# NDC Nonlinearity Passport

- passport_version: `ndc-gp-native-passport/v2`
- model_hash: `9aaea132c94b2252ace311bd61f28604e307fe1b7dfd53427884f29c6c2485d7`
- GP lift feature hash: `a2968a67112596e52dbe968e11e34a917aa8213ee7dc76dcf242c6122bd5e3ba`
- active GP-native readouts: `428`
- active terms: `409`
- estimator: `robust linear estimator over GP-native readouts`

## Channel A

- local slot policy: `Hybrid { self_copies: 3, lag_count: 2, neighbour_count: 1 }`
- self-power max degree: `3`
- local slot count: `6`
- adaptive cascade configured: `true`
- adaptive max depth: `2`

- selected rotor policy: `Learned { channel_a_parameter: 0.173, channel_b_parameter: -0.311, channel_c_parameter: 0.127, seed: 872415233, objective_before: 1.0104950418348104, objective_after: 1.0094845467929756, optimizer_steps: 3 }`

## Channel B

- discovery mode: `grade_resolved_validation_aware`
- candidate edges: `5`
- exported edges: `5`
- exported edge features: `36`
- validation-aware acceptance: `true`
- accepted validation edges: `5`
- rejected candidate edges: `0`
- multi-lag edges: `5`
- group wormholes: `1`
- hierarchical dense pair scans: `5`
- hierarchical cluster pair scans: `1`
- hierarchical site pair scans: `4`

## Channel C

- full-rank mixing: `true`
- mode count: `4`
- connected global cascade count: `1`
- root hash: `05c5847a0854ff47a2d8f0eac7f0fea09413c43db45723bb8ba5d172a20e8fae`

## Validation and deployment

- unsupported extrapolation refusal: `true`
- runtime predictions match compiler predictions: `true`
- runtime hash: `9aaea132c94b2252ace311bd61f28604e307fe1b7dfd53427884f29c6c2485d7`
- export-safe runtime: `true`

## Identifiability, support, residuals, uncertainty

- identification status: `UnobservableDirection`
- active numerical rank: `48` / `409`
- validation unsupported rows: `0` / `16`
- support primary space: `raw_normalized_sensor`; feature-space diagnostic-only: `true`
- primary unsupported fraction: `0.000000`; GP-feature diagnostic unsupported fraction: `0.375000`
- residual whiteness score: `0.928624`; white enough: `true`
- empirical 90pct coverage: `Some(0.0)`; calibrated: `Some(false)`
- risk-coverage points: `10`; mean retained loss: `8.775489e-3`
- runtime parity passed: `true`; checked rows: `16`

## Order, memory and graph semantics

- dominant readout order: `Some(1)`
- nonlinear contribution fraction: `-0.000000`
- effective memory horizon: `0`
- graph semantics: `predictive_interaction_graph_not_causal_graph`

## Statement

observed frames are lifted through per-sensor GP, grade-labelled residual edges, and one connected global GP cascade; coefficients only are estimated Closed-loop free-run stability polish selected persistence blend 0.500000 on heldout validation (free-run NRMSE 9.160289e-1 -> 7.673726e-1).