Corrected source packaging
The initial pilot ended after 72 charged setup failures and zero native calculations. Its Python dependency walk omitted the C file used by the value loader. No scientific fit or forecast ran. The completed run, input arrays, charges and failure records remain unchanged.
The corrected definition explicitly includes both native translation units and requires a smoke test that loads them from an isolated source snapshot. It reuses the verified decoded input bytes from the failed run; it does not read originals again or change the selection, splits, physical bounds, starts, weights, scoring, solver or forecast equations.
Use a separate study identity and separately frozen allowance of 72 starts, 14,472 native calls and a 24-hour cooperative wall ceiling, with one worker and four threads. The previous allowance is closed, not resumed or replenished. Cumulative accounting therefore includes 72 failed setup identities plus up to 72 corrected identities. Preserve both runs in the public record. New failures remain failures; this correction does not authorize retries or limit expansion.