class RubyLLM::Evaluation::Case

One dataset example, with application inputs and optional reference evidence.

Attributes

expected_output [R]

Returns the reference answer, or nil when none was supplied.

inputs [R]

Returns the value passed to Evaluation#perform.

metadata [R]

Returns application-defined reference evidence and category labels.

name [R]

Returns the stable case name used to compare runs.

Public Class Methods

new (name:, inputs:, metadata: {}, **reference)

Source

# File lib/ruby_llm/evaluation/case.rb, line 19
def initialize(name:, inputs:, metadata: {}, **reference)
  unless (reference.keys - [:expected_output]).empty?
    raise ArgumentError,
          "Unknown case fields: #{reference.keys - [:expected_output]}"
  end
  raise ArgumentError, 'A case name cannot be empty' if name.to_s.empty?
  raise ArgumentError, 'Case metadata must be a Hash' unless metadata.is_a?(Hash)

  @name = name.to_s.dup.freeze
  @inputs = Judge::Data.copy(inputs)
  @metadata = Judge::Data.copy(metadata)
  @expected_output = Judge::Data.copy(reference[:expected_output])
  @reference_supplied = reference.key?(:expected_output)
  freeze
end

Creates a case. Inputs and reference evidence must be JSON-compatible.

Public Instance Methods

expected_output? ()

Source

# File lib/ruby_llm/evaluation/case.rb, line 36
def expected_output?
  @reference_supplied
end

Returns whether a reference was supplied, including an explicit nil.

to_h ()

Source

# File lib/ruby_llm/evaluation/case.rb, line 41
def to_h
  data = { name:, inputs:, metadata: }
  data[:expected_output] = expected_output if expected_output?
  data
end

Returns the portable dataset representation of this case.