks.md section 15, research/real-apps-2026-10-03/
- 2026-10-03 and 2026-10-04, the KotlinConf app (Compose Multiplatform). Android, German and English: 6 screenshots, 0 problems. On iOS its first screen came out blank (it draws after its text is readable) and check flagged it; a clean iOS run is still to do. Record: research/frameworks.md sections 15 to 17, research/real-apps-2026-10-04/
- 2026-10-04 and 2026-10-05, Microsoft's .NET MAUI localization sample. Five languages on iOS and two on Android, one screen each: 0 untranslated selectors and 0 problems. Not yet a real MAUI app or a Release build. Record: research/frameworks.md section 5, research/experiments-2026-10-04/C/ and C2/
- 2026-10-04 and 2026-10-05, a small Capacitor test app. The per-app language reached its web view on iOS 26.4, iOS 18.6 and Android 16. Maestro read the web view's text on iOS but not on Android. This tested the language only, not a full shotfleet run. Record: research/frameworks.md section 7, research/experiments-2026-10-04/B/ and B2/
- 2026-10-06, AnywhereRoles, my app. shotfleet 0.1.4, a full run on both platforms: Android, 15 languages × 8 screens = 120 screenshots on 1 emulator, 0 retries; iPhone, 18 languages × 6 screens = 108 screenshots on 2 simulators, 3 languages retried once. Check found 0 problems in both. Before it, check on the app’s old Google Play screenshots found 127 problems: 103 translated captions over English screens, 15 false “missing” lines from a 0.1.4 bug with fastlane supply’s file names (fixed in 0.1.6), and 9 others. Some frames are on the home page. Record: research/anywhereroles-run-2026-10-06/ (run.json and check.json of each run, the console logs and the run report’s findings)
- 2026-10-07, Kuokka, my Hacker News reader. shotfleet 0.1.5 on iPhone: 6 languages × 5 screens = 30 screenshots in three runs of two languages, on 1 iPhone 17 Pro Max simulator (iOS 26.4), 0 retries, 154 s, 132 s and 252 s. Check found 0 problems in all three (the check reads the pictures; it does not prove that every tap hit the right button). A debug build from an unreleased branch, and live Hacker News data at capture time. Six frames are on the home page. Record: research/kuokka-run-2026-10-07/ (run.json and check.json of each run, the console logs, the flow and a note on which frames I left out and why) On 2026-10-08 I ran it on Android too, with shotfleet 0.1.6 on a debug build of 2.0.15 from main, which is a different build from the iPhone one (2.0.14). 1 emulator, two runs of three languages, 5 screens each, 30 screenshots, 0 retries, 139 s and 126 s. Check found 0 problems, with the APK and also with the app’s strings.xml. Maestro 2.0.10 saves no screen hierarchy on Android, so check read the pictures by text recognition only, which is weaker evidence. Six of those frames are on the home page too. Record: research/kuokka-android-run-2026-10-08/ (run.json and both check results of each run, the logs, the flow, and a note on every frame I looked at)
- 2026-10-07, WOD Tracker, my workout timer and log app. shotfleet 0.1.6 on a debug build of version 1.6.7 that was in store review at the time. iPhone: 3 runs of 2 languages, 4 screens each = 24 screenshots on 1 iPhone 17 Pro Max simulator (iOS 26.4), 0 retries, 366 s, 400 s and 384 s. Android: 2 runs of 3 languages, 6 screens each = 36 screenshots on 1 emulator, 0 retries, 553 s and 528 s, with demo data seeded through the app’s debug link. Check found 0 problems in all five runs (the check reads the pictures; it does not prove that every tap hit the right button). On iPhone it had the view hierarchy to read; on Android it read the pictures by text recognition, which is weaker evidence. In setting up, one iPhone attempt was green but a Hindi frame showed the wrong screen; I caught it by looking, and those frames are not shown. Twelve frames are on the home page. Record: research/wod-tracker-run-2026-10-07/ (run.json and check.json of each run, the console logs including the failed attempts, the flows, and the findings for shotfleet)
- 2026-10-08, wger, an open-source Flutter app (not mine). shotfleet 0.1.6 on debug builds I made from the tag 2.1.1, because the latest commit wouldn’t build on Android that day (the maintainer has since pushed the fix). Only the screens you can reach without an account; I never logged in. iPhone: 3 runs of 2 languages, 6 screens each = 36 screenshots on 1 iPhone 17 Pro Max simulator (iOS 26.4), 0 retries, 136 s, 144 s and 129 s. Android: 2 runs of 3 languages, 6 screens each = 36 screenshots on 1 emulator, 0 retries, 177 s and 204 s. Check found 0 problems in all five runs (the check reads the screens; it does not prove that every tap hit the right button). On iPhone it read the view hierarchy; on Android it read the pictures by text recognition, which is weaker evidence. In a failed attempt the first frame was the splash screen; I caught it by looking, and it is not shown. Twelve frames are on the home page. wger is licensed under the AGPL-3.0-or-later and isn’t affiliated with shotfleet. Source: github.com/wger-project/flutter. Record: research/wger-run-2026-10-08/ (run.json and check.json of each run, the console logs including the failed attempts, the flows, and the findings for shotfleet)