Skip to main content

Module scoring

Module scoring 

Source
Expand description

Scorer, rubric, human scale, agreement and origin declarations of rustev.evaluator-config/3 (spec 041, 3.5 to 3.8).

These are declared data: a scorer’s descriptor names its identity, output and rules, so it enters the EvaluatorConfigId. A model scorer is a judge whose plan and backend identities are bound here too. The checks refuse every declaration that would mix a model metric with a deterministic one, read an ordinal scale as the wrong kind, or report an origin family before its evidence is read.

Modules§

code
Refusal codes of spec 041 under invalid_input (3.10.5).

Structs§

AgreementDecl
An agreement between two ordinal sides (3.8.4): each a declared scorer id or human:<scale>.
Declarations
The version 3 members of an evaluator configuration (3.1.3).
HumanScale
A human rating scale (3.6.4).
JudgeRef
A judge plan and the backend identities serving it (3.8.1).
RubricMetric
A rubric metric (3.6.3), reported as rubric/<name> or, for a model scorer, model/rubric/<name>.
ScorerDescriptor
A scorer’s descriptor (3.6.2).

Enums§

OriginFamily
An origin and evidence-quality family (3.5.1).
RubricAggregate
ScorerKind
ScorerOutput
What a score is (3.6.2).

Constants§

HUMAN_SIDE
An agreement side naming a human scale (3.8.4).
MAX_LEVELS
An ordinal scale holds 2 to 16 levels (3.6.2).
MAX_SCORERS
The most scorers one configuration declares (3.6.2).
MODEL_PREFIX
The reserved prefix of every model-derived metric (3.8.5).

Functions§

check_prefix
The model-prefix rule for a metric a gate or formula names (3.8.5): a rubric or agreement metric of a model scorer is named under model/, and a deterministic one never is.