refactor(scoring): reuse the shared measurement mean

src/kb-improvement/evaluation.ts

@@ -16,6 +16,7 @@ import {
   runRagKnowledgeImprovementLoop,
 } from '../rag-improvement-loop'
 import { readinessFor } from '../readiness-helpers'
+import { mean } from '../statistics'
 import { type ValidateKnowledgeResult, validateKnowledgeIndex } from '../validate'
 import type {
   KnowledgeImprovementCandidateRecord,
@@ -497,7 +498,7 @@ function defaultKnowledgeImprovementMetric(
   const blockingReadiness =
     blockingTotal === 0 ? 1 : Math.max(0, blockingTotal - blockingMissing) / blockingTotal
   const answerQuality = lifecycle?.answerQuality
-    ? average(Object.values(lifecycle.answerQuality.metrics).filter(Number.isFinite))
+    ? mean(Object.values(lifecycle.answerQuality.metrics))
     : 1
   const promotionDecision = lifecycle?.promotion ? (lifecycle.promotion.promoted ? 1 : 0) : 1
   const dimensions = {
@@ -515,7 +516,7 @@ function defaultKnowledgeImprovementMetric(
       : `${blockingMissing}/${blockingTotal} blocking knowledge requirements still missing`,
   ].filter((reason): reason is string => Boolean(reason))
   return {
-    score: average(Object.values(dimensions)),
+    score: mean(Object.values(dimensions)),
     passed: failedReasons.length === 0,
     dimensions,
     notes:
@@ -556,9 +557,3 @@ function applyLifecycleFailures(
 function normalizeMetric(metric: KnowledgeImprovementMetric): KnowledgeImprovementMetric {
   return improvementMetricSchema.parse(metric)
 }
-
-function average(values: readonly number[]): number {
-  const finite = values.filter(Number.isFinite)
-  if (finite.length === 0) return 0
-  return finite.reduce((sum, value) => sum + value, 0) / finite.length
-}

src/rag-eval/answer.ts

@@ -1,4 +1,5 @@
 import { groundClaimInText } from '../claim-grounding'
+import { mean } from '../statistics'
 import type {
   RagAnswerEvalArtifact,
   RagAnswerEvalScenario,
@@ -146,7 +147,7 @@ export function scoreAnswerCorrectness(
   const expectedScore =
     expected.length === 0
       ? scoreAnswerRelevance(artifact, scenario, abstained)
-      : average(expected.map((claim) => textSupportScore(claim, artifact.answer)))
+      : mean(expected.map((claim) => textSupportScore(claim, artifact.answer)))
   const forbiddenPenalty =
     forbidden.length === 0
       ? 0
@@ -210,11 +211,6 @@ export function neutralScore(condition: boolean): number {
   return condition ? 1 : 0
 }
 
-export function average(values: readonly number[]): number {
-  const finite = values.filter(Number.isFinite)
-  return finite.length === 0 ? 0 : finite.reduce((sum, value) => sum + value, 0) / finite.length
-}
-
 export function clamp01(value: number): number {
   if (!Number.isFinite(value)) return 0
   return Math.max(0, Math.min(1, value))

src/rag-eval/scoring.ts

@@ -1,7 +1,7 @@
 import type { JudgeConfig } from '@tangle-network/agent-eval/campaign'
 import type { RagGapFinding } from '../rag-improvement-loop'
+import { mean } from '../statistics'
 import {
-  average,
   claimSupport,
   clamp01,
   contextIsRelevant,
@@ -103,7 +103,7 @@ export function scoreRagAnswerArtifact(
       ? scenario.unanswerable
         ? 1
         : 0
-      : average(artifact.contexts.map((context) => contextRelevanceScore(context, scenario)))
+      : mean(artifact.contexts.map((context) => contextRelevanceScore(context, scenario)))
   const contextSufficiency = requiredContextCount === 0 ? contextRelevance : contextRecall
   const support = claims.map((claim) => claimSupport(claim.text, artifact.contexts, options))
   const supportedClaimCount = support.filter(Boolean).length
@@ -241,9 +241,9 @@ export function aggregateRagAnswerMetrics(
   }
   const out: Record<string, number> = {}
   for (const key of [...keys].sort()) {
-    out[key] = average(summaries.map((summary) => summary.metrics[key]))
+    out[key] = mean(summaries.map((summary) => summary.metrics[key]))
   }
-  out.composite = average(summaries.map((summary) => summary.composite))
+  out.composite = mean(summaries.map((summary) => summary.composite))
   return out
 }