Frozen scope: recover missing long-history reports
The preceding six-case software study completed its calculations but rejected exactly matching reports because Python tuples become lists in JSON. Its failure records and all charges remain unchanged.
This separate study can recover only a later cached report that was not saved after that false comparison. Admission requires a completed, hash-verified parent study, exact equality of every available report, complete original and point-oracle records, and compliance with the original time and memory limits. A genuine numerical mismatch or incomplete calculation is not eligible for this repair.
| Limit | Scope |
|---|---|
| Histories | At most four audited globe histories; the two disc histories reuse their complete saved reports |
| Input | The same 96,000 packets, parameters, conditioning proof and two queries at 290 and 438 seconds |
| New work per history | One cached two-query batch, at most 192,000 packet steps |
| New point-oracle or original-cold calculations | Zero |
| New optimizer starts or original aircraft reads | Zero |
| Maximum total new cached batches | Four |
| Maximum charged bound steps | 6,912,000 |
| Numerical allocation | At most four workers, one thread each; frozen before launch |
| Per-batch time limit | 14,400 seconds |
| Whole-study time limit | 21,600 seconds |
| Peak memory | At most 1,536 MiB per worker |
| Retries or allowance expansion | None |
The saved audit fixes the actual task set before preparation. No run can start until the parent closes and the numerical regression passes. The passed regression must still match its recorded code hashes. The repair reuses the common native batch library and accepts no cold-solver or point-oracle callable.
Every returned report and its work accounting are saved before comparison. Acceptance requires exact serialized report equality, agreement of all 51 coefficient columns, and the unchanged 0.00001-metre software arithmetic allowance against the saved independent point checks. This allowance is not a sensor accuracy specification.
All six histories must have complete verified comparisons to qualify long-history reuse. Passing does not qualify whole-record uncertainty, observational coverage, an Earth-model detection or any new use of protected confirmation observations.