akm-cli 0.7.0-rc1 → 0.7.1

package/dist/tests/bench/metrics.test.js

@@ -1,1144 +0,0 @@
-/**
- * Unit tests for outcome / per-task / corpus / trajectory aggregation.
- */
-import { describe, expect, test } from "bun:test";
-import fs from "node:fs";
-import path from "node:path";
-import { aggregateAkmOverhead, aggregateByMemoryAbility, aggregateByTaskFamily, aggregateCorpus, aggregatePerTask, aggregateTrajectory, computeAkmOverhead, computeAssetRegressionCandidates, computeCorpusCoverage, computeCorpusDelta, computeDomainAggregates, computeLearningCurve, computeNegativeTransfer, computeOutcomeAggregate, computePerTaskDelta, computeWorkflowReliability, domainOfTaskId, isPathContained, LEARNING_IMPROVEMENT_THRESHOLD, materialiseMaskedStash, } from "./metrics";
-import { benchMkdtemp } from "./tmp";
-function ptm(overrides = {}) {
-    return {
-        passRate: 0,
-        passAt1: 0,
-        tokensPerPass: null,
-        wallclockMs: 0,
-        passRateStdev: 0,
-        budgetExceededCount: 0,
-        harnessErrorCount: 0,
-        count: 1,
-        runsWithMeasuredTokens: 0,
-        ...overrides,
-    };
-}
-function fakeResult(overrides) {
-    return {
-        schemaVersion: 1,
-        taskId: "t",
-        arm: "akm",
-        seed: 0,
-        model: "m",
-        outcome: "pass",
-        tokens: { input: 0, output: 0 },
-        wallclockMs: 0,
-        trajectory: { correctAssetLoaded: null, feedbackRecorded: null },
-        events: [],
-        verifierStdout: "",
-        verifierExitCode: 0,
-        assetsLoaded: [],
-        ...overrides,
-    };
-}
-describe("computeOutcomeAggregate", () => {
-    test("returns zeros on empty input", () => {
-        expect(computeOutcomeAggregate([])).toEqual({
-            passRate: 0,
-            tokensPerPass: 0,
-            wallclockMs: 0,
-            budgetExceeded: 0,
-            runsWithMeasuredTokens: 0,
-        });
-    });
-    test("computes passRate, tokensPerPass, wallclockMs across mixed outcomes", () => {
-        const results = [
-            fakeResult({ outcome: "pass", tokens: { input: 1000, output: 500 }, wallclockMs: 1000 }),
-            fakeResult({ outcome: "pass", tokens: { input: 2000, output: 1000 }, wallclockMs: 2000 }),
-            fakeResult({ outcome: "fail", tokens: { input: 500, output: 200 }, wallclockMs: 1500 }),
-            fakeResult({ outcome: "budget_exceeded", tokens: { input: 100, output: 50 }, wallclockMs: 500 }),
-        ];
-        const agg = computeOutcomeAggregate(results);
-        expect(agg.passRate).toBeCloseTo(0.5);
-        expect(agg.tokensPerPass).toBeCloseTo((1500 + 3000) / 2);
-        expect(agg.wallclockMs).toBeCloseTo((1000 + 2000 + 1500 + 500) / 4);
-        expect(agg.budgetExceeded).toBe(1);
-    });
-    test("tokensPerPass is 0 (not NaN) when no runs passed", () => {
-        const results = [fakeResult({ outcome: "fail", wallclockMs: 100 })];
-        const agg = computeOutcomeAggregate(results);
-        expect(agg.passRate).toBe(0);
-        expect(agg.tokensPerPass).toBe(0);
-    });
-    test("missing token measurement is NOT silently treated as zero (issue #252)", () => {
-        // Two passes: one parsed at 1000, one missing measurement. The mean must
-        // be 1000 (the measured pass), not (1000+0)/2 = 500.
-        const results = [
-            fakeResult({
-                outcome: "pass",
-                tokens: { input: 700, output: 300 },
-                tokenMeasurement: "parsed",
-            }),
-            fakeResult({
-                outcome: "pass",
-                tokens: { input: 0, output: 0 },
-                tokenMeasurement: "missing",
-            }),
-        ];
-        const agg = computeOutcomeAggregate(results);
-        expect(agg.passRate).toBeCloseTo(1);
-        expect(agg.tokensPerPass).toBeCloseTo(1000);
-        expect(agg.runsWithMeasuredTokens).toBe(1);
-    });
-    test("unsupported token measurement is also skipped from token aggregation", () => {
-        const results = [
-            fakeResult({
-                outcome: "pass",
-                tokens: { input: 0, output: 0 },
-                tokenMeasurement: "unsupported",
-            }),
-        ];
-        const agg = computeOutcomeAggregate(results);
-        // No measured passes → tokensPerPass collapses to 0, but runsWithMeasuredTokens=0
-        // signals that the 0 is "unknown", not "free".
-        expect(agg.tokensPerPass).toBe(0);
-        expect(agg.runsWithMeasuredTokens).toBe(0);
-    });
-});
-describe("aggregatePerTask", () => {
-    test("0 of K passes — tokensPerPass is null, passRate is 0", () => {
-        const runs = [
-            fakeResult({ seed: 0, outcome: "fail", wallclockMs: 1000 }),
-            fakeResult({ seed: 1, outcome: "fail", wallclockMs: 2000 }),
-            fakeResult({ seed: 2, outcome: "harness_error", wallclockMs: 3000 }),
-        ];
-        const m = aggregatePerTask(runs);
-        expect(m.passRate).toBe(0);
-        expect(m.passAt1).toBe(0);
-        expect(m.tokensPerPass).toBeNull();
-        expect(m.wallclockMs).toBe(2000);
-        expect(m.harnessErrorCount).toBe(1);
-        expect(m.budgetExceededCount).toBe(0);
-        expect(m.count).toBe(3);
-    });
-    test("K of K passes — passRate is 1, stdev is 0", () => {
-        const runs = Array.from({ length: 5 }, (_, i) => fakeResult({ seed: i, outcome: "pass", tokens: { input: 1000, output: 0 }, wallclockMs: 1000 }));
-        const m = aggregatePerTask(runs);
-        expect(m.passRate).toBe(1);
-        expect(m.passAt1).toBe(1);
-        expect(m.tokensPerPass).toBe(1000);
-        expect(m.passRateStdev).toBe(0);
-    });
-    test("partial passes — passRate, stdev, and budget_exceeded count are computed", () => {
-        const runs = [
-            fakeResult({ seed: 0, outcome: "pass", tokens: { input: 800, output: 200 }, wallclockMs: 1000 }),
-            fakeResult({ seed: 1, outcome: "pass", tokens: { input: 1200, output: 300 }, wallclockMs: 1500 }),
-            fakeResult({ seed: 2, outcome: "fail", wallclockMs: 2000 }),
-            fakeResult({ seed: 3, outcome: "budget_exceeded", wallclockMs: 3000 }),
-        ];
-        const m = aggregatePerTask(runs);
-        expect(m.passRate).toBeCloseTo(0.5);
-        expect(m.passAt1).toBe(1);
-        expect(m.tokensPerPass).toBeCloseTo((1000 + 1500) / 2);
-        expect(m.budgetExceededCount).toBe(1);
-        // Sample stdev of [1, 1, 0, 0] over 4 samples = sqrt(4/3 * 0.25) — non-zero.
-        expect(m.passRateStdev).toBeGreaterThan(0);
-    });
-    test("passAt1 honours seed=0 specifically when present", () => {
-        const runs = [
-            fakeResult({ seed: 1, outcome: "pass" }),
-            fakeResult({ seed: 0, outcome: "fail" }),
-            fakeResult({ seed: 2, outcome: "pass" }),
-        ];
-        const m = aggregatePerTask(runs);
-        expect(m.passAt1).toBe(0);
-    });
-    test("empty input returns a zeroed envelope", () => {
-        const m = aggregatePerTask([]);
-        expect(m.count).toBe(0);
-        expect(m.passRate).toBe(0);
-        expect(m.tokensPerPass).toBeNull();
-        expect(m.runsWithMeasuredTokens).toBe(0);
-    });
-    test("aggregatePerTask: passes with missing measurement do NOT pull tokensPerPass to zero", () => {
-        const runs = [
-            fakeResult({
-                seed: 0,
-                outcome: "pass",
-                tokens: { input: 800, output: 200 },
-                tokenMeasurement: "parsed",
-                wallclockMs: 1000,
-            }),
-            fakeResult({
-                seed: 1,
-                outcome: "pass",
-                tokens: { input: 0, output: 0 },
-                tokenMeasurement: "missing",
-                wallclockMs: 1000,
-            }),
-        ];
-        const m = aggregatePerTask(runs);
-        expect(m.passRate).toBe(1);
-        // Mean is over the single measured pass, not (1000 + 0) / 2.
-        expect(m.tokensPerPass).toBeCloseTo(1000);
-        expect(m.runsWithMeasuredTokens).toBe(1);
-        expect(m.count).toBe(2);
-    });
-    test("aggregatePerTask: tokensPerPass is null when every pass has missing measurement", () => {
-        const runs = [
-            fakeResult({
-                seed: 0,
-                outcome: "pass",
-                tokens: { input: 0, output: 0 },
-                tokenMeasurement: "missing",
-            }),
-            fakeResult({
-                seed: 1,
-                outcome: "pass",
-                tokens: { input: 0, output: 0 },
-                tokenMeasurement: "unsupported",
-            }),
-        ];
-        const m = aggregatePerTask(runs);
-        expect(m.passRate).toBe(1);
-        expect(m.tokensPerPass).toBeNull();
-        expect(m.runsWithMeasuredTokens).toBe(0);
-    });
-});
-describe("aggregateCorpus", () => {
-    test("weights every task equally regardless of seed count", () => {
-        const perTask = {
-            a: {
-                passRate: 1,
-                passAt1: 1,
-                tokensPerPass: 1000,
-                wallclockMs: 1000,
-                passRateStdev: 0,
-                budgetExceededCount: 0,
-                harnessErrorCount: 0,
-                count: 5,
-                runsWithMeasuredTokens: 5,
-            },
-            b: {
-                passRate: 0,
-                passAt1: 0,
-                tokensPerPass: null,
-                wallclockMs: 2000,
-                passRateStdev: 0,
-                budgetExceededCount: 0,
-                harnessErrorCount: 0,
-                count: 1,
-                runsWithMeasuredTokens: 0,
-            },
-        };
-        const corpus = aggregateCorpus(perTask);
-        expect(corpus.passRate).toBeCloseTo(0.5);
-        expect(corpus.wallclockMs).toBeCloseTo(1500);
-        expect(corpus.tokensPerPass).toBeCloseTo(1000); // null is dropped
-    });
-    test("tokensPerPass is null when every task has null tokensPerPass", () => {
-        const perTask = {
-            a: {
-                passRate: 0,
-                passAt1: 0,
-                tokensPerPass: null,
-                wallclockMs: 1000,
-                passRateStdev: 0,
-                budgetExceededCount: 0,
-                harnessErrorCount: 0,
-                count: 1,
-                runsWithMeasuredTokens: 0,
-            },
-        };
-        const corpus = aggregateCorpus(perTask);
-        expect(corpus.tokensPerPass).toBeNull();
-    });
-    test("empty input returns zeros + null tokens", () => {
-        const corpus = aggregateCorpus({});
-        expect(corpus.passRate).toBe(0);
-        expect(corpus.tokensPerPass).toBeNull();
-    });
-});
-describe("delta helpers", () => {
-    test("computeCorpusDelta — akm − noakm", () => {
-        const noakm = { passRate: 0.3, tokensPerPass: 18000, wallclockMs: 4000 };
-        const akm = { passRate: 0.7, tokensPerPass: 14000, wallclockMs: 3000 };
-        const d = computeCorpusDelta(noakm, akm);
-        expect(d.passRate).toBeCloseTo(0.4);
-        expect(d.tokensPerPass).toBeCloseTo(-4000);
-        expect(d.wallclockMs).toBeCloseTo(-1000);
-    });
-    test("computeCorpusDelta — null tokensPerPass propagates", () => {
-        const noakm = { passRate: 0, tokensPerPass: null, wallclockMs: 1 };
-        const akm = { passRate: 1, tokensPerPass: 5, wallclockMs: 2 };
-        expect(computeCorpusDelta(noakm, akm).tokensPerPass).toBeNull();
-    });
-    test("computePerTaskDelta — same null-safety rule", () => {
-        const noakm = {
-            passRate: 0,
-            passAt1: 0,
-            tokensPerPass: null,
-            wallclockMs: 0,
-            passRateStdev: 0,
-            budgetExceededCount: 0,
-            harnessErrorCount: 0,
-            count: 1,
-            runsWithMeasuredTokens: 0,
-        };
-        const akm = {
-            passRate: 1,
-            passAt1: 1,
-            tokensPerPass: 1000,
-            wallclockMs: 100,
-            passRateStdev: 0,
-            budgetExceededCount: 0,
-            harnessErrorCount: 0,
-            count: 1,
-            runsWithMeasuredTokens: 1,
-        };
-        expect(computePerTaskDelta(noakm, akm).tokensPerPass).toBeNull();
-    });
-});
-describe("aggregateTrajectory", () => {
-    test("returns null/0 on empty input", () => {
-        const t = aggregateTrajectory([]);
-        expect(t.correctAssetLoaded).toBeNull();
-        expect(t.feedbackRecorded).toBe(0);
-    });
-    test("correctAssetLoaded is null when no run had a known goldRef", () => {
-        const runs = [
-            fakeResult({ trajectory: { correctAssetLoaded: null, feedbackRecorded: false } }),
-            fakeResult({ trajectory: { correctAssetLoaded: null, feedbackRecorded: true } }),
-        ];
-        const t = aggregateTrajectory(runs);
-        expect(t.correctAssetLoaded).toBeNull();
-        expect(t.feedbackRecorded).toBeCloseTo(0.5);
-    });
-    test("correctAssetLoaded is fraction over runs with goldRef", () => {
-        const runs = [
-            fakeResult({ trajectory: { correctAssetLoaded: true, feedbackRecorded: false } }),
-            fakeResult({ trajectory: { correctAssetLoaded: false, feedbackRecorded: false } }),
-            fakeResult({ trajectory: { correctAssetLoaded: null, feedbackRecorded: false } }),
-        ];
-        const t = aggregateTrajectory(runs);
-        expect(t.correctAssetLoaded).toBeCloseTo(0.5);
-        expect(t.feedbackRecorded).toBe(0);
-    });
-});
-describe("domainOfTaskId", () => {
-    test("returns the segment before the first slash", () => {
-        expect(domainOfTaskId("docker-homelab/redis-healthcheck")).toBe("docker-homelab");
-    });
-    test("falls back to 'unknown' when there is no slash", () => {
-        expect(domainOfTaskId("noslash")).toBe("unknown");
-    });
-    test("falls back to 'unknown' when the slash is at index 0", () => {
-        expect(domainOfTaskId("/leading")).toBe("unknown");
-    });
-});
-describe("computeNegativeTransfer", () => {
-    test("returns zero count and severity when no regressions are present", () => {
-        const tasks = [
-            { id: "d/a", noakm: ptm({ passRate: 0.4 }), akm: ptm({ passRate: 0.8 }) },
-            { id: "d/b", noakm: ptm({ passRate: 0.5 }), akm: ptm({ passRate: 0.5 }) },
-        ];
-        const out = computeNegativeTransfer(tasks);
-        expect(out.count).toBe(0);
-        expect(out.severity).toBe(0);
-        expect(out.topRegressedTasks).toEqual([]);
-    });
-    test("captures a single regression with correct delta and severity", () => {
-        const tasks = [
-            { id: "d/a", noakm: ptm({ passRate: 0.4 }), akm: ptm({ passRate: 0.8 }) },
-            { id: "d/regressed", noakm: ptm({ passRate: 0.6 }), akm: ptm({ passRate: 0.2 }) },
-        ];
-        const out = computeNegativeTransfer(tasks);
-        expect(out.count).toBe(1);
-        expect(out.severity).toBeCloseTo(0.4);
-        expect(out.topRegressedTasks).toHaveLength(1);
-        const row = out.topRegressedTasks[0];
-        if (!row)
-            throw new Error("expected row");
-        expect(row.taskId).toBe("d/regressed");
-        expect(row.domain).toBe("d");
-        expect(row.delta).toBeCloseTo(-0.4);
-        expect(row.severity).toBeCloseTo(0.4);
-    });
-    test("multiple regressions are sorted by severity desc with deterministic tiebreak", () => {
-        const tasks = [
-            // Mild regression -0.1.
-            { id: "alpha/x", noakm: ptm({ passRate: 0.6 }), akm: ptm({ passRate: 0.5 }) },
-            // Tied severity -0.3 (first tiebreaks by taskId asc).
-            { id: "beta/y", noakm: ptm({ passRate: 0.8 }), akm: ptm({ passRate: 0.5 }) },
-            { id: "alpha/z", noakm: ptm({ passRate: 0.8 }), akm: ptm({ passRate: 0.5 }) },
-            // Improvement (no regression).
-            { id: "alpha/w", noakm: ptm({ passRate: 0.1 }), akm: ptm({ passRate: 0.9 }) },
-        ];
-        const out = computeNegativeTransfer(tasks);
-        expect(out.count).toBe(3);
-        expect(out.severity).toBeCloseTo(0.7);
-        expect(out.topRegressedTasks.map((r) => r.taskId)).toEqual(["alpha/z", "beta/y", "alpha/x"]);
-    });
-    test("a task with equal pass rate is not counted as regressed", () => {
-        const tasks = [{ id: "d/eq", noakm: ptm({ passRate: 0.5 }), akm: ptm({ passRate: 0.5 }) }];
-        expect(computeNegativeTransfer(tasks).count).toBe(0);
-    });
-});
-describe("computeDomainAggregates", () => {
-    test("groups tasks by domain prefix", () => {
-        const tasks = [
-            {
-                id: "alpha/a",
-                noakm: ptm({ passRate: 0.4, tokensPerPass: 10000, wallclockMs: 1000 }),
-                akm: ptm({ passRate: 0.8, tokensPerPass: 8000, wallclockMs: 900 }),
-            },
-            {
-                id: "alpha/b",
-                noakm: ptm({ passRate: 0.6, tokensPerPass: 12000, wallclockMs: 2000 }),
-                akm: ptm({ passRate: 0.4, tokensPerPass: 9000, wallclockMs: 1500 }),
-            },
-            {
-                id: "beta/c",
-                noakm: ptm({ passRate: 0.2, tokensPerPass: null, wallclockMs: 500 }),
-                akm: ptm({ passRate: 0.5, tokensPerPass: 5000, wallclockMs: 600 }),
-            },
-        ];
-        const rows = computeDomainAggregates(tasks);
-        expect(rows.map((r) => r.domain)).toEqual(["alpha", "beta"]);
-        const alpha = rows.find((r) => r.domain === "alpha");
-        if (!alpha)
-            throw new Error("alpha missing");
-        expect(alpha.taskCount).toBe(2);
-        expect(alpha.regressionCount).toBe(1);
-        expect(alpha.passRateNoakm).toBeCloseTo(0.5);
-        expect(alpha.passRateAkm).toBeCloseTo(0.6);
-        expect(alpha.passRateDelta).toBeCloseTo(0.1);
-        expect(alpha.tokensPerPassDelta).toBeCloseTo(8500 - 11000);
-        expect(alpha.wallclockMsDelta).toBeCloseTo(1200 - 1500);
-        const beta = rows.find((r) => r.domain === "beta");
-        if (!beta)
-            throw new Error("beta missing");
-        expect(beta.regressionCount).toBe(0);
-        // Single-side null tokensPerPass yields null delta.
-        expect(beta.tokensPerPassDelta).toBeNull();
-    });
-    test("emits an empty array on no tasks", () => {
-        expect(computeDomainAggregates([])).toEqual([]);
-    });
-});
-describe("computeAssetRegressionCandidates", () => {
-    function fakeRun(taskId, assets) {
-        return {
-            schemaVersion: 1,
-            taskId,
-            arm: "akm",
-            seed: 0,
-            model: "m",
-            outcome: "pass",
-            tokens: { input: 0, output: 0 },
-            wallclockMs: 0,
-            trajectory: { correctAssetLoaded: null, feedbackRecorded: null },
-            events: [],
-            verifierStdout: "",
-            verifierExitCode: 0,
-            assetsLoaded: assets,
-        };
-    }
-    test("returns empty when no regressed tasks were provided", () => {
-        expect(computeAssetRegressionCandidates([], [fakeRun("d/a", ["skill:x"])])).toEqual([]);
-    });
-    test("counts distinct regressed tasks per asset and totals raw load volume", () => {
-        const akmRuns = [
-            // task d/r1 across two seeds, same asset.
-            fakeRun("d/r1", ["skill:foo", "skill:bar"]),
-            fakeRun("d/r1", ["skill:foo"]),
-            // task d/r2 loads skill:foo (again) plus skill:baz.
-            fakeRun("d/r2", ["skill:foo", "skill:baz"]),
-            // Non-regressed task is ignored entirely.
-            fakeRun("d/clean", ["skill:foo", "skill:bar", "skill:baz"]),
-        ];
-        const rows = computeAssetRegressionCandidates(["d/r1", "d/r2"], akmRuns);
-        expect(rows.map((r) => r.assetRef)).toEqual(["skill:foo", "skill:bar", "skill:baz"]);
-        const foo = rows[0];
-        if (!foo)
-            throw new Error("foo missing");
-        expect(foo.regressedTaskCount).toBe(2);
-        expect(foo.regressedTaskIds).toEqual(["d/r1", "d/r2"]);
-        expect(foo.totalLoadCount).toBe(3);
-        const bar = rows[1];
-        if (!bar)
-            throw new Error("bar missing");
-        expect(bar.regressedTaskCount).toBe(1);
-        expect(bar.totalLoadCount).toBe(1);
-    });
-});
-// ── Memory-operation aggregations (#262) ───────────────────────────────────
-describe("aggregateByMemoryAbility / aggregateByTaskFamily (#262)", () => {
-    function entry(id, noakmPass, akmPass, extras = {}) {
-        return {
-            id,
-            noakm: ptm({ passRate: noakmPass }),
-            akm: ptm({ passRate: akmPass }),
-            ...extras,
-        };
-    }
-    test("returns empty when no entries carry the keying tag", () => {
-        const entries = [entry("d/a", 0.4, 0.6), entry("d/b", 0.5, 0.7)];
-        expect(aggregateByMemoryAbility(entries)).toEqual([]);
-        expect(aggregateByTaskFamily(entries)).toEqual([]);
-    });
-    test("aggregateByMemoryAbility groups tasks, computes deltas + negative transfer", () => {
-        const entries = [
-            entry("d/lookup-1", 0.4, 0.8, { memoryAbility: "procedural_lookup" }),
-            entry("d/lookup-2", 0.6, 0.4, { memoryAbility: "procedural_lookup" }),
-            entry("d/compose-1", 0.0, 1.0, { memoryAbility: "multi_asset_composition" }),
-            entry("d/no-tag", 0.5, 0.7),
-        ];
-        const rows = aggregateByMemoryAbility(entries);
-        expect(rows.map((r) => r.category)).toEqual(["multi_asset_composition", "procedural_lookup"]);
-        const lookup = rows.find((r) => r.category === "procedural_lookup");
-        expect(lookup?.taskCount).toBe(2);
-        expect(lookup?.passRateNoakm).toBeCloseTo(0.5);
-        expect(lookup?.passRateAkm).toBeCloseTo(0.6);
-        expect(lookup?.passRateDelta).toBeCloseTo(0.1);
-        // d/lookup-2 regressed (akm < noakm).
-        expect(lookup?.negativeTransferCount).toBe(1);
-        expect(lookup?.workflowCompliance).toBeNull();
-    });
-    test("aggregateByMemoryAbility folds workflow_compliance when at least one task supplies it", () => {
-        const entries = [
-            entry("d/a", 0.5, 0.7, { memoryAbility: "procedural_lookup", workflowCompliance: 0.8 }),
-            entry("d/b", 0.5, 0.7, { memoryAbility: "procedural_lookup" }),
-            entry("d/c", 0.5, 0.7, { memoryAbility: "procedural_lookup", workflowCompliance: 0.6 }),
-        ];
-        const [row] = aggregateByMemoryAbility(entries);
-        expect(row?.workflowCompliance).toBeCloseTo(0.7);
-    });
-    test("aggregateByTaskFamily groups by family", () => {
-        const entries = [
-            entry("d/a", 0.4, 0.6, { taskFamily: "d/group-1" }),
-            entry("d/b", 0.4, 0.4, { taskFamily: "d/group-1" }),
-            entry("d/c", 0.0, 1.0, { taskFamily: "d/group-2" }),
-        ];
-        const rows = aggregateByTaskFamily(entries);
-        expect(rows.map((r) => r.category)).toEqual(["d/group-1", "d/group-2"]);
-        const g1 = rows.find((r) => r.category === "d/group-1");
-        expect(g1?.taskCount).toBe(2);
-        expect(g1?.passRateDelta).toBeCloseTo(0.1);
-    });
-    test("computeCorpusCoverage counts every closed-set ability + an untagged bucket", () => {
-        const cov = computeCorpusCoverage([
-            { memoryAbility: "procedural_lookup", taskFamily: "d/family-a" },
-            { memoryAbility: "procedural_lookup", taskFamily: "d/family-a" },
-            { memoryAbility: "abstention", taskFamily: "d/family-b" },
-            { taskFamily: "d/family-c" },
-            {},
-        ]);
-        expect(cov.totalTasks).toBe(5);
-        expect(cov.memoryAbilityCounts.procedural_lookup).toBe(2);
-        expect(cov.memoryAbilityCounts.abstention).toBe(1);
-        expect(cov.memoryAbilityCounts.conflict_resolution).toBe(0);
-        expect(cov.memoryAbilityCounts.untagged).toBe(2);
-        expect(cov.taskFamilyCounts["d/family-a"]).toBe(2);
-        expect(cov.taskFamilyCounts.untagged).toBe(1);
-    });
-});
-// ── AKM overhead (#263) ────────────────────────────────────────────────────
-function akmEvent(eventType, ts, ref, metadata) {
-    return {
-        schemaVersion: 1,
-        id: 0,
-        ts,
-        eventType,
-        ...(ref ? { ref } : {}),
-        ...(metadata ? { metadata } : {}),
-    };
-}
-function metaMap(entries) {
-    const m = new Map();
-    for (const e of entries)
-        m.set(e.id, { goldRef: e.goldRef, expectedTransferFrom: e.expectedTransferFrom });
-    return m;
-}
-describe("computeAkmOverhead — no AKM calls", () => {
-    test("zero counts and null timings when run had no AKM events", () => {
-        const run = fakeResult({ taskId: "demo/none", events: [] });
-        const rows = computeAkmOverhead([run]);
-        expect(rows).toHaveLength(1);
-        const r = rows[0];
-        expect(r.searchCount).toBe(0);
-        expect(r.showCount).toBe(0);
-        expect(r.feedbackCount).toBe(0);
-        expect(r.totalToolCalls).toBe(0);
-        expect(r.assetsLoadedCount).toBe(0);
-        expect(r.timeToFirstSearchMs).toBeNull();
-        expect(r.timeToFirstCorrectAssetMs).toBeNull();
-        expect(r.contextBytesLoaded).toBeNull();
-        expect(r.assetBytesLoaded).toBeNull();
-        // Without metadata, irrelevance is unjudgeable -> null.
-        expect(r.irrelevantAssetsLoadedCount).toBeNull();
-    });
-    test("aggregate over empty array is the zero envelope", () => {
-        const agg = aggregateAkmOverhead([]);
-        expect(agg.totalRuns).toBe(0);
-        expect(agg.passingRuns).toBe(0);
-        expect(agg.toolCallsPerSuccess).toBeNull();
-        expect(agg.costPerSuccess).toBeNull();
-        expect(agg.meanTimeToFirstSearchMs).toBeNull();
-    });
-});
-describe("computeAkmOverhead — successful AKM use", () => {
-    test("counts search/show/feedback, computes timings and relevance", () => {
-        const run = fakeResult({
-            taskId: "demo/ok",
-            outcome: "pass",
-            tokenMeasurement: "parsed",
-            tokens: { input: 100, output: 50 },
-            events: [
-                akmEvent("search", "2026-04-27T10:00:00.000Z", undefined, { query: "deploy" }),
-                akmEvent("show", "2026-04-27T10:00:00.500Z", "skill:deploy"),
-                akmEvent("feedback", "2026-04-27T10:00:01.000Z", "skill:deploy"),
-            ],
-        });
-        const tasks = metaMap([{ id: "demo/ok", goldRef: "skill:deploy", expectedTransferFrom: [] }]);
-        const rows = computeAkmOverhead([run], { taskMetadata: tasks });
-        const r = rows[0];
-        expect(r.searchCount).toBe(1);
-        expect(r.showCount).toBe(1);
-        expect(r.feedbackCount).toBe(1);
-        expect(r.totalToolCalls).toBe(3);
-        expect(r.assetsLoadedCount).toBe(1);
-        expect(r.irrelevantAssetsLoadedCount).toBe(0);
-        expect(r.timeToFirstSearchMs).toBe(0); // first search IS the run-start anchor
-        expect(r.timeToFirstCorrectAssetMs).toBe(500);
-        const agg = aggregateAkmOverhead(rows, [run]);
-        expect(agg.passingRuns).toBe(1);
-        expect(agg.toolCallsPerSuccess).toBe(3);
-        expect(agg.costPerSuccess).toBe(150);
-    });
-    test("expected_transfer_from refs are not counted as irrelevant", () => {
-        const run = fakeResult({
-            taskId: "demo/transfer",
-            events: [
-                akmEvent("show", "2026-04-27T10:00:00.000Z", "skill:foo"),
-                akmEvent("show", "2026-04-27T10:00:01.000Z", "skill:helper"),
-            ],
-        });
-        const tasks = metaMap([{ id: "demo/transfer", goldRef: "skill:foo", expectedTransferFrom: ["skill:helper"] }]);
-        const rows = computeAkmOverhead([run], { taskMetadata: tasks });
-        expect(rows[0].assetsLoadedCount).toBe(2);
-        expect(rows[0].irrelevantAssetsLoadedCount).toBe(0);
-    });
-});
-describe("computeAkmOverhead — excessive AKM calls", () => {
-    test("high counts and low calls-per-success are surfaced", () => {
-        const goldRef = "skill:gold";
-        const noisyRun = fakeResult({
-            taskId: "demo/noisy",
-            outcome: "fail",
-            events: [
-                akmEvent("search", "2026-04-27T10:00:00.000Z"),
-                akmEvent("search", "2026-04-27T10:00:00.100Z"),
-                akmEvent("search", "2026-04-27T10:00:00.200Z"),
-                akmEvent("show", "2026-04-27T10:00:00.300Z", "skill:other"),
-                akmEvent("show", "2026-04-27T10:00:00.400Z", "skill:other2"),
-                akmEvent("show", "2026-04-27T10:00:00.500Z", "skill:other3"),
-                akmEvent("show", "2026-04-27T10:00:00.600Z", goldRef),
-            ],
-        });
-        const passingRun = fakeResult({
-            taskId: "demo/easy",
-            outcome: "pass",
-            tokenMeasurement: "parsed",
-            tokens: { input: 10, output: 10 },
-            events: [akmEvent("search", "2026-04-27T10:00:00.000Z"), akmEvent("show", "2026-04-27T10:00:00.100Z", goldRef)],
-        });
-        const tasks = metaMap([
-            { id: "demo/noisy", goldRef, expectedTransferFrom: [] },
-            { id: "demo/easy", goldRef, expectedTransferFrom: [] },
-        ]);
-        const rows = computeAkmOverhead([noisyRun, passingRun], { taskMetadata: tasks });
-        expect(rows[0].totalToolCalls).toBe(7);
-        expect(rows[0].irrelevantAssetsLoadedCount).toBe(3);
-        expect(rows[0].timeToFirstCorrectAssetMs).toBe(600);
-        expect(rows[1].totalToolCalls).toBe(2);
-        const agg = aggregateAkmOverhead(rows, [noisyRun, passingRun]);
-        expect(agg.totalToolCalls).toBe(9);
-        expect(agg.passingRuns).toBe(1);
-        // 9 tool calls for one passing run = high overhead per success.
-        expect(agg.toolCallsPerSuccess).toBe(9);
-    });
-});
-describe("computeAkmOverhead — missing timing/byte data", () => {
-    test("event without ts -> null first-search timing (NOT zero)", () => {
-        const run = fakeResult({
-            taskId: "demo/notime",
-            events: [
-                // No ts on event — workflow-trace assigns a synthetic order hint but
-                // ts stays undefined, so we cannot anchor a real time-offset.
-                { schemaVersion: 1, id: 0, eventType: "search" },
-            ],
-        });
-        const rows = computeAkmOverhead([run]);
-        expect(rows[0].searchCount).toBe(1);
-        expect(rows[0].timeToFirstSearchMs).toBeNull();
-        expect(rows[0].timeToFirstCorrectAssetMs).toBeNull();
-    });
-    test("byte sizes are always null for now (NOT zero)", () => {
-        const run = fakeResult({
-            events: [akmEvent("show", "2026-04-27T10:00:00.000Z", "skill:foo")],
-        });
-        const rows = computeAkmOverhead([run]);
-        expect(rows[0].contextBytesLoaded).toBeNull();
-        expect(rows[0].assetBytesLoaded).toBeNull();
-        const agg = aggregateAkmOverhead(rows, [run]);
-        expect(agg.meanContextBytesLoaded).toBeNull();
-        expect(agg.meanAssetBytesLoaded).toBeNull();
-    });
-    test("cost_per_success is null when any passing run lacks parsed token measurement", () => {
-        const passParsed = fakeResult({
-            taskId: "t1",
-            outcome: "pass",
-            tokenMeasurement: "parsed",
-            tokens: { input: 10, output: 5 },
-            events: [akmEvent("search", "2026-04-27T10:00:00.000Z")],
-        });
-        const passMissing = fakeResult({
-            taskId: "t2",
-            outcome: "pass",
-            tokenMeasurement: "missing",
-            tokens: { input: 0, output: 0 },
-            events: [akmEvent("search", "2026-04-27T10:00:00.000Z")],
-        });
-        const rows = computeAkmOverhead([passParsed, passMissing]);
-        const agg = aggregateAkmOverhead(rows, [passParsed, passMissing]);
-        expect(agg.passingRuns).toBe(2);
-        expect(agg.costPerSuccess).toBeNull();
-    });
-    test("missing task metadata -> irrelevantAssetsLoadedCount is null (not 0)", () => {
-        const run = fakeResult({
-            taskId: "demo/unknown",
-            events: [akmEvent("show", "2026-04-27T10:00:00.000Z", "skill:foo")],
-        });
-        // No metadata supplied for this task.
-        const rows = computeAkmOverhead([run]);
-        expect(rows[0].assetsLoadedCount).toBe(1);
-        expect(rows[0].irrelevantAssetsLoadedCount).toBeNull();
-    });
-    test("aggregate skips null timings rather than zero-filling", () => {
-        const noTime = fakeResult({
-            taskId: "t1",
-            outcome: "fail",
-            events: [{ schemaVersion: 1, id: 0, eventType: "search" }],
-        });
-        const withTime = fakeResult({
-            taskId: "t2",
-            outcome: "fail",
-            events: [akmEvent("search", "2026-04-27T10:00:01.000Z")],
-        });
-        const rows = computeAkmOverhead([noTime, withTime]);
-        // First run: search event has no ts -> no run-start anchor, timing null.
-        // Second run: search event IS the only event with ts, so it's both the
-        // anchor and the first search -> offset 0.
-        expect(rows[0].timeToFirstSearchMs).toBeNull();
-        expect(rows[1].timeToFirstSearchMs).toBe(0);
-        const agg = aggregateAkmOverhead(rows, [noTime, withTime]);
-        // Mean honours only the parseable observation; the null is skipped, NOT
-        // treated as zero in the numerator.
-        expect(agg.meanTimeToFirstSearchMs).toBe(0);
-        // tool_calls_per_success is null because no run passed.
-        expect(agg.toolCallsPerSuccess).toBeNull();
-    });
-});
-// ── computeWorkflowReliability (#258) ──────────────────────────────────────
-function wfCheck(overrides = {}) {
-    return {
-        schemaVersion: 1,
-        workflowId: "wf-1",
-        taskId: "t1",
-        arm: "akm",
-        seed: 0,
-        status: "pass",
-        score: 1,
-        requiredPassed: 1,
-        requiredTotal: 1,
-        violations: [],
-        evidence: {
-            matchedEvents: 1,
-            feedbackRecorded: false,
-            goldAssetLoaded: false,
-            traceTruncated: false,
-        },
-        ...overrides,
-    };
-}
-// ── Learning curve across episodes (issue #265) ────────────────────────────
-function ep(overrides) {
-    return {
-        delta_from_previous_episode: 0,
-        cumulative_feedback_events: 0,
-        cumulative_proposals_created: 0,
-        cumulative_proposals_accepted: 0,
-        cumulative_lessons_created: 0,
-        lesson_reuse_rate: null,
-        ...overrides,
-    };
-}
-function statuses(workflowId, taskId, statusList) {
-    return statusList.map((status, seed) => wfCheck({ workflowId, taskId, seed, status }));
-}
-describe("computeWorkflowReliability (#258)", () => {
-    test("empty input yields zeroed corpus + empty by_workflow", () => {
-        const result = computeWorkflowReliability([]);
-        expect(result.byWorkflow).toEqual({});
-        expect(result.corpus).toEqual({ pass_at_k: 0, pass_all_k: 0, groups: 0, tasks: 0 });
-    });
-    test("all-pass: every (task, seed) is pass → pass@k=1, pass^k=1", () => {
-        const checks = [
-            ...statuses("wf-1", "t1", ["pass", "pass", "pass"]),
-            ...statuses("wf-1", "t2", ["pass", "pass", "pass"]),
-        ];
-        const result = computeWorkflowReliability(checks);
-        expect(result.byWorkflow["wf-1"].pass_at_k).toBe(1);
-        expect(result.byWorkflow["wf-1"].pass_all_k).toBe(1);
-        expect(result.byWorkflow["wf-1"].tasks).toBe(2);
-        expect(result.byWorkflow["wf-1"].k).toBe(3);
-        expect(result.corpus.pass_at_k).toBe(1);
-        expect(result.corpus.pass_all_k).toBe(1);
-        expect(result.corpus.groups).toBe(2);
-        expect(result.corpus.tasks).toBe(2);
-    });
-    test("none-pass: no seed is pass → pass@k=0, pass^k=0", () => {
-        const checks = [
-            ...statuses("wf-1", "t1", ["fail", "fail", "fail"]),
-            ...statuses("wf-1", "t2", ["partial", "fail", "harness_error"]),
-        ];
-        const result = computeWorkflowReliability(checks);
-        expect(result.byWorkflow["wf-1"].pass_at_k).toBe(0);
-        expect(result.byWorkflow["wf-1"].pass_all_k).toBe(0);
-        expect(result.corpus.pass_at_k).toBe(0);
-        expect(result.corpus.pass_all_k).toBe(0);
-    });
-    test("some-pass: pass@k > 0, pass^k < pass@k when seeds disagree per task", () => {
-        // t1: 1 pass, 2 fail → counts toward pass@k (anyPass) but NOT pass^k
-        // t2: 3 pass → counts toward both
-        const checks = [
-            ...statuses("wf-1", "t1", ["pass", "fail", "fail"]),
-            ...statuses("wf-1", "t2", ["pass", "pass", "pass"]),
-        ];
-        const result = computeWorkflowReliability(checks);
-        expect(result.byWorkflow["wf-1"].pass_at_k).toBeCloseTo(1); // both tasks have at least one pass
-        expect(result.byWorkflow["wf-1"].pass_all_k).toBeCloseTo(0.5); // only t2 is all-pass
-        expect(result.corpus.pass_at_k).toBeCloseTo(1);
-        expect(result.corpus.pass_all_k).toBeCloseTo(0.5);
-    });
-    test("mixed partial/fail: partial does NOT count as pass for reliability", () => {
-        // partial is non-pass per the strict reliability bucketing.
-        const checks = [...statuses("wf-1", "t1", ["partial", "partial", "partial"])];
-        const result = computeWorkflowReliability(checks);
-        expect(result.byWorkflow["wf-1"].pass_at_k).toBe(0);
-        expect(result.byWorkflow["wf-1"].pass_all_k).toBe(0);
-    });
-    test("not_applicable seeds are excluded from numerator and denominator", () => {
-        // Only the 2 applicable seeds matter; both are pass → 1 task all-pass.
-        const checks = [...statuses("wf-1", "t1", ["not_applicable", "pass", "pass", "not_applicable"])];
-        const result = computeWorkflowReliability(checks);
-        expect(result.byWorkflow["wf-1"].pass_at_k).toBe(1);
-        expect(result.byWorkflow["wf-1"].pass_all_k).toBe(1);
-        expect(result.byWorkflow["wf-1"].tasks).toBe(1);
-        expect(result.byWorkflow["wf-1"].k).toBe(2);
-    });
-    test("workflow with every check not_applicable is omitted (no group counted)", () => {
-        const checks = [...statuses("wf-skips", "t1", ["not_applicable", "not_applicable"])];
-        const result = computeWorkflowReliability(checks);
-        expect(result.byWorkflow["wf-skips"]).toBeUndefined();
-        expect(result.corpus.groups).toBe(0);
-        expect(result.corpus.tasks).toBe(0);
-    });
-    test("multiple workflows compute independently; corpus weights groups equally", () => {
-        // wf-a: t1 all pass (1/1 = 1, 1/1 = 1)
-        // wf-b: t2 mixed (anyPass=1, allPass=0); t3 none (0, 0)
-        //   per-workflow: pass@k=0.5, pass^k=0
-        // corpus: 3 groups → pass@k = (1+1+0)/3 = 2/3; pass^k = (1+0+0)/3 = 1/3
-        const checks = [
-            ...statuses("wf-a", "t1", ["pass", "pass"]),
-            ...statuses("wf-b", "t2", ["pass", "fail"]),
-            ...statuses("wf-b", "t3", ["fail", "fail"]),
-        ];
-        const result = computeWorkflowReliability(checks);
-        expect(result.byWorkflow["wf-a"].pass_at_k).toBe(1);
-        expect(result.byWorkflow["wf-a"].pass_all_k).toBe(1);
-        expect(result.byWorkflow["wf-b"].pass_at_k).toBeCloseTo(0.5);
-        expect(result.byWorkflow["wf-b"].pass_all_k).toBe(0);
-        expect(result.corpus.pass_at_k).toBeCloseTo(2 / 3);
-        expect(result.corpus.pass_all_k).toBeCloseTo(1 / 3);
-        expect(result.corpus.groups).toBe(3);
-        expect(result.corpus.tasks).toBe(3);
-    });
-    test("harness_error is treated as non-pass (consistent with #257 bucketing)", () => {
-        const checks = [...statuses("wf-1", "t1", ["pass", "harness_error", "pass"])];
-        const result = computeWorkflowReliability(checks);
-        expect(result.byWorkflow["wf-1"].pass_at_k).toBe(1);
-        expect(result.byWorkflow["wf-1"].pass_all_k).toBe(0);
-    });
-});
-describe("computeLearningCurve", () => {
-    test("monotonic improvement: positive slope, time_to_improvement at first crossing", () => {
-        const episodes = [
-            ep({ episode_index: 0, pass_rate: 0.4, cumulative_feedback_events: 10 }),
-            ep({ episode_index: 1, pass_rate: 0.5, cumulative_feedback_events: 22 }),
-            ep({ episode_index: 2, pass_rate: 0.6, cumulative_feedback_events: 35 }),
-            ep({ episode_index: 3, pass_rate: 0.7, cumulative_feedback_events: 48 }),
-        ];
-        const curve = computeLearningCurve(episodes);
-        expect(curve.pass_rate_by_episode).toEqual([0.4, 0.5, 0.6, 0.7]);
-        // Slope is exactly 0.1 per episode for evenly spaced 0.1 increments.
-        expect(curve.learning_slope).toBeCloseTo(0.1, 6);
-        // Episode 1 first exceeds 0.4 + 0.05 = 0.45 (0.5 > 0.45).
-        expect(curve.time_to_improvement).toBe(1);
-        // Deltas: 0, 0.1, 0.1, 0.1
-        expect(curve.episodes[0].delta_from_previous_episode).toBe(0);
-        expect(curve.episodes[1].delta_from_previous_episode).toBeCloseTo(0.1);
-        expect(curve.episodes[3].delta_from_previous_episode).toBeCloseTo(0.1);
-    });
-    test("no improvement: flat pass rate yields zero slope and null time_to_improvement", () => {
-        const episodes = [
-            ep({ episode_index: 0, pass_rate: 0.5 }),
-            ep({ episode_index: 1, pass_rate: 0.5 }),
-            ep({ episode_index: 2, pass_rate: 0.51 }),
-            ep({ episode_index: 3, pass_rate: 0.52 }),
-        ];
-        const curve = computeLearningCurve(episodes);
-        expect(curve.learning_slope).toBeCloseTo(0.0073, 3);
-        // Never crosses 0.5 + 0.05 = 0.55.
-        expect(curve.time_to_improvement).toBeNull();
-    });
-    test("regression mid-episode: slope still computed, time_to_improvement honours first qualifying episode", () => {
-        // Pass rate climbs, then regresses below baseline+threshold, then recovers.
-        const episodes = [
-            ep({ episode_index: 0, pass_rate: 0.4 }),
-            ep({ episode_index: 1, pass_rate: 0.6 }), // > 0.45 → first crossing
-            ep({ episode_index: 2, pass_rate: 0.42 }), // mid-episode regression
-            ep({ episode_index: 3, pass_rate: 0.55 }),
-        ];
-        const curve = computeLearningCurve(episodes);
-        // First crossing wins, even though episode 2 regresses below threshold.
-        expect(curve.time_to_improvement).toBe(1);
-        expect(curve.episodes[2].delta_from_previous_episode).toBeCloseTo(-0.18);
-        // Slope is computed across all four points; should be positive overall.
-        expect(curve.learning_slope).toBeGreaterThan(0);
-    });
-    test("single-episode degenerate input: slope is 0, time_to_improvement is null", () => {
-        const curve = computeLearningCurve([ep({ episode_index: 0, pass_rate: 0.7 })]);
-        expect(curve.pass_rate_by_episode).toEqual([0.7]);
-        expect(curve.learning_slope).toBe(0);
-        expect(curve.time_to_improvement).toBeNull();
-        // Episode 0's delta is always 0 by definition.
-        expect(curve.episodes[0].delta_from_previous_episode).toBe(0);
-    });
-    test("empty input is degenerate: empty arrays, zero slope, null time", () => {
-        const curve = computeLearningCurve([]);
-        expect(curve.episodes).toEqual([]);
-        expect(curve.pass_rate_by_episode).toEqual([]);
-        expect(curve.learning_slope).toBe(0);
-        expect(curve.time_to_improvement).toBeNull();
-    });
-    test("unsorted input is sorted by episode_index before processing", () => {
-        const episodes = [
-            ep({ episode_index: 2, pass_rate: 0.6 }),
-            ep({ episode_index: 0, pass_rate: 0.4 }),
-            ep({ episode_index: 1, pass_rate: 0.5 }),
-        ];
-        const curve = computeLearningCurve(episodes);
-        expect(curve.episodes.map((e) => e.episode_index)).toEqual([0, 1, 2]);
-        expect(curve.pass_rate_by_episode).toEqual([0.4, 0.5, 0.6]);
-        expect(curve.time_to_improvement).toBe(1);
-    });
-    test("delta_from_previous_episode is recomputed defensively from sorted pass_rates", () => {
-        // Caller stamps wrong deltas — function recomputes.
-        const episodes = [
-            ep({ episode_index: 0, pass_rate: 0.4, delta_from_previous_episode: 99 }),
-            ep({ episode_index: 1, pass_rate: 0.6, delta_from_previous_episode: -42 }),
-        ];
-        const curve = computeLearningCurve(episodes);
-        expect(curve.episodes[0].delta_from_previous_episode).toBe(0);
-        expect(curve.episodes[1].delta_from_previous_episode).toBeCloseTo(0.2);
-    });
-    test("threshold is strictly greater-than (exact baseline+threshold does not count)", () => {
-        // baseline = 0.5; threshold = 0.05 → must exceed 0.55.
-        const episodes = [
-            ep({ episode_index: 0, pass_rate: 0.5 }),
-            ep({ episode_index: 1, pass_rate: 0.5 + LEARNING_IMPROVEMENT_THRESHOLD }), // 0.55 exactly
-            ep({ episode_index: 2, pass_rate: 0.5 + LEARNING_IMPROVEMENT_THRESHOLD + 0.001 }),
-        ];
-        const curve = computeLearningCurve(episodes);
-        expect(curve.time_to_improvement).toBe(2);
-    });
-    test("cumulative counters are echoed verbatim (caller-provided)", () => {
-        const episodes = [
-            ep({
-                episode_index: 0,
-                pass_rate: 0.4,
-                cumulative_feedback_events: 10,
-                cumulative_proposals_created: 0,
-                cumulative_proposals_accepted: 0,
-                cumulative_lessons_created: 0,
-                lesson_reuse_rate: null,
-            }),
-            ep({
-                episode_index: 1,
-                pass_rate: 0.55,
-                cumulative_feedback_events: 25,
-                cumulative_proposals_created: 4,
-                cumulative_proposals_accepted: 3,
-                cumulative_lessons_created: 3,
-                lesson_reuse_rate: 0.42,
-            }),
-        ];
-        const curve = computeLearningCurve(episodes);
-        expect(curve.episodes[1].cumulative_feedback_events).toBe(25);
-        expect(curve.episodes[1].cumulative_proposals_accepted).toBe(3);
-        expect(curve.episodes[1].cumulative_lessons_created).toBe(3);
-        expect(curve.episodes[1].lesson_reuse_rate).toBeCloseTo(0.42);
-    });
-});
-// ── #271: masked-stash path-traversal hardening ─────────────────────────────
-describe("materialiseMaskedStash stashName containment (#271)", () => {
-    function makeFixturesRoot() {
-        const fixturesRoot = benchMkdtemp("akm-bench-fixtures-");
-        // Plant a sibling outside fixturesRoot that a traversal-shaped stashName
-        // could otherwise reach. The MANIFEST.json existence check would pass
-        // there if containment were not enforced.
-        const sibling = path.join(path.dirname(fixturesRoot), `sibling-${path.basename(fixturesRoot)}`);
-        fs.mkdirSync(sibling, { recursive: true });
-        fs.writeFileSync(path.join(sibling, "MANIFEST.json"), "{}");
-        return {
-            fixturesRoot,
-            cleanup: () => {
-                fs.rmSync(fixturesRoot, { recursive: true, force: true });
-                fs.rmSync(sibling, { recursive: true, force: true });
-            },
-        };
-    }
-    test("rejects stashName starting with '..' (relative traversal)", () => {
-        const { fixturesRoot, cleanup } = makeFixturesRoot();
-        try {
-            const sibling = `../sibling-${path.basename(fixturesRoot)}`;
-            const result = materialiseMaskedStash(fixturesRoot, sibling, "skill:foo");
-            expect(result).toBeNull();
-        }
-        finally {
-            cleanup();
-        }
-    });
-    test("rejects absolute stashName", () => {
-        const { fixturesRoot, cleanup } = makeFixturesRoot();
-        try {
-            const result = materialiseMaskedStash(fixturesRoot, "/etc", "skill:foo");
-            expect(result).toBeNull();
-        }
-        finally {
-            cleanup();
-        }
-    });
-    test("rejects nested traversal that would escape fixturesRoot", () => {
-        const { fixturesRoot, cleanup } = makeFixturesRoot();
-        try {
-            // path.resolve(fixturesRoot, "a/../../sibling-xyz") would land on
-            // the sibling directory if containment is not enforced.
-            const escapePath = `a/../../sibling-${path.basename(fixturesRoot)}`;
-            const result = materialiseMaskedStash(fixturesRoot, escapePath, "skill:foo");
-            expect(result).toBeNull();
-        }
-        finally {
-            cleanup();
-        }
-    });
-    test("returns null (not a crash) for a contained stashName with no MANIFEST", () => {
-        const { fixturesRoot, cleanup } = makeFixturesRoot();
-        try {
-            // 'inner' is inside fixturesRoot but has no MANIFEST.json. Containment
-            // passes; the existing MANIFEST gate returns null. Sanity check that
-            // the new containment check did not accidentally reject the happy path.
-            fs.mkdirSync(path.join(fixturesRoot, "inner"));
-            const result = materialiseMaskedStash(fixturesRoot, "inner", "skill:foo");
-            expect(result).toBeNull();
-        }
-        finally {
-            cleanup();
-        }
-    });
-});
-describe("isPathContained symlink resolution (#271)", () => {
-    test("rejects a symlink inside root that points outside (alignment with isWithin)", () => {
-        const tmpRoot = benchMkdtemp("akm-bench-contain-root-");
-        const outside = benchMkdtemp("akm-bench-contain-outside-");
-        try {
-            // The actual file lives outside `tmpRoot`. Without realpath alignment,
-            // path.resolve(tmpRoot, "escape") looks contained ('escape' is just a
-            // basename) and the masking heuristic would happily rmSync it.
-            const escapeTarget = path.join(outside, "victim");
-            fs.writeFileSync(escapeTarget, "do-not-delete");
-            const symlinkPath = path.join(tmpRoot, "escape");
-            try {
-                fs.symlinkSync(escapeTarget, symlinkPath);
-            }
-            catch (err) {
-                // Some sandboxes (e.g. Windows w/o dev mode) deny symlink creation —
-                // skip rather than fail in those environments.
-                if (process.platform === "win32")
-                    return;
-                throw err;
-            }
-            // Without #271 alignment: rel === "escape" (contained). With
-            // safeRealpath: target resolves to `outside/victim` → rel starts with
-            // "..".
-            expect(isPathContained(tmpRoot, symlinkPath)).toBe(false);
-            // The victim file must still exist after the rejection (sanity check
-            // for the test fixture, not the function under test).
-            expect(fs.existsSync(escapeTarget)).toBe(true);
-        }
-        finally {
-            fs.rmSync(tmpRoot, { recursive: true, force: true });
-            fs.rmSync(outside, { recursive: true, force: true });
-        }
-    });
-    test("accepts a symlink inside root that points back inside root", () => {
-        const tmpRoot = benchMkdtemp("akm-bench-contain-inside-");
-        try {
-            const realFile = path.join(tmpRoot, "real");
-            fs.writeFileSync(realFile, "ok");
-            const linkPath = path.join(tmpRoot, "link");
-            try {
-                fs.symlinkSync(realFile, linkPath);
-            }
-            catch (err) {
-                if (process.platform === "win32")
-                    return;
-                throw err;
-            }
-            expect(isPathContained(tmpRoot, linkPath)).toBe(true);
-        }
-        finally {
-            fs.rmSync(tmpRoot, { recursive: true, force: true });
-        }
-    });
-    test("accepts a non-existent child path under root (covers safeRealpath ancestor walk)", () => {
-        const tmpRoot = benchMkdtemp("akm-bench-contain-pending-");
-        try {
-            const pending = path.join(tmpRoot, "not-yet-created", "child.md");
-            expect(isPathContained(tmpRoot, pending)).toBe(true);
-        }
-        finally {
-            fs.rmSync(tmpRoot, { recursive: true, force: true });
-        }
-    });
-    test("rejects an absolute target outside root", () => {
-        const tmpRoot = benchMkdtemp("akm-bench-contain-abs-");
-        const outside = benchMkdtemp("akm-bench-contain-abs-outside-");
-        try {
-            expect(isPathContained(tmpRoot, path.join(outside, "x"))).toBe(false);
-        }
-        finally {
-            fs.rmSync(tmpRoot, { recursive: true, force: true });
-            fs.rmSync(outside, { recursive: true, force: true });
-        }
-    });
-});