Agent-based model · one step = one week
Each row is an evaluation heading for its decision window, the pale band on its track. Reaching the desk is not enough: the last hurdle is being believed. The λ dial prices how heavily users discount an involved evaluation's independence; the guard is the reform that protects how involvement is read.
grey line: scoped · teal line: delivered · teal ring: window open · grey blocks: competing demands · bar: commissioner belief
One seeded run of eval_use_abm_v3 · one evaluation in two shown · the λ dial and guard re-score the credibility gate live, mirroring the model's gate-5 formula (ρ=0.60, penalty cap 0.60) · costs use a US$350k average per evaluation, scaled by length · the statistics that matter come from 300-seed averages, not this single trace.