Decision-Closure Boundary Mapping in Gemini: An Exploratory CFC Benchmark Series

This record freezes an exploratory black-box benchmark series developed under Comparative Feedback Control (CFC) to examine whether Gemini preserves decision-closure boundaries under synthetic changes in applicability, record identity, conflict coverage, cross-audit transfer, priority resolution, referential integrity, execution state, and context identity.

The historical series uses identifiers V1-V100. Original materials for V1-V4 were not recovered and are deliberately left unscored. V6 was an unstable unstructured baseline for which an exact aggregate could not be recovered. The reconstructable/scorable set therefore contains 95 variants and 285 replications: 283 semantic PASS, 2 semantic PARTIAL, and 0 recovered semantic FAIL, corresponding to a strict semantic PASS rate of 283/285 = 99.3%.

This value is a reconstructed scorable-set statistic, not an exact V1-V100 benchmark accuracy and not a general estimate of CFC or Gemini effectiveness. In the strongest exact block, V78-V100, all 69/69 scheduled clean replications scored semantic PASS. The series is exploratory, adaptive, and was not preregistered.