Interpreting Residual Control Gains in Quadrotor Simulation - simulation records and code

S5 contains holdout trace accounting and a 480-episode targeted force-path intervention. S6 contains the complete corrected-platform comparison: six classical families, five SAC training seeds, 400 checkpoint-development evaluations and 17,280 confirmation evaluations, including all 5,038 failures and independent reaggregation code. The archives preserve their original bytes and identifiers.

Historical PyBullet results use the legacy force path; S6 uses the corrected path. The independent rigid-body implementation uses RK4 integration. These are simulation results for one nominal vehicle and finite development budgets, not real-flight validation.

Full high-frequency trajectories, learning checkpoints and replay buffers are not included in this compact deposit; they remain in the authors archive. The deposited compact records support the reported aggregate comparisons. See README and the SHA-256 manifest for file scope and verification instructions.