D2: Adaptation: labelers and verdicts

Model: gpt-6-astra. Effort: a: high.

Use one leg at a time in a study. Each leg continues the previous native checkpoint. Reader jobs and separate studies can run in parallel. Closure automatically validates, archives and dispatches. Repeat counts and pilot rounds below are starting examples.


python3 r-place/aggregator/handler/codex/run.py step STUDY D2a --start

Talk through the live panel or run.py chat ACTOR. Open a checkpoint as an interview without changing the study. See OpenAI operation.