A 600-run experiment using Gemma on cyber CTF labs found no difference in solve rates when models were told how many steps remained. Baseline runs hit 67.6% success compared to 65.5% for step-aware runs.