Enum EffectVerdict 

pub enum EffectVerdict {
    ExceedsNoise {
        effect: f64,
        noise: f64,
        pairs: usize,
    },
    WithinNoise {
        effect: f64,
        noise: f64,
        pairs: usize,
    },
}

What the A/A arm says about the size of the measured effect.

Deliberately not a p-value or a confidence interval: with one replicate there is nothing to compute either from, and rendering a statistic that the design cannot support would be worse than rendering none. The two arms of this enum are the honest resolution of a one-pair comparison — the effect is clearly bigger than the drift, or it is not clearly bigger than the drift.

Variants

ExceedsNoise

|gglib − raw| is at least EFFECT_NOISE_RATIO times the A/A gap.

Fields

effect: f64

gglib − raw, signed: a negative effect that clears the noise floor is still a finding, just not the hoped-for one.

noise: f64

The mean pairwise drift among identical raw runs.

pairs: usize

Pairwise gaps behind noise — the estimate’s degrees of freedom. A verdict over one pair and one over six are not the same strength of claim, and rendering them identically invites exactly the misreading the A/A arm exists to prevent.

WithinNoise

The effect is not clearly larger than the drift between two runs of the same arm. It is not thereby absent — it is unresolved at this seed count, and the fix is more seeds rather than a different conclusion.

Fields

effect: f64

gglib − raw, signed.

noise: f64

The mean pairwise drift among identical raw runs.

pairs: usize

Pairwise gaps behind noise.

Implementations

impl EffectVerdict

pub fn ratio(&self) -> Option<f64>

|effect| ÷ noise, or None when the two arms landed on exactly the same composite and the ratio would divide by zero.

A zero denominator is not a licence to report an infinite ratio: two identical scores on a suite this small means the drift went unresolved, not that there is none.

pub const fn effect(&self) -> f64

The signed gglib − raw difference this verdict is about.

pub const fn pairs(&self) -> usize

How many pairwise drift gaps stand behind Self::noise.

pub const fn noise(&self) -> f64

The A/A drift this verdict measured the effect against.

pub const fn exceeds_noise(&self) -> bool

Whether the effect cleared the drift by EFFECT_NOISE_RATIO.

Trait Implementations

impl Clone for EffectVerdict

fn clone(&self) -> EffectVerdict

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for EffectVerdict

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for EffectVerdict

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for EffectVerdict

fn eq(&self, other: &EffectVerdict) -> 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 Serialize for EffectVerdict

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

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl Copy for EffectVerdict

impl StructuralPartialEq for EffectVerdict