Giving this a body, from the roadmap audit 2026-09-05. It has been open since February with a one-word title and an empty description, while being the only thing users report unprompted - so it reads as ignorable and it is not.
What it should mean now: the non-macOS performance gap. All profiling to date has happened on one macOS machine. The bad numbers come from Windows and Linux users on scenes that should be trivial, and nobody has reproduced them on hardware.
Three things have to happen before any optimisation work is worth starting, and none of them are code:
Establish whether the number is real. Users are photographing the fps readout in scene_stats_hud.dart. That readout has never been validated - see the perf HUD issue filed today. If it is misleading, it is manufacturing this issue.
Control for the toolchain. Flutter 3.47-beta Impeller renders strokes about 2.5x too thick under canvas zoom, and the same code renders differently depending on whether it was launched from the terminal (stable 3.44.8) or Android Studio. Reporters on mixed toolchains are not comparable to each other.
Get a save file with a repro. Same ask as the PNG crop bug - the shape is almost certainly scene-specific rather than something a demo map shows.
STRATEGY.md section 11 sizes the eventual pass at 20-40 hrs and calls it both a consumer conversion risk and an authority evaluation risk.
Note the board also currently has this sitting in the Done column, which given the above is almost certainly wrong.
Giving this a body, from the roadmap audit 2026-09-05. It has been open since February with a one-word title and an empty description, while being the only thing users report unprompted - so it reads as ignorable and it is not.
**What it should mean now: the non-macOS performance gap.** All profiling to date has happened on one macOS machine. The bad numbers come from Windows and Linux users on scenes that should be trivial, and nobody has reproduced them on hardware.
Three things have to happen before any optimisation work is worth starting, and none of them are code:
1. **Establish whether the number is real.** Users are photographing the fps readout in `scene_stats_hud.dart`. That readout has never been validated - see the perf HUD issue filed today. If it is misleading, it is manufacturing this issue.
2. 2. **Control for the toolchain.** Flutter 3.47-beta Impeller renders strokes about 2.5x too thick under canvas zoom, and the same code renders differently depending on whether it was launched from the terminal (stable 3.44.8) or Android Studio. Reporters on mixed toolchains are not comparable to each other.
3. 3. **Get a save file with a repro.** Same ask as the PNG crop bug - the shape is almost certainly scene-specific rather than something a demo map shows.
STRATEGY.md section 11 sizes the eventual pass at 20-40 hrs and calls it both a consumer conversion risk and an authority evaluation risk.
Note the board also currently has this sitting in the **Done** column, which given the above is almost certainly wrong.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Giving this a body, from the roadmap audit 2026-09-05. It has been open since February with a one-word title and an empty description, while being the only thing users report unprompted - so it reads as ignorable and it is not.
What it should mean now: the non-macOS performance gap. All profiling to date has happened on one macOS machine. The bad numbers come from Windows and Linux users on scenes that should be trivial, and nobody has reproduced them on hardware.
Three things have to happen before any optimisation work is worth starting, and none of them are code:
scene_stats_hud.dart. That readout has never been validated - see the perf HUD issue filed today. If it is misleading, it is manufacturing this issue.STRATEGY.md section 11 sizes the eventual pass at 20-40 hrs and calls it both a consumer conversion risk and an authority evaluation risk.
Note the board also currently has this sitting in the Done column, which given the above is almost certainly wrong.