Separating width-search and final-basis effects
The preceding refinement changed the width-search basis together with the final solve basis. Its large flux differences therefore could not distinguish those effects. A bounded factorial diagnostic now varies each setting while holding the other fixed, preserving every remaining saved input.
The reporting checks pass (54 focused synthetic checks pass in each mode, including separated basis effects, interactions, signed channels, incomplete evidence and consumed-control drift; 553 unchanged recorded checks are retained separatelyC-204). They exercise separated effects, interactions, signed flux, missing evidence and disagreement with consumed controls. The original output driver now records native selected widths and search growth alongside the retained mode spectra. It links the same pinned solver library. Preparation checks the single-pass conditions needed to interpret the registers; the wrapper’s missing diagnostics remain an explicit limit.
The package is frozen, verified from its actual execution directory and submitted once under the standing bounded-compute and diagnostic-storage permissions. Controls reproduce the preceding basis cells before scientific interpretation. Submission does not establish convergence. The current model, guard and acceptance criteria remain fixed; the separate feedback-sampling proposal remains unrun.
Technical details → Verification › reference basis study
Decisions → OQ-33 · OQ-35