CORRECTED SUCCESSOR APPARATUS - NOT USED IN FROZEN MAIN EXPERIMENT
Pure wrapper fixes omitted recovery max_steps and inaccurate public message. New partial score adds an explicit length group and is not interchangeable with frozen partial scores. Original controller/tests/traces stay unchanged for exact replay. Any future inference requires a new independent labeled protocol/freeze; do not install this as an unrecorded patch to the original study. It does not repair semantic split overlap; future banks must enforce cross-split payload uniqueness before evaluation.
