Documentation
¶
Overview ¶
Package scoring implements the consistency scorers used to detect whether an agent's arguments contradict each other under cross-examination.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type LLMScorer ¶
type LLMScorer struct {
Model model.BaseModel[*schema.Message]
Logger *slog.Logger
// NLIProbe builds the evaluation prompt. When nil, the built-in template
// is used.
NLIProbe func(question, r1, r2 string) string
}
LLMScorer implements core.ConsistencyScorer by having a model run a logical entailment / contradiction check between two arguments.
type LexicalScorer ¶
type LexicalScorer struct{}
LexicalScorer implements core.ConsistencyScorer deterministically with zero extra model calls: it returns the Jaccard similarity of the two arguments' word sets. It is best used as a cheap approximation when an LLM evaluator is not available, e.g. while debugging the pipeline.
Source Files
Click to show internal directories.
Click to hide internal directories.