Rendered from the original saved readout; no rescoring, fitting or selection.
Synthetic16 generated identities in2 policy languages;2 stochastic blocks.
Each hypothetical complete policy pays for the common initial stage once.
Physical study expense counts the shared initial stage once total.
The primary interval clusters identity and retains both blocks and revisions.
Factor and transition summaries are descriptive, not new selection tests.
Missingness bounds are not confidence intervals. Costs are lower bounds when incomplete.
This is not a Harness-Bench result, a proof of correctness, or a general frontier.
