@bolt-foundry/gambit 0.8.6-rc.1 → 0.8.6-rc.3

package/esm/simulator-ui/src/verify_metrics.d.ts

@@ -1,50 +0,0 @@
-export type VerifyVerdict = "PASS" | "WARN" | "FAIL";
-export type VerifyCalibrationRun = {
-    id: string;
-    status: "running" | "completed" | "error";
-    runAt?: string;
-    result?: unknown;
-};
-export type VerifyOutlier = {
-    key: string;
-    label: string;
-    sampleSize: number;
-    agreementRate: number | null;
-    scoreDelta: number | null;
-    passFlip: boolean;
-    instability: boolean;
-    minScore: number | null;
-    maxScore: number | null;
-    minRunId?: string;
-    maxRunId?: string;
-    turnIndex?: number;
-    messageRefId?: string;
-};
-export type VerifyConsistencyReport = {
-    sampleSize: number;
-    comparableExampleCount: number;
-    agreementRate: number | null;
-    scoreSpreadMin: number | null;
-    scoreSpreadMedian: number | null;
-    scoreSpreadMax: number | null;
-    instabilityCount: number;
-    verdict: VerifyVerdict;
-    verdictReason: string;
-    outliers: Array<VerifyOutlier>;
-};
-export declare const VERIFY_CONSISTENCY_THRESHOLDS: {
-    readonly minSampleSize: 6;
-    readonly instabilityScoreDelta: 1.5;
-    readonly pass: {
-        readonly agreementMin: 0.9;
-        readonly maxSpread: 1;
-        readonly maxInstabilityCount: 0;
-    };
-    readonly warn: {
-        readonly agreementMin: 0.75;
-        readonly maxSpread: 2;
-        readonly maxInstabilityCount: 2;
-    };
-};
-export declare function buildVerifyConsistencyReport(runs: Array<VerifyCalibrationRun>): VerifyConsistencyReport;
-//# sourceMappingURL=verify_metrics.d.ts.map

package/esm/simulator-ui/src/verify_metrics.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"verify_metrics.d.ts","sourceRoot":"","sources":["../../../src/simulator-ui/src/verify_metrics.ts"],"names":[],"mappings":"AAAA,MAAM,MAAM,aAAa,GAAG,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;AAErD,MAAM,MAAM,oBAAoB,GAAG;IACjC,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,SAAS,GAAG,WAAW,GAAG,OAAO,CAAC;IAC1C,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,OAAO,CAAC;CAClB,CAAC;AAkBF,MAAM,MAAM,aAAa,GAAG;IAC1B,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,QAAQ,EAAE,OAAO,CAAC;IAClB,WAAW,EAAE,OAAO,CAAC;IACrB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB,CAAC;AAEF,MAAM,MAAM,uBAAuB,GAAG;IACpC,UAAU,EAAE,MAAM,CAAC;IACnB,sBAAsB,EAAE,MAAM,CAAC;IAC/B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,gBAAgB,EAAE,MAAM,CAAC;IACzB,OAAO,EAAE,aAAa,CAAC;IACvB,aAAa,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,KAAK,CAAC,aAAa,CAAC,CAAC;CAChC,CAAC;AAEF,eAAO,MAAM,6BAA6B;;;;;;;;;;;;;CAahC,CAAC;AA0JX,wBAAgB,4BAA4B,CAC1C,IAAI,EAAE,KAAK,CAAC,oBAAoB,CAAC,GAChC,uBAAuB,CA0HzB"}

package/esm/simulator-ui/src/verify_metrics.js

@@ -1,260 +0,0 @@
-export const VERIFY_CONSISTENCY_THRESHOLDS = {
-    minSampleSize: 6,
-    instabilityScoreDelta: 1.5,
-    pass: {
-        agreementMin: 0.9,
-        maxSpread: 1,
-        maxInstabilityCount: 0,
-    },
-    warn: {
-        agreementMin: 0.75,
-        maxSpread: 2,
-        maxInstabilityCount: 2,
-    },
-};
-const pickPayload = (result) => {
-    if (!result || typeof result !== "object")
-        return {};
-    const record = result;
-    if (record.payload &&
-        typeof record.payload === "object" &&
-        !Array.isArray(record.payload)) {
-        return record.payload;
-    }
-    return record;
-};
-const extractScoreReasonPass = (result) => {
-    const payload = pickPayload(result);
-    const score = typeof payload.score === "number" && Number.isFinite(payload.score)
-        ? payload.score
-        : undefined;
-    const reason = typeof payload.reason === "string"
-        ? payload.reason
-        : undefined;
-    if (typeof payload.pass === "boolean") {
-        return { score, reason, pass: payload.pass };
-    }
-    if (typeof payload.passed === "boolean") {
-        return { score, reason, pass: payload.passed };
-    }
-    if (typeof payload.verdict === "string") {
-        const verdict = payload.verdict.trim().toLowerCase();
-        if (verdict === "pass")
-            return { score, reason, pass: true };
-        if (verdict === "fail")
-            return { score, reason, pass: false };
-    }
-    if (typeof score === "number") {
-        return { score, reason, pass: score >= 0 };
-    }
-    return { score, reason };
-};
-const flattenRunExamples = (run) => {
-    if (!run.result || typeof run.result !== "object")
-        return [];
-    const record = run.result;
-    if (record.mode === "turns" && Array.isArray(record.turns)) {
-        const buckets = [];
-        record.turns.forEach((turn, fallbackIndex) => {
-            if (!turn || typeof turn !== "object")
-                return;
-            const turnRecord = turn;
-            const index = typeof turnRecord.index === "number" &&
-                Number.isFinite(turnRecord.index)
-                ? Math.max(0, Math.round(turnRecord.index))
-                : fallbackIndex;
-            const messageRefId = typeof turnRecord.messageRefId === "string" &&
-                turnRecord.messageRefId.trim().length > 0
-                ? turnRecord.messageRefId
-                : undefined;
-            const key = messageRefId ? `ref:${messageRefId}` : `turn:${index}`;
-            const label = `Assistant turn ${fallbackIndex + 1}`;
-            const parsed = extractScoreReasonPass(turnRecord.result);
-            buckets.push({
-                key,
-                label,
-                points: [{
-                        runId: run.id,
-                        runAt: run.runAt,
-                        score: parsed.score,
-                        pass: parsed.pass,
-                        reason: parsed.reason,
-                        turnIndex: index,
-                        messageRefId,
-                    }],
-            });
-        });
-        return buckets;
-    }
-    const parsed = extractScoreReasonPass(run.result);
-    return [{
-            key: "conversation",
-            label: "Conversation score",
-            points: [{
-                    runId: run.id,
-                    runAt: run.runAt,
-                    score: parsed.score,
-                    pass: parsed.pass,
-                    reason: parsed.reason,
-                }],
-        }];
-};
-const median = (values) => {
-    if (!values.length)
-        return null;
-    const sorted = [...values].sort((a, b) => a - b);
-    const mid = Math.floor(sorted.length / 2);
-    if (sorted.length % 2 === 1)
-        return sorted[mid];
-    return (sorted[mid - 1] + sorted[mid]) / 2;
-};
-const round2 = (value) => Math.round(value * 100) / 100;
-const resolveVerdict = (input) => {
-    const t = VERIFY_CONSISTENCY_THRESHOLDS;
-    if (input.sampleSize < t.minSampleSize) {
-        return {
-            verdict: "WARN",
-            reason: `Need at least ${t.minSampleSize} samples before issuing a firm verdict.`,
-        };
-    }
-    if (input.agreementRate === null) {
-        return {
-            verdict: "WARN",
-            reason: "No comparable pass/fail evidence was found in the sampled runs.",
-        };
-    }
-    const spreadMax = input.spreadMax ?? 0;
-    if (input.agreementRate >= t.pass.agreementMin &&
-        spreadMax <= t.pass.maxSpread &&
-        input.instabilityCount <= t.pass.maxInstabilityCount) {
-        return {
-            verdict: "PASS",
-            reason: "Agreement, spread, and instability all meet PASS thresholds.",
-        };
-    }
-    if (input.agreementRate >= t.warn.agreementMin &&
-        spreadMax <= t.warn.maxSpread &&
-        input.instabilityCount <= t.warn.maxInstabilityCount) {
-        return {
-            verdict: "WARN",
-            reason: "Some variation was detected, but results remain within WARN thresholds.",
-        };
-    }
-    return {
-        verdict: "FAIL",
-        reason: "Agreement/spread instability exceeds WARN thresholds.",
-    };
-};
-export function buildVerifyConsistencyReport(runs) {
-    const completedRuns = runs.filter((run) => run.status === "completed");
-    const sampleSize = completedRuns.length;
-    const bucketsByKey = new Map();
-    completedRuns.forEach((run) => {
-        flattenRunExamples(run).forEach((entry) => {
-            const existing = bucketsByKey.get(entry.key);
-            if (!existing) {
-                bucketsByKey.set(entry.key, {
-                    key: entry.key,
-                    label: entry.label,
-                    points: [...entry.points],
-                });
-                return;
-            }
-            existing.points.push(...entry.points);
-        });
-    });
-    const outliers = [];
-    let agreementVotes = 0;
-    let agreementTotal = 0;
-    const scoreDeltas = [];
-    bucketsByKey.forEach((bucket) => {
-        const scores = bucket.points
-            .map((point) => point.score)
-            .filter((score) => typeof score === "number" && Number.isFinite(score));
-        const minScore = scores.length ? Math.min(...scores) : null;
-        const maxScore = scores.length ? Math.max(...scores) : null;
-        const scoreDelta = minScore !== null && maxScore !== null
-            ? round2(maxScore - minScore)
-            : null;
-        const passVotes = bucket.points
-            .map((point) => point.pass)
-            .filter((pass) => typeof pass === "boolean");
-        const passCount = passVotes.filter((value) => value).length;
-        const failCount = passVotes.length - passCount;
-        const agreementRate = passVotes.length > 0
-            ? round2(Math.max(passCount, failCount) / passVotes.length)
-            : null;
-        if (passVotes.length > 0) {
-            agreementVotes += Math.max(passCount, failCount);
-            agreementTotal += passVotes.length;
-        }
-        if (scoreDelta !== null) {
-            scoreDeltas.push(scoreDelta);
-        }
-        const passFlip = passCount > 0 && failCount > 0;
-        const instability = passFlip ||
-            (scoreDelta !== null &&
-                scoreDelta > VERIFY_CONSISTENCY_THRESHOLDS.instabilityScoreDelta);
-        const minPoint = minScore === null
-            ? undefined
-            : bucket.points.find((point) => point.score === minScore);
-        const maxPoint = maxScore === null
-            ? undefined
-            : bucket.points.find((point) => point.score === maxScore);
-        outliers.push({
-            key: bucket.key,
-            label: bucket.label,
-            sampleSize: bucket.points.length,
-            agreementRate,
-            scoreDelta,
-            passFlip,
-            instability,
-            minScore,
-            maxScore,
-            minRunId: minPoint?.runId,
-            maxRunId: maxPoint?.runId,
-            turnIndex: maxPoint?.turnIndex ?? minPoint?.turnIndex,
-            messageRefId: maxPoint?.messageRefId ?? minPoint?.messageRefId,
-        });
-    });
-    outliers.sort((a, b) => {
-        if (a.instability !== b.instability)
-            return a.instability ? -1 : 1;
-        if (a.passFlip !== b.passFlip)
-            return a.passFlip ? -1 : 1;
-        const aDelta = a.scoreDelta ?? -1;
-        const bDelta = b.scoreDelta ?? -1;
-        if (aDelta !== bDelta)
-            return bDelta - aDelta;
-        if (a.sampleSize !== b.sampleSize)
-            return b.sampleSize - a.sampleSize;
-        return a.label.localeCompare(b.label);
-    });
-    const agreementRate = agreementTotal > 0
-        ? round2(agreementVotes / agreementTotal)
-        : null;
-    const scoreSpreadMin = scoreDeltas.length ? Math.min(...scoreDeltas) : null;
-    const scoreSpreadMax = scoreDeltas.length ? Math.max(...scoreDeltas) : null;
-    const scoreSpreadMedian = median(scoreDeltas);
-    const instabilityCount = outliers.filter((entry) => entry.instability).length;
-    const verdict = resolveVerdict({
-        sampleSize,
-        agreementRate,
-        spreadMax: scoreSpreadMax,
-        instabilityCount,
-    });
-    return {
-        sampleSize,
-        comparableExampleCount: outliers.length,
-        agreementRate,
-        scoreSpreadMin: scoreSpreadMin === null ? null : round2(scoreSpreadMin),
-        scoreSpreadMedian: scoreSpreadMedian === null
-            ? null
-            : round2(scoreSpreadMedian),
-        scoreSpreadMax: scoreSpreadMax === null ? null : round2(scoreSpreadMax),
-        instabilityCount,
-        verdict: verdict.verdict,
-        verdictReason: verdict.reason,
-        outliers,
-    };
-}

package/script/simulator-ui/src/verify_metrics.d.ts

@@ -1,50 +0,0 @@
-export type VerifyVerdict = "PASS" | "WARN" | "FAIL";
-export type VerifyCalibrationRun = {
-    id: string;
-    status: "running" | "completed" | "error";
-    runAt?: string;
-    result?: unknown;
-};
-export type VerifyOutlier = {
-    key: string;
-    label: string;
-    sampleSize: number;
-    agreementRate: number | null;
-    scoreDelta: number | null;
-    passFlip: boolean;
-    instability: boolean;
-    minScore: number | null;
-    maxScore: number | null;
-    minRunId?: string;
-    maxRunId?: string;
-    turnIndex?: number;
-    messageRefId?: string;
-};
-export type VerifyConsistencyReport = {
-    sampleSize: number;
-    comparableExampleCount: number;
-    agreementRate: number | null;
-    scoreSpreadMin: number | null;
-    scoreSpreadMedian: number | null;
-    scoreSpreadMax: number | null;
-    instabilityCount: number;
-    verdict: VerifyVerdict;
-    verdictReason: string;
-    outliers: Array<VerifyOutlier>;
-};
-export declare const VERIFY_CONSISTENCY_THRESHOLDS: {
-    readonly minSampleSize: 6;
-    readonly instabilityScoreDelta: 1.5;
-    readonly pass: {
-        readonly agreementMin: 0.9;
-        readonly maxSpread: 1;
-        readonly maxInstabilityCount: 0;
-    };
-    readonly warn: {
-        readonly agreementMin: 0.75;
-        readonly maxSpread: 2;
-        readonly maxInstabilityCount: 2;
-    };
-};
-export declare function buildVerifyConsistencyReport(runs: Array<VerifyCalibrationRun>): VerifyConsistencyReport;
-//# sourceMappingURL=verify_metrics.d.ts.map

package/script/simulator-ui/src/verify_metrics.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"verify_metrics.d.ts","sourceRoot":"","sources":["../../../src/simulator-ui/src/verify_metrics.ts"],"names":[],"mappings":"AAAA,MAAM,MAAM,aAAa,GAAG,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;AAErD,MAAM,MAAM,oBAAoB,GAAG;IACjC,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,SAAS,GAAG,WAAW,GAAG,OAAO,CAAC;IAC1C,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,OAAO,CAAC;CAClB,CAAC;AAkBF,MAAM,MAAM,aAAa,GAAG;IAC1B,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,QAAQ,EAAE,OAAO,CAAC;IAClB,WAAW,EAAE,OAAO,CAAC;IACrB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB,CAAC;AAEF,MAAM,MAAM,uBAAuB,GAAG;IACpC,UAAU,EAAE,MAAM,CAAC;IACnB,sBAAsB,EAAE,MAAM,CAAC;IAC/B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,gBAAgB,EAAE,MAAM,CAAC;IACzB,OAAO,EAAE,aAAa,CAAC;IACvB,aAAa,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,KAAK,CAAC,aAAa,CAAC,CAAC;CAChC,CAAC;AAEF,eAAO,MAAM,6BAA6B;;;;;;;;;;;;;CAahC,CAAC;AA0JX,wBAAgB,4BAA4B,CAC1C,IAAI,EAAE,KAAK,CAAC,oBAAoB,CAAC,GAChC,uBAAuB,CA0HzB"}

package/script/simulator-ui/src/verify_metrics.js

@@ -1,264 +0,0 @@
-"use strict";
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.VERIFY_CONSISTENCY_THRESHOLDS = void 0;
-exports.buildVerifyConsistencyReport = buildVerifyConsistencyReport;
-exports.VERIFY_CONSISTENCY_THRESHOLDS = {
-    minSampleSize: 6,
-    instabilityScoreDelta: 1.5,
-    pass: {
-        agreementMin: 0.9,
-        maxSpread: 1,
-        maxInstabilityCount: 0,
-    },
-    warn: {
-        agreementMin: 0.75,
-        maxSpread: 2,
-        maxInstabilityCount: 2,
-    },
-};
-const pickPayload = (result) => {
-    if (!result || typeof result !== "object")
-        return {};
-    const record = result;
-    if (record.payload &&
-        typeof record.payload === "object" &&
-        !Array.isArray(record.payload)) {
-        return record.payload;
-    }
-    return record;
-};
-const extractScoreReasonPass = (result) => {
-    const payload = pickPayload(result);
-    const score = typeof payload.score === "number" && Number.isFinite(payload.score)
-        ? payload.score
-        : undefined;
-    const reason = typeof payload.reason === "string"
-        ? payload.reason
-        : undefined;
-    if (typeof payload.pass === "boolean") {
-        return { score, reason, pass: payload.pass };
-    }
-    if (typeof payload.passed === "boolean") {
-        return { score, reason, pass: payload.passed };
-    }
-    if (typeof payload.verdict === "string") {
-        const verdict = payload.verdict.trim().toLowerCase();
-        if (verdict === "pass")
-            return { score, reason, pass: true };
-        if (verdict === "fail")
-            return { score, reason, pass: false };
-    }
-    if (typeof score === "number") {
-        return { score, reason, pass: score >= 0 };
-    }
-    return { score, reason };
-};
-const flattenRunExamples = (run) => {
-    if (!run.result || typeof run.result !== "object")
-        return [];
-    const record = run.result;
-    if (record.mode === "turns" && Array.isArray(record.turns)) {
-        const buckets = [];
-        record.turns.forEach((turn, fallbackIndex) => {
-            if (!turn || typeof turn !== "object")
-                return;
-            const turnRecord = turn;
-            const index = typeof turnRecord.index === "number" &&
-                Number.isFinite(turnRecord.index)
-                ? Math.max(0, Math.round(turnRecord.index))
-                : fallbackIndex;
-            const messageRefId = typeof turnRecord.messageRefId === "string" &&
-                turnRecord.messageRefId.trim().length > 0
-                ? turnRecord.messageRefId
-                : undefined;
-            const key = messageRefId ? `ref:${messageRefId}` : `turn:${index}`;
-            const label = `Assistant turn ${fallbackIndex + 1}`;
-            const parsed = extractScoreReasonPass(turnRecord.result);
-            buckets.push({
-                key,
-                label,
-                points: [{
-                        runId: run.id,
-                        runAt: run.runAt,
-                        score: parsed.score,
-                        pass: parsed.pass,
-                        reason: parsed.reason,
-                        turnIndex: index,
-                        messageRefId,
-                    }],
-            });
-        });
-        return buckets;
-    }
-    const parsed = extractScoreReasonPass(run.result);
-    return [{
-            key: "conversation",
-            label: "Conversation score",
-            points: [{
-                    runId: run.id,
-                    runAt: run.runAt,
-                    score: parsed.score,
-                    pass: parsed.pass,
-                    reason: parsed.reason,
-                }],
-        }];
-};
-const median = (values) => {
-    if (!values.length)
-        return null;
-    const sorted = [...values].sort((a, b) => a - b);
-    const mid = Math.floor(sorted.length / 2);
-    if (sorted.length % 2 === 1)
-        return sorted[mid];
-    return (sorted[mid - 1] + sorted[mid]) / 2;
-};
-const round2 = (value) => Math.round(value * 100) / 100;
-const resolveVerdict = (input) => {
-    const t = exports.VERIFY_CONSISTENCY_THRESHOLDS;
-    if (input.sampleSize < t.minSampleSize) {
-        return {
-            verdict: "WARN",
-            reason: `Need at least ${t.minSampleSize} samples before issuing a firm verdict.`,
-        };
-    }
-    if (input.agreementRate === null) {
-        return {
-            verdict: "WARN",
-            reason: "No comparable pass/fail evidence was found in the sampled runs.",
-        };
-    }
-    const spreadMax = input.spreadMax ?? 0;
-    if (input.agreementRate >= t.pass.agreementMin &&
-        spreadMax <= t.pass.maxSpread &&
-        input.instabilityCount <= t.pass.maxInstabilityCount) {
-        return {
-            verdict: "PASS",
-            reason: "Agreement, spread, and instability all meet PASS thresholds.",
-        };
-    }
-    if (input.agreementRate >= t.warn.agreementMin &&
-        spreadMax <= t.warn.maxSpread &&
-        input.instabilityCount <= t.warn.maxInstabilityCount) {
-        return {
-            verdict: "WARN",
-            reason: "Some variation was detected, but results remain within WARN thresholds.",
-        };
-    }
-    return {
-        verdict: "FAIL",
-        reason: "Agreement/spread instability exceeds WARN thresholds.",
-    };
-};
-function buildVerifyConsistencyReport(runs) {
-    const completedRuns = runs.filter((run) => run.status === "completed");
-    const sampleSize = completedRuns.length;
-    const bucketsByKey = new Map();
-    completedRuns.forEach((run) => {
-        flattenRunExamples(run).forEach((entry) => {
-            const existing = bucketsByKey.get(entry.key);
-            if (!existing) {
-                bucketsByKey.set(entry.key, {
-                    key: entry.key,
-                    label: entry.label,
-                    points: [...entry.points],
-                });
-                return;
-            }
-            existing.points.push(...entry.points);
-        });
-    });
-    const outliers = [];
-    let agreementVotes = 0;
-    let agreementTotal = 0;
-    const scoreDeltas = [];
-    bucketsByKey.forEach((bucket) => {
-        const scores = bucket.points
-            .map((point) => point.score)
-            .filter((score) => typeof score === "number" && Number.isFinite(score));
-        const minScore = scores.length ? Math.min(...scores) : null;
-        const maxScore = scores.length ? Math.max(...scores) : null;
-        const scoreDelta = minScore !== null && maxScore !== null
-            ? round2(maxScore - minScore)
-            : null;
-        const passVotes = bucket.points
-            .map((point) => point.pass)
-            .filter((pass) => typeof pass === "boolean");
-        const passCount = passVotes.filter((value) => value).length;
-        const failCount = passVotes.length - passCount;
-        const agreementRate = passVotes.length > 0
-            ? round2(Math.max(passCount, failCount) / passVotes.length)
-            : null;
-        if (passVotes.length > 0) {
-            agreementVotes += Math.max(passCount, failCount);
-            agreementTotal += passVotes.length;
-        }
-        if (scoreDelta !== null) {
-            scoreDeltas.push(scoreDelta);
-        }
-        const passFlip = passCount > 0 && failCount > 0;
-        const instability = passFlip ||
-            (scoreDelta !== null &&
-                scoreDelta > exports.VERIFY_CONSISTENCY_THRESHOLDS.instabilityScoreDelta);
-        const minPoint = minScore === null
-            ? undefined
-            : bucket.points.find((point) => point.score === minScore);
-        const maxPoint = maxScore === null
-            ? undefined
-            : bucket.points.find((point) => point.score === maxScore);
-        outliers.push({
-            key: bucket.key,
-            label: bucket.label,
-            sampleSize: bucket.points.length,
-            agreementRate,
-            scoreDelta,
-            passFlip,
-            instability,
-            minScore,
-            maxScore,
-            minRunId: minPoint?.runId,
-            maxRunId: maxPoint?.runId,
-            turnIndex: maxPoint?.turnIndex ?? minPoint?.turnIndex,
-            messageRefId: maxPoint?.messageRefId ?? minPoint?.messageRefId,
-        });
-    });
-    outliers.sort((a, b) => {
-        if (a.instability !== b.instability)
-            return a.instability ? -1 : 1;
-        if (a.passFlip !== b.passFlip)
-            return a.passFlip ? -1 : 1;
-        const aDelta = a.scoreDelta ?? -1;
-        const bDelta = b.scoreDelta ?? -1;
-        if (aDelta !== bDelta)
-            return bDelta - aDelta;
-        if (a.sampleSize !== b.sampleSize)
-            return b.sampleSize - a.sampleSize;
-        return a.label.localeCompare(b.label);
-    });
-    const agreementRate = agreementTotal > 0
-        ? round2(agreementVotes / agreementTotal)
-        : null;
-    const scoreSpreadMin = scoreDeltas.length ? Math.min(...scoreDeltas) : null;
-    const scoreSpreadMax = scoreDeltas.length ? Math.max(...scoreDeltas) : null;
-    const scoreSpreadMedian = median(scoreDeltas);
-    const instabilityCount = outliers.filter((entry) => entry.instability).length;
-    const verdict = resolveVerdict({
-        sampleSize,
-        agreementRate,
-        spreadMax: scoreSpreadMax,
-        instabilityCount,
-    });
-    return {
-        sampleSize,
-        comparableExampleCount: outliers.length,
-        agreementRate,
-        scoreSpreadMin: scoreSpreadMin === null ? null : round2(scoreSpreadMin),
-        scoreSpreadMedian: scoreSpreadMedian === null
-            ? null
-            : round2(scoreSpreadMedian),
-        scoreSpreadMax: scoreSpreadMax === null ? null : round2(scoreSpreadMax),
-        instabilityCount,
-        verdict: verdict.verdict,
-        verdictReason: verdict.reason,
-        outliers,
-    };
-}