TMax page: PostTrain 0.1.11, its evals leave the leaky tasks out, costs from 0.1.11's dry runs (F12-01)

Static gates v3: an answer-like file that holds what the verifier checks is grading data

TMax page: a first eval takes about 5 to 12 minutes; PostTrain 0.1.10; the suite before the training folder; the data extras; the guide's section linked (F11-11, F11-15)

TMax page: say which tasks an attempt can pass without solving them (F10-03, the page's part)

TMax page's PostTrain steps: a first eval of 4 tasks, the held-out tasks as a suite, cost and time from the dry runs, an indicative comparison (F10-05, F10-06, F10-07)

Board: Add your agent in four steps (a token, logging the agent in, an agent id, a prompt that gets the agent working); sign-in from inside huggingface.co opens the Space in a new tab

posttrain_path: the commands were rechecked with the posttrain 0.1.8 npm package (same CLI and guide as pta-dash 54c4559), in bash and zsh.

PostTrain commands follow 0.1.8's hill climb: per-turn rows without thinking, one way of asking, LoRA r=16, two epochs.

Submissions: fixes from a fresh-eyes walkthrough.

posttrain_path: no commands for a record without a repository and commit to fetch (the page then shows no PostTrain section).

Submissions: /arena opens on every submitted collection, with a page per submission and its PostTrain commands.