Enum RailTailEvidence 

pub enum RailTailEvidence {
    AnalyticFaceProof {
        min_curvature: f64,
        curvature_margin: f64,
    },
    ProbedTail {
        noise_floor: f64,
        drift_band: f64,
    },
}

What established a rail coordinate’s tail law, and the standard it cleared (#2348 Inc 5 build-out).

The two routes are not two ways of measuring one quantity. They apply different standards to different evidence, and the floors they clear are not comparable: one is a threshold on a finite-difference estimate of ĉ, the other a backward-error threshold on the smallest eigenvalue of a matrix. Carrying both in a single noise_margin: f64 — documented as “the pencil-constant noise floor for a measured tail, or the eigen-backward-error margin of the analytic face form” — left every reader unable to tell which number they held, including this module’s own well-formedness guard.

Variants

AnalyticFaceProof

PROVEN at the face. The λ = ∞ limit was formed exactly and the first-order form C on the released subspace is positive definite, so the criterion strictly increases for every finite smoothing parameter on the face, and on every sub-face, at once.

A coordinate whose own penalty releases nothing once the REST of the face is at λ = ∞ is unidentified there, and the proof derives tail_constant = 0 for it exactly: V does not depend on its smoothing parameter at all. That zero is a result of the proof, not a missing measurement — which is precisely why the two routes cannot share one well-formedness rule.

Fields

min_curvature: f64

λ_min(C), the smallest curvature of the face’s first-order form.

curvature_margin: f64

The floor it cleared: q·ε·‖C‖·(1 + cond), the eigenvalue backward error amplified by the Z-block solve that formed C.

ProbedTail

MEASURED by probing back from the rail: the pencil constant ĉ = −e^{ρ}·∂V/∂ρ held across a finite-difference-clean window. The evidence is a window at finite λ, so a constant that is not strictly positive is the instrument reporting itself rather than a tail law.

Fields

noise_floor: f64

The finite-difference floor ĉ had to exceed to count as signal.

drift_band: f64

The relative drift band the window was held to.

Implementations

impl RailTailEvidence

pub fn admits(&self, tail_constant: f64) -> bool

Whether tail_constant is well formed FOR THIS ROUTE.

The routes disagree about zero, and both are right. A probed tail reads ĉ off a finite-difference window, where a non-positive constant is noise being reported as a law. A proven face derives ĉ from the limit itself, and derives exactly zero for a coordinate the rest of the face has already pinned; refusing that would refuse a face the proof covers. Such a coordinate needs another face coordinate to pin its directions, so it can only arise on a face of two or more — which is why a single-coordinate certificate never exercised the disagreement.

pub fn route(&self) -> &'static str

A short label naming the route, for logs and refusal summaries.

Trait Implementations

impl Clone for RailTailEvidence

fn clone(&self) -> RailTailEvidence

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for RailTailEvidence

impl Debug for RailTailEvidence

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for RailTailEvidence

fn deserialize<__D>( __deserializer: __D, ) -> Result<RailTailEvidence, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for RailTailEvidence

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

Equality operator ==.

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

Inequality operator !=.

impl Serialize for RailTailEvidence

fn serialize<__S>( &self, __serializer: __S, ) -> Result<<__S as Serializer>::Ok, <__S as Serializer>::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for RailTailEvidence