Common-acceleration duration growth

Repeat the completed duration diagnostic with the qualified acceleration-aware construction. Keep the same 48 identities: all three models, both processing cases and eight original yaw sectors. Each uses 400/800/1,600-packet prefixes at 1.8/3.6/7.2 seconds, the original 27 physical bounds and all 15 common error coordinates. Extended generated inputs must be byte-identical to the earlier duration screen; their first 400 packets must also match the completed prediction comparison. No observed records or protected ESA outcomes are read.

For each prefix, preserve one native center with all 27 physical tangent columns, one bound and three native parameter probes (both corners and the same fixed interior point). Subtract the actual known-control 15-vector only when checking the factored remainder. Keep native/bound agreement within 1 micrometre, normalized common vectors within 1+1e-12 and probe residuals within the unshared allowance plus 10 micrometres. Retain all failures and unsupported domains.

Report the unshared contribution, position/lever parts, bindings and common radii at every horizon. These are conservative calculation terms, not complete training-constrained prediction widths, sensor specifications or calibrated intervals. The completed 48-sector acceleration-aware comparison is a prerequisite; passing this screen cannot replace a longer full prediction test.

Limits: 576 native evaluations / 537,600 native packet visits, 144 bound evaluations / 134,400 reserved bound steps, one worker with four numerical threads and 3,600 cooperative seconds. A bound may use at most 120 cooperative seconds; the longer CPU allowance changes no mathematical term, physical interval or comparison tolerance. Charge and reserve before operations. No retries, expansions, process signals, observed fits or completed-budget reuse.

Numerical timings from the preceding native/full-cover studies motivate this separate finite allowance. Freeze sources, deterministic inputs and environment before launch. Whole-program floating-point enclosure, actual receiver-error adequacy and independent confirmation remain separate requirements.

Complete short comparison · Previous duration screen · Native construction · Roadmap.