Nu Gundam (BG) vs Strike Freedom (BW)

AI summary

Validation of the 10-game adaptive series doctrine (Nu LLM pilots went 9-1 vs SF). Four heuristic combos, 300 games each. RESULT: the series learnings do NOT move the knob-level matchup — all four combos land 68-71% for Nu, within noise of the pre-series 65-70% band. The profile knobs (aggression, blocker discipline) are global scalars and cannot express what actually won the LLM series: per-turn event-count arithmetic (attack events vs ready blockers + shields) and exact-kill sequencing. Nu's engine remains dominant: cheap early curve floods events faster than SF's L3-L4 defense curve can answer; every end is shield attrition (100% shield_empty). To encode the real doctrine, the HeuristicPolicy needs an event-count term in its face-attack scoring, not new profile weights. How to play Nu: curve out, pair Amuro on an expendable body as pre-combat removal, keep attacking in events not damage. How to play SF: blocker count is a resource below 2 total deploy instead of attack; Argama at <=3 shields; count races in attack events; survive to the total-resources-8 LR turn.