Muse Wonder, Ling 3.1 Flash, LFM 2.5 and Gemini 4 Argon added

Ling 3.1 Flash kept returning to the browser between revisions. In a same-task sample drawn from 10 earlier models, its browser-call count exceeded the reference median on 33 of 34 uninterrupted runs. Its Brick Breaker transcript shows why those extra visits can matter: browser checks exposed invalid Phaser method calls, followed by code edits, reloads and keyboard and drag inputs. The later console observations no longer contained those errors.

Muse Wonder took a different path: all 35 saved sessions included mouse, keyboard or scroll input, while 15 contained a recorded edit-and-recheck sequence. Its Drum Machine session combined clicks and dragging with keyboard input—an informative place to inspect how it exercised the controls.

The other additions need a closer look at individual results. Gemini 4 Argon encountered rate-limit and upstream-timeout interruptions; nine LFM 2.5 sessions saved no index.html. These records describe the observed build process, not a ranking of the finished apps.