# AP1-R2c p=42 bounded-method diagnostic checkpoint

Timestamp: `2026-09-02T08:45:54Z`

The authoritative continuation parent remains the repeatedly validated
41-node PASS. No 42-node vector from this diagnostic was stored or used as a
seed.

An endpoint-preserving degree-2/3 B-spline extrapolation with `beta=3`, a
partial `rho_chi` projection with `alpha=1.75e-5`, and three tiny frozen
Lobatto defect dampings reduced the 42-node full DAE norm from
`4.4107284402342004e-4` to `2.981570089038665e-4`. Three independent full
checks returned the same value.

The DAE threshold remains `1e-4`, so the candidate is NONPASS. All other
unchanged gates pass on the diagnostic candidate:

- Ward: `4.6404608962232175e-7` (threshold `2e-6`)
- source closure: `2.1148579406445302e-5` (threshold `1e-4`)
- Friedmann residual: `1.512764475823987e-16`
- Wronskian relative error: `4.4e-50`
- minimum `H`: `8.312598005592009e-8`
- minimum `P_X`: `1.8132335484729594`
- minimum `K`: `3.016188545394498`

A 68-candidate two-parent p-secant bank, an 18-candidate blockwise spline
bank, and a five-coordinate linearized minimax proposal were worse and are
excluded for this checkpoint. Equations, physics, and gates were unchanged.
G22 remains ORANGE and production curves remain locked.