Enum EvalArm 

pub enum EvalArm {
    Raw,
    Gglib,
    RawReplicate,
    Control,
}

Which arm a task ran under.

Variants

Raw

Pipeline bypassed — bare llama-server behaviour.

Gglib

The full gglib request/response pipeline.

RawReplicate

A/A control. The raw arm again, on a disjoint seed set.

Nothing about the request differs from Self::Raw — same bypass, same tasks, same machine, same loaded model — so any gap between the two is the eval measuring itself. That gap is the floor a raw-versus-gglib delta has to clear, and without it a small delta has two readings that the report cannot separate: the pipeline helped a little, or five seeds is not enough seeds.

It answers a strictly different question from Self::Control. The control establishes that a large change registers; this establishes how large a change has to be before it registers as anything but drift. A run carrying only the control can say “the apparatus works” about an effect it has no ability to resolve.

Control

Positive control. The gglib pipeline with the temperature forced to CONTROL_TEMPERATURE, which should sample visibly worse.

It exists to answer a question the other two arms cannot: can this apparatus detect a sampling change at all? A raw-versus-gglib run showing no difference has two explanations — the pipeline does not help, or the harness cannot see — and nothing in that run distinguishes them.

This arm is a deliberate, large, known-bad change. If it does not score below the gglib arm, the apparatus failed to move under a difference that should be impossible to miss, and no other number in the report can be believed. That is the same discipline ADR 0004 applies to its instruments: a comparison in which nothing could have varied, reporting that nothing varied, is not evidence.

Trait Implementations

impl Clone for EvalArm

fn clone(&self) -> EvalArm

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for EvalArm

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for EvalArm

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for EvalArm

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

Formats the value using the given formatter.

impl PartialEq for EvalArm

fn eq(&self, other: &EvalArm) -> 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 EvalArm

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 EvalArm

impl Eq for EvalArm

impl StructuralPartialEq for EvalArm