Recommendation Admissibility for Finite Computational Benchmarks under Empirical-Resampling Variability and Aggregation-Design Uncertainty

Computational benchmarks may identify a point winner even when the represented evidence does not justify releasing a stronger recommendation. C7 is an executable protocol for deciding whether a bounded algorithm recommendation is admissible on a declared finite computational-benchmark evidence surface. It evaluates two distinct components separately: empirical-resampling variability under a frozen perturbation contract and aggregation-design uncertainty over a prespecified admissible design set. Complete-portfolio support and aggregation-design behavior are then combined through the A1–A5 release grammar, with mandatory evidence or authority failures handled by fail-closed gates. In the FJSP demonstration, V1 retains MWKR as the point recommendation but no candidate achieves complete support, yielding A3 — Sampling-inconclusive. In V2, ECT is point-recommended and supported at both required endpoints; under the additive/common-calibration geometry, that support extends across the declared segment, yielding A1 — Supported and aggregation-design-stable. These conclusions remain limited to the declared finite surfaces, matched portfolios, represented conditions, replication structures, and admissible aggregation-design domains; they do not establish universal solver superiority, future-instance inference, deployment validity, or certification.