Measured. The same question (why did checkout conversion fall last week, one cause planted in synthetic data) ran through each wiring on gemini-3.8-flash with thinking off, 12 seeds each, scored against the planted truth. Accuracy: solo 100%, flat 100%, supervised 100%. Cost per run: solo $0.0027, flat $0.0082, supervised $0.0083. The runs in that table: 181 calls, $0.23. A first pass with a contradictory cause label scored solo 75%, flat 75%, supervised 83%: the solo agent said 'unclear', the flat crew voted confidently for the wrong cause, because scouts whose slice could not tell copied the first draft. One sentence fixed, every wiring 100%. The replay above is seed 10 (truth: pricing change), real events with time compressed. Code and every run log: crew-wiring-lab. By Nikunj Kaushik.