Fixed six-target hybrid assessment

This separate freeze covers exactly the six unresolved stretches from the completed nine-target assessment that have not been tested with the qualified hybrid solver. Selection uses saved completion status, not favorable model costs or gyro values.

The finite allowance is 24 starts / 4,528 native sweeps, with four total numerical threads, no retries and no allowance expansion. Eleven failed primary model/case identities are each fitted from their latest saved primary endpoint and repeated from that same endpoint with the fixed inward 0.01 normalized perturbation. Every primary and repeat permits 200 fitting calls plus two diagnostics. Twenty-five passing companion primaries are reused with their original artifacts and hashes. Two completed controls permit forty fitting calls plus two diagnostics each.

Some stretches contain multiple failed model/case identities. Every newly fitted primary must pass its own repeat check; a later passing pair cannot overwrite an earlier failure. Per-pair results and an all-pairs-per-stretch gate are preserved. All six primaries, every required repeat and both fresh controls must pass before any stretch changes disposition. A passing repeat never replaces a failed primary.

The source targets are 0e73879f98c3b701fd7c42df, 370cec66878c4c1cf84b4ceb, 50f24131843071ae9312c082, b4e44ddc0f130ecd45674c78, bb8061a43b4f14b6262ceda1 and da346f70a6661ee42cecea19. The seventh unresolved stretch is excluded and has a separate saved-tangent audit.

The solver, physical models, two processing cases, original measurements, selected intervals, clock, covariance, calibration bounds and final checks stay unchanged. Known-answer software qualification and the completed matched trial are preserved; partial success in that trial does not establish that every target will resolve.

Applying the prior per-identity runtimes to every new evaluation cap gives an estimated ceiling of 28.8 hours. Actual completion may be earlier; the binding limits are starts and native evaluations. Freeze sources, inputs, environment and the task list before launch. Preserve charged interruptions, exclusive locks, start/sweep journals and atomic results. Completed work cannot resume, and unused allowances do not transfer. Publication performs no fits or automatic rollout.