Complete the long comparison without repeating valid references
The initial freeze has a report-writing incompatibility: it assigns one artifact name to several progress records, whereas accepted study artifacts are immutable. The numerical calculation is unchanged. Long-history qualification remains pending.
Allow started reference calculations to finish. Do not signal workers or edit their frozen sources. Admit no additional work in that freeze. Preserve its completion, failures, charged starts and any complete cold-reference reports.
Before another freeze:
- Exercise the comparison with the real immutable artifact writer. Confirm the regression fails, then give each phase record a unique name and rerun it.
- Inspect the terminal first attempt. A completed 290-second original reference can be reused only if its sealed artifact and full input/source/proof/environment bindings verify. A partial report or mere charged call does not qualify.
- Freeze the same six cases, 290/438-second queries, physical domains and all acceptance rules in a separate study. Reuse verified original references; calculate 290-second references only for previously uncharged cases. An incomplete charged original call blocks this bookkeeping-only repair. The new batched calculation still runs for every case and must match both complete original reports.
- Charge all new calls before execution. Preserve interrupted allowances and distinguish prior reference work from the new study's measured work. Never restart the old study or transfer its unused allowance.
The test of scientific assumptions is unchanged. This repairs study bookkeeping; it does not loosen numerical equivalence, physical bounds or prediction targets.