Compress the P50→P90→P99 band the market prices as risk. Lower variance is lower risk premium: more debt, cheaper equity, lower premiums. The real $$cost of capital$$ move.
Per-panel observability makes output recoverable and loss improvable (did the stow fire at 70°? is a module degrading pre-failure?). Measured excess return the model leaves on the table.
Solar is underwritten off priors (NOAA hail contours, TMY weather), not the asset’s measured reality, so the P50 you sell and the P90/P99 you are financed against sit a wide, expensive band apart. Verified, panel-level data is the Bayesian update: it compresses the band and lifts the financeable floor (+9 at P90, +16 at P99), lowering the cost of capital and insurance exposure.
The 9–13% of output lost to undiagnosed faults, soiling and mismatch is invisible to today’s sensing. Per-panel truth makes it observable and recoverable, and makes loss improvable per asset: measured alpha, not a modelled promise.