M01 first execution: failed evidence serialization
================================================
The first execution reached final summary writing after producing18response CSVs
and3finest-mesh volume NPZs. JSON encoding rejected NumPy boolean scalar values.
Per-run record writes had also encountered that encoding issue and were caught
as failures; the original progress file therefore does not describe12 valid saved
reports. No successful final summary was written for this attempt.

The original partial outputs remain in ../results/. The exact original runner,
test script and test recorder are preserved beside this note. The engine and frozen
plan were not changed. The repaired runner converts NumPy scalar values to native
JSON scalars and records successful rows only after writing their report. A small
serialization regression test was added. No scientific parameter or gate changed.

The identical prescribed matrix is executed again into ../results-v2/. The two
attempts must not be called two independent experimental validations. There were
no new ODE integrations or physical trials. Current source hashes and final reports
identify the repaired execution; this failed attempt remains part of the journey.
