Corrected timing diagnostic

The first timing check completed its control but rejected both target endpoints before native evaluation because of one floating-point step beyond a parameter bound. This separately frozen diagnostic corrects that endpoint construction and reuses the completed control.

The finite plan allows 24 native predictions on two saved endpoints, with four numerical threads, no optimizer starts and unchanged scientific checks. Every rounding adjustment is recorded; larger bound violations remain errors. The current tally remains 47 globe preferences, zero flat and six unresolved.

The diagnostic completed both endpoints and all 24 native calls in about seven minutes, with no errors. The primary is about 23 ns before a common packet boundary; the repeat is about 72 ns after it. Full and half coupled corrections predict a decrease but increase the actual objective. See the preserved results. The next step is a separately qualified packet-aware timing solver. This completed diagnostic cannot restart or reuse its allowance.

The launcher is analysis/tests/ilvis0_timing_worker_v2.py, through analysis/tests/run_numerical.py --threads 4. Runtime metadata lives in data/ilvis0-timing-derivative-v2-20261016/. --prepare-only freezes the study, --detach launches it, and --package-only publishes completed evidence without predictions. The preceding completed study cannot restart or reuse its allowance.