Historical: Extended task catalog and calibration utilities

TaskSpec and task_outcome(sim) are stable Core contracts: a task names its setup, rollout defaults, optional primary metric, and anchors, and the result exposes that declared key with its task-native value. A measured anchor also declares the scored_ticks interval where its normalised value is valid. A different scoring window keeps the raw value and withholds normalisation with an explicit status. This feature record tracks the extended task implementations, calibration utilities, and control settings composed through that interface.

Calibration and outcome definitions operationalise behaviour. They do not validate the named cognitive capacity, make raw metrics comparable across tasks, or turn one rollout into an independent scientific result.

cartpole_plank_easy is now a core coordinate of the frozen direct-control benchmark. It retains experimental task status, and earlier scope-setting Falandays measurements remain exploratory rather than benchmark evidence. Poor performance is reported directly; the protocol does not exclude the task or hide it inside a cross-task aggregate.