Worker software qualification

All 39 targeted checks passed. They cover durable pre-call charges, failed and interrupted work, separate fit/diagnostic limits, changed authorization, duplicate workers, journal integrity, failed-fit abstention, all 27 native derivative columns and the actual fit-to-prediction artifact path.

An independent analytic geographic control exercises the real native value and tangent implementations, fits a training subset, predicts the remaining epochs, and writes compressed receiver-error output. Its maximum position error is 2.94756e-05 m, below the predeclared 0.002 m control limit. Training and full-record predictions agree on their shared prefix. The descriptive probe stage charges exactly 13 native sweeps for this six-parameter software case. The archive cases permit all 26 or 27 active directions and reserve at most 55.

The full worker-task check also verifies training-only initialization, saved receiver/probe hashes and accounting against the on-disk journal. A changed charge journal is refused even when its total count matches. Publication cannot launch the worker; a completed worker cannot enter native integration again.

These are software controls, not NASA measurements or validation of archive coverage. They do not prove global optima, unique calibration, calibrated receiver error or nonlinear confidence intervals. No NASA originals were read and no observed-data starts were made. Native scientific studies remain separately frozen.