Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Targeted random-arm L1 follow-up

The controls were predeclared from the 15 L0-admissible random routes before any L1 result was inspected: rank 1, the median rank 8, and rank 15. L1 is the impulsive Sims–Flanagan approximation, not the continuous-thrust L2 validation.

L0 rankVariantL0 diagnosticL1 outcomeL1 scoreL0−L1 gapMax mismatchMax throttleWorker-sActual evals
1`3-3-3-6-100000`658588.701RefinementNotClosed289300.288369288.4131.0744401.0286991393.1
8`3-3-3-2-6-1000011`212995.658PropagationFailure3738.0
15`3-2-2-3-2-2-2-2-3-2-3-1000000000100`7005.533PropagationFailure97.3

No promotion passed the declared L1 threshold. The leader returned a finite diagnostic score, but its normalized endpoint mismatch remained far above 1e-7. Both controls encountered typed Kepler propagation failures; their zero actual-evaluation fields mean the exception occurred before the retry layer returned its evaluation counter, not that the failed promotion consumed no compute. Worker-seconds retain the observed cost.