experiment-reconcile

Reconcile field logs from a real machine with the experiment plan that produced them - find each planned run in the unannotated logs, infer what the operator actually did against what was asked (skipped, partial, different setpoints, extra manoeuvres), re-simulate what was actually done, and compare predicted vs actual on the same graphs, run by run, against the plan's tolerances. Then turn the gap into a minimal plant-model refinement on the planned path and test it against the same logs. Use when logs come back from a field session run with a cppmodel:experiment-design plan, when asked how the machine compared to the model's prediction, or when asked to refine a plant model from field data.