The loop that worked was procedural: the Steering Committee proposes rulebook changes (three-quarters vote in Cycle 1, a quorum afterwards) and the Council approves or changes them by majority vote, so every cycle got a new rulebook. Rules traceable to findings include the three-stage vetting after the Cycle 2 outcry, the CDBG program-and-service pot after Cycle 1 projects ran into public-funds law, nonprofit prequalification after partners without capacity stalled projects, the Cycle 3 focus on 20 to 34 year olds and online voting after two cycles of under-representation, lowering the voting age to 14 in Cycle 4, targeting assemblies at areas where fewer residents voted, and slimming the ballot from 33 to 12 proposals. The measurement was anonymous surveys and exit polls at assemblies, delegate meetings and polls, compared at p=0.01 with Census and ACS data, written up by the City for Cycles 1 to 3 only; no summary report, Research Board output or survey data exists in the archive after Cycle 3, and the Cycle 2 and 3 reports were both published in May 2016, well after the cycles. Cycle 3 evaluation also found something the convener did not want: online voting made representation less proportional, which the City answered with Stanford-designed targeted online advertising rather than a rule change.