Skeleton first vs told to render early: time to first render

24 runs. 12 of them put a partial page on screen before the finished one; the other 12 went from nothing to their final page in one step.

Nothing scoreable on screenUnder half the briefHalf or moreThe whole briefFirst render
0s30s60s90s120s150s180s210s240s
First renderWhole briefStages
cc-sonnet-5.5
toldrun 1 19.9s19.9s 1
toldrun 2 24.9s24.9s 1
toldrun 3 19.6s19.6s 1
skeleton firstrun 1 7.5s37.8s 4
skeleton firstrun 2 7.5s55.8s 5
skeleton firstrun 3 8.7s50.9s 5
cc-opus-5.5
toldrun 1 31.1s31.1s 1
toldrun 2 36.9s36.9s 1
toldrun 3 33.0s33.0s 1
skeleton firstrun 1 16.6s62.6s 5
skeleton firstrun 2 13.7s40.1s 4
skeleton firstrun 3 12.7s57.7s 5
pi-gpt-5.5
toldrun 1 28.4s28.4s 1
toldrun 2 27.1s27.1s 1
toldrun 3 30.6s30.6s 1
skeleton firstrun 1 27.5s54.1s 4
skeleton firstrun 2 20.9s72.0s 5
skeleton firstrun 3 12.5s-- 4
pi-gemini-3.8-flash
toldrun 1 44.0s44.0s 1
toldrun 2 45.1s45.1s 1
toldrun 3 45.3s45.3s 1
skeleton firstrun 1 19.9s112.2s 5
skeleton firstrun 2 28.3s89.7s 4
skeleton firstrun 3 16.1s71.5s 5

Each bar is one run, from the prompt to the moment the run ended (where the bar stops), shaded by how much of the brief the judge could see on screen. Agents are ordered by median first render. Stages counts the distinct scores the page passed through: 1 means it went from nothing to its final page in a single step.