Offline practice validation

Runtime: Python 3.12.14, standard library only.
Unittest distinct test methods: 46. All 46 passed.
Subtest repetitions and repeated runs are not additional test methods.
Original invalid JSON counterexamples rejected: 12.
Actual CLI result: complete JSON equals expected_costs.json.
Seven independently checked totals, USD:
0.208, 0.54, 0.208, 0.238, 0.64, 0.081, 0.2025.
Separate storage-only examples, USD: 0.005 and 0.01.
Explicit generation totals: null; cache-creation price not confirmed.
Input fixture SHA-256 preserved before/after final executions.
Canonical synthetic policy source and Q01-Q03 match the article example.
ZIP member SHA-256 and CRC-32 checked in delivery preparation.

Provider API calls: 0.
Actual model token measurement: NOT_RUN.
Actual model answer quality: NOT_RUN.
Actual model speed/cache observations: NOT_RUN.
Included rates and token counts are stipulated arithmetic inputs.
