venture-learnlisted

# Venture Learn Target venture directory: `$ARGUMENTS` Read the target venture's completed experiment definition and raw results. Require at least one recorded observation and a completion timestamp before treating the experiment as completed; a cancelled experiment may inform a later decision only through an explicit learning note about the cancellation, not fabricated result evidence. Persist a learning note under `learning/` that separates: - observations; - interpretation; - unexpected behavior; - evidence created; - assumptions strengthened or weakened; - experiment limitations. Classify the result against the locked `preregistration.design.signals`: set `results.branch` to `success`, `failure` or `ambiguous`, and quote in the learning note the preregistered signal the observations meet. Classify from the observations, not from the decision you expect to make; when they meet signals of more than one branch, say so and choose the branch the signals themselves rank first, never the more convenient one. Create stable first-party evidence IDs for material observations. Update the venture state without deleting contrary evidence. Replace current `next_action` with the next stable `N###` object whose single instruction is to run `venture-decide` again. Set `type: decision`; point `assumption_id` at the assumption most changed by the experiment when one clearly dominates, otherwise null. Never rewrite `latest_decision.snapshot` while recording learning. The snapshot rem