Saved-primary artifact inventory

Inventory the final six declared model/case primaries for each of the 53 selected stretches: 318 identities, once each. Follow the current verified disposition ledger to its authoritative comparison. Never select a historical fit by cost, substitute a repeat for a primary or treat a fixed-time profile as a full fit.

The batch permits at most 318 primary linearization reads and 318 epoch-artifact reads, with no original recordings, native integration, optimizer starts, dense factorizations or singular decompositions. It is a metadata/array inspection. One batch, no retries after a charged start; rendering completion performs no inspection. Freeze the selection, source metadata, software, environment and this plan before the first primary inspection. Preserve missing and incompatible artifact outcomes explicitly; do not repair a scientific artifact.

Check a declared artifact's SHA-256 against the authoritative fit. A published copy named in that comparison's manifest may supply the same exact bytes. Verify the full parameter order and point, residual cost, all Jacobian columns, receiver position layout and saved epoch times against the frozen receiver selection. For separate epoch artifacts, check that their relative times and selected UTC times differ by one constant origin within eight floating-point ulps at their recorded time scale. This is a metadata consistency check, not independent clock calibration. Do not interpolate positions or alter times.

Distinguish a complete model comparison from reusable saved arrays. If a passing primary has no saved linearization, record that fact; it does not reverse its existing disposition. The inventory may expose array/metadata mismatches that prevent reuse, and those must remain explicit. No missing artifact authorizes a new prediction or fit. A later fixed-parameter projection audit needs its own finite inventory, covariance/source freeze and operation limits.