Serial execution passes without changing the mathematics
All six cases completed in 38.12–41.27 seconds per bound. Every saved report field matches the predecessor exactly. The screen took 246.08 seconds in total and used one worker with one numerical thread.
The algorithm, physical domains, reference values, conditioning and 120-second per-bound ceiling remain unchanged. This establishes a usable finite execution recipe on the six declared domains. It does not prove a CPU quota, isolate the cause of earlier timing differences or change the failed shortcut's outcome.
The next separately frozen study generates only the twelve missing future certificates. All completed training certificates will be reused. Earlier timeouts remain recorded; none of the completed studies is restarted or given extra allowance.
After those certificates pass, a new saved-only comparison must check complete region covers against the original width target. Recording-length usefulness and the observation and arithmetic assumptions remain further requirements. No new aircraft model preference follows here.