CALIBER
Quickstart
Use CALIBER

Evaluation and test sets

The practical trust loop: define evidence, run scorecards, inspect judges, and decide what still needs review before live change.

System userEvaluatorDeveloperHow ToGA
PrerequisitesA CALIBER deployment with evaluation access
Reviewed 2026-08-10 · current main branch docs contract

Use this page when you need the practical trust loop: assemble evidence, run scorecards, inspect judges and per-example results, and decide what still needs review before live change.

At a glance

TaskStart hereDeep reference
define the evidence basecreate or inspect a test setTest sets architecture
score a candidaterun evaluation against that evidenceEvaluation architecture
inspect judge behaviorunderstand the scorer and custom judge surfacesEvaluation architecture
connect runtime QA to engineering evaluationcompare both trust loops explicitlyQA plan

1. What this path is for

CALIBER separates evidence from rollout. Evaluation tells you how something performed against a known set of cases. It does not release or apply a change.

2. Common tasks

You want to...Read this next
curate representative examplesTest sets architecture
run a scorecard or judge-backed evaluationEvaluation architecture
connect evaluation to calibrationCalibration
connect evaluation to runtime review and releaseReview and release flows

3. Common failure modes

SymptomFirst thing to check
Scores are high but production still looks weakthe evidence set may not represent the live workload
A judge looks inconsistentinspect custom-judge setup and agreement expectations
A good score did not change productionevaluation is evidence, not apply/release

CALIBER : Contextual Adaptive Lifecycle for Intelligent Build, Evaluation, and Refinement — this page is generated from the authoritative Markdown sources in docs/ and the repository-level ARCHITECTURE.md.