Joint constraints sharply narrow acceleration
All six generated cases passed independent exact support checks. Their projected acceleration-change radii are 0.467–0.506 m/s², compared with the earlier full-box radius of 6.6685 m/s²: a reduction of about 92–93% in every group.
The improvement comes from retaining relationships between calibration terms. Their individual allowed ranges remain wide, but the combination that determines acceleration is much better constrained by training observations. No known simulated calibration or later observation constructed these bounds.
The six cases completed in 2.9 seconds, within the frozen 36 LP / 108 exact support allowance. All 51 coordinates and original physical bounds remain. Known anchors were opened only after saving the bounds, and all stayed feasible and contained. There were no aircraft reads, fits or native integrations.
This passes the acceleration engineering target, not the full prediction target. The next test must carry the verified change bounds into propagation while preserving the original expansion, shared coefficients and unqualified remainder allowances. It must reject certificates belonging to another source or training region. The 53-stretch findings remain unchanged.