Left and Right Target Balancing

Replays show the first two reserved starts for each of the first four agents, where present, in every training run. Selection does not depend on success. The aggregate includes every agent and reserved test. Upright success requires completing the test and keeping the final-half angle RMS below five degrees. Except for the supplied-reward baseline, test pushes have a 5% reversal probability. Paired conditions use the same starts and reversal streams.