Function compareMetricWithinTolerance

src/eval/discovery-harness.ts:297–321

(
  actualValue: number | null,
  comparatorValue: number | null,
  metric: DiscoveryMetricName,
  tolerancePercent: number
)

}

function compareMetricWithinTolerance(
  actualValue: number | null,
  comparatorValue: number | null,
  metric: DiscoveryMetricName,
  tolerancePercent: number
): DiscoveryMetricComparison {
  const lowerIsBetter =
    metric === 'averagePayloadBytes' ||
    metric === 'averageEstimatedTokens' ||
    metric === 'averageFirstRelevantHit';
  const multiplier = 1 + tolerancePercent / 100;
  const passes =
    actualValue !== null &&
    comparatorValue !== null &&
    (lowerIsBetter
      ? actualValue <= comparatorValue * multiplier
      : actualValue >= comparatorValue * (1 - tolerancePercent / 100));

  return {
    metric,
    comparatorValue,
    actualValue,
    passes
  };
}

function evaluateBaselineGate(
  summary: DiscoverySummaryType,