Struct SteerPlan 

pub struct SteerPlan {
    pub atom: usize,
    pub atom_name: String,
    pub t_from: Vec<f64>,
    pub t_to: Vec<f64>,
    pub amplitude: f64,
    pub measured_row: usize,
    pub delta: ArrayBase<OwnedRepr<f64>, Dim<[usize; 1]>>,
    pub predicted_nats: Option<f64>,
    pub validity_radius: Option<f64>,
    pub off_manifold_norm: f64,
    pub metric_provenance: MetricProvenance,
}

The actionable output of a steering query over one atom.

Fields

atom: usize

Which atom was steered (index into SaeManifoldTerm::atoms).

atom_name: String

The atom’s name (mirrors crate::manifold::SaeManifoldAtom::name).

t_from: Vec<f64>

The source latent coordinate t_from (length = atom’s latent_dim).

t_to: Vec<f64>

The target latent coordinate t_to (length = atom’s latent_dim).

amplitude: f64

The amplitude a the on-manifold move was scaled by (the atom’s mean active assignment mass; 1.0 if the atom is active on no row).

measured_row: usize

The row whose per-row output-Fisher metric the dose was measured through (the atom’s most-active row; 0 if active nowhere).

delta: ArrayBase<OwnedRepr<f64>, Dim<[usize; 1]>>

The activation-space delta: δ = a · (g_k(t_to) − g_k(t_from)), a length-p vector in the reconstruction/output space — the actual move to add to a hidden state.

predicted_nats: Option<f64>

DOSIMETRY: predicted output effect of the move in nats of KL, integrated along the decoder curve through the output-Fisher metric. None when the metric carries no behavioral information (Euclidean provenance) — the dose is not available, not zero.

validity_radius: Option<f64>

VALIDITY RADIUS: the latent step size (Euclidean norm of the move from t_from) at which the path-integrated dose first diverges from the straight endpoint quadratic form by more than [VALIDITY_DIVERGENCE_FRACTION]. Equals the full move length when the linearization is trusted all the way to t_to. None under a no-behavior metric (there is no dose to validate).

off_manifold_norm: f64

OFF-MANIFOLD GUARD: the norm of δ’s component outside the span of the atom’s local decoder tangents ∂g_k/∂t at t_from. ≈ 0 by construction (the move is a chord of the curve); a large value flags a move that left the learned surface.

metric_provenance: MetricProvenance

The provenance of the metric the dose was read through, echoed so a consumer can certify why predicted_nats is None when it is.

Trait Implementations

impl Clone for SteerPlan

fn clone(&self) -> SteerPlan

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for SteerPlan

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl PartialEq for SteerPlan

fn eq(&self, other: &SteerPlan) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for SteerPlan