Skip to main content
MeanScore computes the arithmetic mean of a named score field over every EvalRow in a run. It is the primary quality signal for most benchmarks. 
from context_bench.metrics import MeanScore

​
Constructor parameters

​
score_field
string
default:"score"
Name of the score key to average. Must match a key emitted by an evaluator’s score() method — for example "f1", "exact_match", or "mc_accuracy".

​
Return value

compute() returns a dict[str, float] with the following key:
​
mean_score
float
Mean value of score_field across all examples. Range [0.0, 1.0] for standard evaluators. Returns 0.0 when the row list is empty.

​
Usage

from context_bench import evaluate
from context_bench.evaluators import AnswerQuality
from context_bench.metrics import MeanScore

result = evaluate(
    systems=[my_system],
    dataset=my_dataset,
    evaluators=[AnswerQuality()],
    metrics=[MeanScore(score_field="f1")],
)
print(result.summary["my-system"]["mean_score"])  # e.g. 0.742

MeanScore(score_field="f1")

​
When it is enabled

MeanScore is included in every CLI run by default. The CLI flag --score-field controls which field it reads (default: f1).
score_field must match a key produced by an evaluator that is also registered for the run. If the field is absent on a row, that row contributes 0.0 to the mean.

Build docs developers (and LLMs) love

Get started for freeTalk to us