Check serial execution without changing the calculation
The single-thread whole-bound profile completed the original specified-field calculation, including all required prefixes, inside its 90-second bound allowance despite profiler overhead. Earlier four-worker calls sometimes exceeded 120 seconds. This is evidence to test allocation, not proof of a CPU quota or a causal speed improvement.
Use the unchanged direct-source implementation, not the midpoint shortcut. Retain all six original sector-zero child domains, complete 1,600-packet inputs, 7.2-second query, clock, source/reference and conditioning. Compare every converted report field exactly with the sealed original six-case preflight. Run one case at a time without a profiler. Record bound wall time and separate whole-evaluator wall/CPU times; the latter also include setup and compilation.
Separate finite allowance
- Six identities and six bound calls; at most 86,400 logical bound steps.
- Six reference setups and child proofs; 36 parent support checks.
- Six source/derivative/point setups and at most 9,600 visits in each separately reserved stream, retaining the original per-task resource recipe.
- One worker, one numerical thread, no other study active.
- Preserve the original 120-second cooperative ceiling per bound. Use a new 900-second overall ceiling for this serial study. No retries or signals.
- Zero native trajectory oracles, LP proposals, optimizer starts, original aircraft reads or protected ESA outcomes.
All six cases must finish inside the unchanged per-bound limit and reproduce their complete saved reports exactly. This qualifies a finite execution recipe, not a new mathematical method, an independent hardware benchmark or a claim about concurrency's cause. Earlier failures and budgets remain unchanged.
Only a passing screen permits a separately frozen serial generation of missing future certificates. The complete-cover width target, recording-scale bounds and observation-error assessment remain separate requirements.