Testing the time-alignment derivative

The current recovery total is 47 globe preferences, zero flat preferences and six unresolved stretches. The remaining failed fits stop early. This focused check asks whether the fitted timing direction predicts actual position changes across nearby IMU packet intervals.

The finite plan fixes three saved endpoints and at most 34 native predictions, with no optimization. It preserves all scientific results and limits. Software controls check the derivative comparison, packet convention, feasibility and exact call counts before launch.

This run has finished: the control completed ten native calls, while both target identities failed preflight before native evaluation. Their saved coupled corrections exceeded a normalized bound by one floating-point step. The errors and results remain preserved. A separate corrected freeze reuses the completed control and tests the two previously unevaluated endpoints. This completed run cannot restart or spend its unused allowance.

The worker is analysis/tests/ilvis0_timing_worker.py, invoked through analysis/tests/run_numerical.py --threads 4. Runtime state and source snapshots are kept in data/ilvis0-timing-derivative-20261016/. --prepare-only freezes inputs; --detach starts the finite diagnostic; --package-only publishes completed results without numerical predictions. No completed identity may be retried.