Loading battlefield
Recorded Run
Model & Recording Identity
Frozen Guided Comparison
Five paired seeds (71-75) per mission, three controllers. Same map, opponent, physics, and scoring.
| Mission | Original Rules | Tactical Rules | Guided Nimble |
|---|
Tactical rules met all 15 objectives. Guided Nimble met 5, all in withdrawal, saving two units per run rather than three. This is a small harness comparison, not a Jev benchmark or a claim of model superiority.
| Mission | Seed | Controller | Objective | Survivors | Time | Switches | Fallbacks | Discards | Mean Latency | Input Tokens | Known API Cost | Replay |
|---|