Implementation:Arize ai Phoenix ExactMatch

Overview

exact_match is a code-based evaluator function in the arize-phoenix-evals package that performs strict string equality comparison between an output and an expected value. It is decorated with @create_evaluator to conform to the standard Evaluator interface, returning a Score of 1.0 for an exact match and 0.0 otherwise.

Description

The exact_match evaluator is the simplest metric in the phoenix-evals metrics module. It performs a direct equality check (output == expected) with no text normalization applied -- no case folding, whitespace trimming, or Unicode normalization. The comparison is case-sensitive and whitespace-sensitive.

The function is wrapped with the @create_evaluator decorator, which:

• Registers it as an evaluator with the name "exact_match" and kind "code".
• Enables it to accept a dictionary eval_input and optional field_mapping parameter.
• Makes it compatible with the standard evaluator interface used by evaluate_dataframe and other evaluation utilities.

Usage

from phoenix.evals.metrics import exact_match

Code Reference

Property	Value
Source File	packages/phoenix-evals/src/phoenix/evals/metrics/exact_match.py
Module	phoenix.evals.metrics.exact_match
Function	exact_match(output: str, expected: str) -> Score
Decorator	@create_evaluator(name="exact_match", kind="code")
Lines	~37
Kind	"code"
Direction	"maximize"
Domain	LLM Evaluation, Metrics

Implementation

The core logic is minimal:

@create_evaluator(name="exact_match", kind="code")
def exact_match(output: str, expected: str) -> Score:
    correct = output == expected
    return Score(score=float(correct))

I/O Contract

Input

Field	Type	Required	Description
output	str	Yes	The output text to evaluate.
expected	str	Yes	The expected (ground-truth) text.

Note: Input field names can be remapped using the field_mapping parameter if the eval input dictionary uses different key names.

Output

Returns a list containing one Score object:

Field	Description
name	"exact_match"
score	1.0 if output == expected, 0.0 otherwise.
label	True or False.
explanation	None
kind	"code"
direction	"maximize"

Usage Examples

Direct Usage (No Field Mapping)

from phoenix.evals.metrics import exact_match

eval_input = {"output": "no", "expected": "yes"}
scores = exact_match(eval_input)
print(scores)
# [Score(score=0.0, name='exact_match', label=False, explanation=None,
#  direction='maximize', kind='code', metadata={})]

Usage with Field Mapping

from phoenix.evals.metrics import exact_match

eval_input = {"prediction": "yes", "gold": "yes"}
field_mapping = {"output": "prediction", "expected": "gold"}
scores = exact_match(eval_input, field_mapping=field_mapping)
print(scores)
# [Score(score=1.0, name='exact_match', label=True, explanation=None,
#  direction='maximize', kind='code', metadata={})]

Case Sensitivity

from phoenix.evals.metrics import exact_match

# Case-sensitive: "Paris" != "paris"
eval_input = {"output": "paris", "expected": "Paris"}
scores = exact_match(eval_input)
# score=0.0 -- no normalization is performed

Related Pages

• Arize_ai_Phoenix_MatchesRegex -- Regex-based code evaluator for pattern matching.
• Arize_ai_Phoenix_PrecisionRecallFScore -- Code-based precision/recall evaluator for classification metrics.
• Arize_ai_Phoenix_CorrectnessEvaluator -- LLM-based correctness evaluation for semantic comparison.
• Arize_ai_Phoenix_Evals_Public_API -- The top-level phoenix.evals public API surface.