@opensearch-project/agent-health 0.6.0 → 0.7.0

package/lib/dist/lib/runStats.js

@@ -2,6 +2,11 @@
  * Copyright OpenSearch Contributors
  * SPDX-License-Identifier: Apache-2.0
  */
+/** Run statuses after which no further per-case progress can happen. */
+export const TERMINAL_RUN_STATUSES = new Set(['completed', 'failed', 'cancelled']);
+export function isTerminalRunStatus(status) {
+    return status !== undefined && TERMINAL_RUN_STATUSES.has(status);
+}
 /**
  * Bucket a run's per-test-case results into passed/failed/errored/pending using
  * ONLY the persisted result fields (status + passFailStatus) — no reports
@@ -18,16 +23,46 @@
  * the evaluator couldn't produce one (judge validation error, trace timeout).
  * Excluded from passed/failed — exactly as calculateRunStats does via the
  * report's metricsStatus.
+ *
+ * `plannedTotal` (bug: runs list showed no in-flight indication, 2026-09-01):
+ * for a genuinely in-progress run, `results` only gets an entry once a test
+ * case has actually STARTED — a run 9 cases into a planned 62 has a
+ * `results` map of length 9, so `total` here would report 9, not 62. Pass
+ * the run's snapshotted test-case count (`testCaseSnapshots.length`) as
+ * `plannedTotal` and the shortfall (planned - observed) is folded into
+ * `pending` so the invariant `total === passed+failed+errored+pending+notRun`
+ * still holds and callers see the true "53 more to go" total instead of a
+ * misleadingly small, already-100%-accounted-for total that looks finished.
+ *
+ * `runStatus` (bug: cancelled/failed runs rendered a phantom "n pending ⟳"
+ * forever, 2026-09-04): the shortfall above is only *pending* while the run
+ * can still make progress. Once the run is TERMINAL (cancelled / failed /
+ * completed) nothing will ever start those cases, so they are bucketed as
+ * `notRun` instead — a cancelled 34/62 run reads "34 executed · 28 not run",
+ * not "28 pending" with a spinner. The same applies to per-case entries a
+ * terminal run left behind in `pending`/`running` (an executor that died
+ * mid-case) and to entries explicitly marked `status: 'cancelled'` (written
+ * for never-started cases when a run is cancelled — see
+ * `EvaluationRun.results` in types/index.ts). Without `runStatus` the legacy
+ * behaviour (shortfall = pending) is preserved.
  */
-export function bucketRunResults(results) {
-    let passed = 0, failed = 0, errored = 0, pending = 0, total = 0;
+export function bucketRunResults(results, plannedTotal, runStatus) {
+    const terminal = isTerminalRunStatus(runStatus);
+    let passed = 0, failed = 0, errored = 0, pending = 0, notRun = 0, total = 0;
     for (const r of Object.values(results || {})) {
         total++;
+        if (r.status === 'cancelled') {
+            notRun++;
+            continue;
+        }
         if (r.status === 'pending' || r.status === 'running') {
-            pending++;
+            if (terminal)
+                notRun++;
+            else
+                pending++;
             continue;
         }
-        if (r.status === 'failed' || r.status === 'cancelled') {
+        if (r.status === 'failed') {
             failed++;
             continue;
         }
@@ -40,9 +75,39 @@ export function bucketRunResults(results) {
                 errored++; // completed without a verdict = judge errored (#242)
             continue;
         }
-        pending++; // unknown / no status
+        // Unknown / missing status: while the run is live this is "not settled
+        // yet" (pending). On a TERMINAL run an entry with no recognised status is
+        // schema drift or a bad writer — surface it as `errored` (amber ⚠) rather
+        // than hide it in the neutral notRun bucket (codex review).
+        if (terminal)
+            errored++;
+        else
+            pending++;
     }
-    return { passed, failed, errored, pending, total };
+    if (plannedTotal !== undefined && plannedTotal > total) {
+        if (terminal)
+            notRun += plannedTotal - total;
+        else
+            pending += plannedTotal - total;
+        total = plannedTotal;
+    }
+    return { passed, failed, errored, pending, notRun, total };
+}
+/**
+ * Pass rate (0-100) over the JUDGED set only: `passed / (passed + failed)`.
+ * Excluded: errored (executed, but the evaluator produced no verdict — the
+ * repo-wide #242 convention, same as the benchmark runs table and the run
+ * inspector), pending (not finished) and notRun (never started). So a run
+ * cancelled at 34/62 reports the pass rate of the cases that were actually
+ * judged, and an in-flight run doesn't read as "12%" because 50 cases
+ * haven't happened yet. Named for what it is — callers that surface it
+ * should footnote the excluded errored/notRun counts (the detail page does).
+ * Returns `null` when nothing has been judged so callers can render "—"
+ * rather than a fabricated 0%.
+ */
+export function passRateOverJudged(stats) {
+    const judged = stats.passed + stats.failed;
+    return judged > 0 ? Math.round((stats.passed / judged) * 100) : null;
 }
 /**
  * Calculate statistics for a benchmark run.
@@ -61,6 +126,7 @@ export function calculateRunStats(run, reports) {
     let failed = 0;
     let pending = 0;
     let errored = 0;
+    let notRun = 0;
     let total = 0;
     Object.entries(run.results || {}).forEach(([testCaseId, result]) => {
         total++;
@@ -69,7 +135,12 @@ export function calculateRunStats(run, reports) {
             pending++;
             return;
         }
-        if (result.status === 'failed' || result.status === 'cancelled') {
+        // Never started (run cancelled before reaching it) — not a failure.
+        if (result.status === 'cancelled') {
+            notRun++;
+            return;
+        }
+        if (result.status === 'failed') {
             failed++;
             return;
         }
@@ -107,15 +178,16 @@ export function calculateRunStats(run, reports) {
             pending++;
         }
     });
-    // Pass rate is computed over the evaluable set (total minus errored).
-    // If every run errored, expose 0% rather than dividing by zero.
-    const evaluable = Math.max(0, total - errored);
+    // Pass rate is computed over the evaluable set (total minus errored minus
+    // never-run). If every run errored, expose 0% rather than dividing by zero.
+    const evaluable = Math.max(0, total - errored - notRun);
     const passRate = evaluable > 0 ? Math.round((passed / evaluable) * 100) : 0;
     return {
         passed,
         failed,
         pending,
         errored,
+        notRun,
         total,
         passRate,
     };
@@ -161,19 +233,67 @@ export function getReportIdsFromRun(run) {
  * simply ignore the extra `pending` field.
  */
 export function computeRunStats(run) {
+    const plannedTotal = run.testCaseSnapshots?.length;
+    // Terminal-aware bucketing (see bucketRunResults): the planned-but-absent
+    // remainder is `pending` only while the run is still running. Only the
+    // EXPLICIT persisted status is trusted here — the results-derived legacy
+    // fallback in getEffectiveRunStatus reports 'completed' for any status-less
+    // run whose observed results happen to all be settled, which is exactly
+    // what an in-flight legacy run looks like between two cases.
+    const effectiveStatus = run.status;
     if (run.results && Object.keys(run.results).length > 0) {
-        return bucketRunResults(run.results);
+        const bucketed = bucketRunResults(run.results, plannedTotal, effectiveStatus);
+        // Legacy-shape fallback (goyamegh/pr-run-report-v2, run-list all-errored bug):
+        // runs created before per-result passFailStatus denormalization landed
+        // (evaluationRunner.ts writing `results[testCaseId].passFailStatus`) only
+        // persisted `{ reportId, status: 'completed' }` on each result -- no
+        // verdict field at all. bucketRunResults correctly treats a 'completed'
+        // result with no passFailStatus as `errored` (#242), which is right for a
+        // genuine judge failure, but wrong here: EVERY case in the run buckets as
+        // errored (0 passed, 0 failed) even though the run actually has real
+        // verdicts. Detect that specific signature and, only when the
+        // denormalized `run.stats` independently shows real pass/fail evidence
+        // (computed by the report-fetching `computeStatsForRun`, verified against
+        // linked reports' actual passFailStatus -- see tests/unit/lib/runStats.test.ts
+        // 'legacy run shape' cases), trust `run.stats` instead of asserting the
+        // whole run errored. A run that is genuinely all-errored (no result ever
+        // resolved a verdict) will have run.stats.passed === run.stats.failed === 0
+        // too, so it still falls through to the errored bucketing below.
+        const allBucketedAsErrored = bucketed.errored > 0 && bucketed.passed === 0 && bucketed.failed === 0;
+        const statsHaveRealVerdicts = !!run.stats && ((run.stats.passed ?? 0) > 0 || (run.stats.failed ?? 0) > 0);
+        if (allBucketedAsErrored && statsHaveRealVerdicts && run.stats) {
+            return statsFallback(run.stats, run.stats.total ?? bucketed.total, effectiveStatus);
+        }
+        return bucketed;
     }
     if (run.stats && (run.stats.total ?? 0) > 0) {
-        return {
-            passed: run.stats.passed ?? 0,
-            failed: run.stats.failed ?? 0,
-            errored: run.stats.errored ?? 0,
-            pending: run.stats.pending ?? 0,
-            total: run.stats.total ?? 0,
-        };
+        return statsFallback(run.stats, run.stats.total ?? 0, effectiveStatus);
     }
-    return { passed: 0, failed: 0, errored: 0, pending: 0, total: 0 };
+    // No results and no stats: a run that never started a case. Terminal →
+    // every planned case is "not run"; otherwise they're all still pending.
+    const planned = plannedTotal || 0;
+    return isTerminalRunStatus(effectiveStatus)
+        ? { passed: 0, failed: 0, errored: 0, pending: 0, notRun: planned, total: planned }
+        : { passed: 0, failed: 0, errored: 0, pending: planned, notRun: 0, total: planned };
+}
+/**
+ * Denormalized `run.stats` fallback, made terminal-aware: stats persisted
+ * before `notRun` existed carry the never-started remainder in `pending`
+ * even for a cancelled run. Re-home that remainder so a legacy cancelled
+ * run can't render a phantom spinner either.
+ */
+function statsFallback(stats, total, effectiveStatus) {
+    const pending = stats.pending ?? 0;
+    const notRun = stats.notRun ?? 0;
+    const terminal = isTerminalRunStatus(effectiveStatus);
+    return {
+        passed: stats.passed ?? 0,
+        failed: stats.failed ?? 0,
+        errored: stats.errored ?? 0,
+        pending: terminal ? 0 : pending,
+        notRun: terminal ? notRun + pending : notRun,
+        total,
+    };
 }
 export function computeRunStatsFromReports(run, reports) {
     const reportsMap = {};
@@ -186,7 +306,38 @@ export function computeRunStatsFromReports(run, reports) {
         failed: fullStats.failed,
         pending: fullStats.pending,
         errored: fullStats.errored,
+        notRun: fullStats.notRun,
         total: fullStats.total,
     };
 }
+/**
+ * Single canonical "is this run actively in progress" check, shared by every
+ * runs-list surface (bug: the Evaluation Runs page and the benchmark-scoped
+ * Runs panel each grew their own copy of this, and only one of them actually
+ * rendered a running indicator). `status` is authoritative when present
+ * (always true for `EvaluationRun`; `BenchmarkRun.status` is only undefined
+ * for legacy pre-status data). Falls back to inspecting `results` so legacy
+ * runs without a top-level `status` still get a sensible effective status.
+ */
+export function getEffectiveRunStatus(run) {
+    if (run.status)
+        return run.status;
+    const results = Object.values(run.results || {});
+    if (results.some(r => r.status === 'running'))
+        return 'running';
+    if (results.some(r => r.status === 'pending') &&
+        !results.some(r => r.status === 'completed' || r.status === 'failed'))
+        return 'running';
+    if (results.some(r => r.status === 'completed') || results.some(r => r.status === 'failed'))
+        return 'completed';
+    return 'failed';
+}
+/**
+ * True when a run has neither reached a terminal status nor accounted for
+ * every planned test case yet. Used to decide whether a runs-list page
+ * should keep polling for live updates.
+ */
+export function isRunInProgress(run) {
+    return getEffectiveRunStatus(run) === 'running';
+}
 //# sourceMappingURL=runStats.js.map

package/lib/dist/lib/runStats.js.map

@@ -1 +1 @@
-{"version":3,"file":"runStats.js","sourceRoot":"","sources":["../../runStats.ts"],"names":[],"mappings":"AAAA;;;GAGG;AA8CH;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,gBAAgB,CAC9B,OAAiF;IAEjF,IAAI,MAAM,GAAG,CAAC,EAAE,MAAM,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC;IAChE,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC,EAAE,CAAC;QAC7C,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YAAC,OAAO,EAAE,CAAC;YAAC,SAAS;QAAC,CAAC;QAC9E,IAAI,CAAC,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAAC,MAAM,EAAE,CAAC;YAAC,SAAS;QAAC,CAAC;QAC9E,IAAI,CAAC,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAC7B,IAAI,CAAC,CAAC,cAAc,KAAK,QAAQ;gBAAE,MAAM,EAAE,CAAC;iBACvC,IAAI,CAAC,CAAC,cAAc,KAAK,QAAQ;gBAAE,MAAM,EAAE,CAAC;;gBAC5C,OAAO,EAAE,CAAC,CAAC,qDAAqD;YACrE,SAAS;QACX,CAAC;QACD,OAAO,EAAE,CAAC,CAAC,sBAAsB;IACnC,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;AACrD,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,iBAAiB,CAC/B,GAAiB,EACjB,OAAgD;IAEhD,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,KAAK,GAAG,CAAC,CAAC;IAEd,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,EAAE,MAAM,CAAC,EAAE,EAAE;QACjE,KAAK,EAAE,CAAC;QAER,4BAA4B;QAC5B,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YAC/D,OAAO,EAAE,CAAC;YACV,OAAO;QACT,CAAC;QAED,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAChE,MAAM,EAAE,CAAC;YACT,OAAO;QACT,CAAC;QAED,2DAA2D;QAC3D,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACrD,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;YAExC,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,yCAAyC;gBACzC,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,oDAAoD;YACpD,IAAI,MAAM,CAAC,aAAa,KAAK,SAAS,IAAI,MAAM,CAAC,aAAa,KAAK,aAAa,EAAE,CAAC;gBACjF,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,sEAAsE;YACtE,8DAA8D;YAC9D,+DAA+D;YAC/D,IAAI,MAAM,CAAC,aAAa,KAAK,OAAO,EAAE,CAAC;gBACrC,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,+CAA+C;YAC/C,IAAI,MAAM,CAAC,cAAc,KAAK,QAAQ,EAAE,CAAC;gBACvC,MAAM,EAAE,CAAC;YACX,CAAC;iBAAM,CAAC;gBACN,6DAA6D;gBAC7D,MAAM,EAAE,CAAC;YACX,CAAC;QACH,CAAC;aAAM,CAAC;YACN,yDAAyD;YACzD,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,sEAAsE;IACtE,gEAAgE;IAChE,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,GAAG,OAAO,CAAC,CAAC;IAC/C,MAAM,QAAQ,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,GAAG,SAAS,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAE5E,OAAO;QACL,MAAM;QACN,MAAM;QACN,OAAO;QACP,OAAO;QACP,KAAK;QACL,QAAQ;KACT,CAAC;AACJ,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CAAC,GAAiB;IACnD,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;IAEpC,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE;QAClD,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACpB,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QACjC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,OAAO,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;AAC/B,CAAC;AAED;;;;;;;GAOG;AACH;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,eAAe,CAC7B,GAGC;IAED,IAAI,GAAG,CAAC,OAAO,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvD,OAAO,gBAAgB,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IACvC,CAAC;IACD,IAAI,GAAG,CAAC,KAAK,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC;QAC5C,OAAO;YACL,MAAM,EAAE,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC;YAC7B,MAAM,EAAE,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC;YAC7B,OAAO,EAAE,GAAG,CAAC,KAAK,CAAC,OAAO,IAAI,CAAC;YAC/B,OAAO,EAAE,GAAG,CAAC,KAAK,CAAC,OAAO,IAAI,CAAC;YAC/B,KAAK,EAAE,GAAG,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC;SAC5B,CAAC;IACJ,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;AACpE,CAAC;AAED,MAAM,UAAU,0BAA0B,CACxC,GAAiB,EACjB,OAA2B;IAE3B,MAAM,UAAU,GAA4C,EAAE,CAAC;IAC/D,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE;QACvB,UAAU,CAAC,MAAM,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,iBAAiB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;IAErD,OAAO;QACL,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,KAAK,EAAE,SAAS,CAAC,KAAK;KACvB,CAAC;AACJ,CAAC"}
+{"version":3,"file":"runStats.js","sourceRoot":"","sources":["../../runStats.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAuDH,wEAAwE;AACxE,MAAM,CAAC,MAAM,qBAAqB,GAAoC,IAAI,GAAG,CAAC,CAAC,WAAW,EAAE,QAAQ,EAAE,WAAW,CAAC,CAAC,CAAC;AAEpH,MAAM,UAAU,mBAAmB,CAAC,MAAsC;IACxE,OAAO,MAAM,KAAK,SAAS,IAAI,qBAAqB,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;AACnE,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AACH,MAAM,UAAU,gBAAgB,CAC9B,OAAiF,EACjF,YAAqB,EACrB,SAA8B;IAE9B,MAAM,QAAQ,GAAG,mBAAmB,CAAC,SAAS,CAAC,CAAC;IAChD,IAAI,MAAM,GAAG,CAAC,EAAE,MAAM,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,MAAM,GAAG,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC;IAC5E,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC,EAAE,CAAC;QAC7C,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAAC,MAAM,EAAE,CAAC;YAAC,SAAS;QAAC,CAAC;QACrD,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YACrD,IAAI,QAAQ;gBAAE,MAAM,EAAE,CAAC;;gBAAM,OAAO,EAAE,CAAC;YACvC,SAAS;QACX,CAAC;QACD,IAAI,CAAC,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAAC,MAAM,EAAE,CAAC;YAAC,SAAS;QAAC,CAAC;QAClD,IAAI,CAAC,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAC7B,IAAI,CAAC,CAAC,cAAc,KAAK,QAAQ;gBAAE,MAAM,EAAE,CAAC;iBACvC,IAAI,CAAC,CAAC,cAAc,KAAK,QAAQ;gBAAE,MAAM,EAAE,CAAC;;gBAC5C,OAAO,EAAE,CAAC,CAAC,qDAAqD;YACrE,SAAS;QACX,CAAC;QACD,uEAAuE;QACvE,0EAA0E;QAC1E,0EAA0E;QAC1E,4DAA4D;QAC5D,IAAI,QAAQ;YAAE,OAAO,EAAE,CAAC;;YAAM,OAAO,EAAE,CAAC;IAC1C,CAAC;IACD,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY,GAAG,KAAK,EAAE,CAAC;QACvD,IAAI,QAAQ;YAAE,MAAM,IAAI,YAAY,GAAG,KAAK,CAAC;;YACxC,OAAO,IAAI,YAAY,GAAG,KAAK,CAAC;QACrC,KAAK,GAAG,YAAY,CAAC;IACvB,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;AAC7D,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,kBAAkB,CAAC,KAAyC;IAC1E,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC;IAC3C,OAAO,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,MAAM,GAAG,MAAM,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;AACvE,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,iBAAiB,CAC/B,GAAiB,EACjB,OAAgD;IAEhD,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,KAAK,GAAG,CAAC,CAAC;IAEd,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,EAAE,MAAM,CAAC,EAAE,EAAE;QACjE,KAAK,EAAE,CAAC;QAER,4BAA4B;QAC5B,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YAC/D,OAAO,EAAE,CAAC;YACV,OAAO;QACT,CAAC;QAED,oEAAoE;QACpE,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAClC,MAAM,EAAE,CAAC;YACT,OAAO;QACT,CAAC;QAED,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAC/B,MAAM,EAAE,CAAC;YACT,OAAO;QACT,CAAC;QAED,2DAA2D;QAC3D,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACrD,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;YAExC,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,yCAAyC;gBACzC,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,oDAAoD;YACpD,IAAI,MAAM,CAAC,aAAa,KAAK,SAAS,IAAI,MAAM,CAAC,aAAa,KAAK,aAAa,EAAE,CAAC;gBACjF,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,sEAAsE;YACtE,8DAA8D;YAC9D,+DAA+D;YAC/D,IAAI,MAAM,CAAC,aAAa,KAAK,OAAO,EAAE,CAAC;gBACrC,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,+CAA+C;YAC/C,IAAI,MAAM,CAAC,cAAc,KAAK,QAAQ,EAAE,CAAC;gBACvC,MAAM,EAAE,CAAC;YACX,CAAC;iBAAM,CAAC;gBACN,6DAA6D;gBAC7D,MAAM,EAAE,CAAC;YACX,CAAC;QACH,CAAC;aAAM,CAAC;YACN,yDAAyD;YACzD,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,0EAA0E;IAC1E,4EAA4E;IAC5E,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,GAAG,OAAO,GAAG,MAAM,CAAC,CAAC;IACxD,MAAM,QAAQ,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,GAAG,SAAS,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAE5E,OAAO;QACL,MAAM;QACN,MAAM;QACN,OAAO;QACP,OAAO;QACP,MAAM;QACN,KAAK;QACL,QAAQ;KACT,CAAC;AACJ,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CAAC,GAAiB;IACnD,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;IAEpC,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE;QAClD,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACpB,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QACjC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,OAAO,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;AAC/B,CAAC;AAED;;;;;;;GAOG;AACH;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,eAAe,CAC7B,GAKC;IAED,MAAM,YAAY,GAAG,GAAG,CAAC,iBAAiB,EAAE,MAAM,CAAC;IACnD,0EAA0E;IAC1E,uEAAuE;IACvE,yEAAyE;IACzE,4EAA4E;IAC5E,wEAAwE;IACxE,6DAA6D;IAC7D,MAAM,eAAe,GAAG,GAAG,CAAC,MAAM,CAAC;IACnC,IAAI,GAAG,CAAC,OAAO,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvD,MAAM,QAAQ,GAAG,gBAAgB,CAAC,GAAG,CAAC,OAAO,EAAE,YAAY,EAAE,eAAe,CAAC,CAAC;QAE9E,+EAA+E;QAC/E,uEAAuE;QACvE,0EAA0E;QAC1E,qEAAqE;QACrE,wEAAwE;QACxE,0EAA0E;QAC1E,0EAA0E;QAC1E,qEAAqE;QACrE,8DAA8D;QAC9D,uEAAuE;QACvE,0EAA0E;QAC1E,+EAA+E;QAC/E,wEAAwE;QACxE,yEAAyE;QACzE,4EAA4E;QAC5E,iEAAiE;QACjE,MAAM,oBAAoB,GAAG,QAAQ,CAAC,OAAO,GAAG,CAAC,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,CAAC;QACpG,MAAM,qBAAqB,GAAG,CAAC,CAAC,GAAG,CAAC,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAC1G,IAAI,oBAAoB,IAAI,qBAAqB,IAAI,GAAG,CAAC,KAAK,EAAE,CAAC;YAC/D,OAAO,aAAa,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,KAAK,CAAC,KAAK,IAAI,QAAQ,CAAC,KAAK,EAAE,eAAe,CAAC,CAAC;QACtF,CAAC;QAED,OAAO,QAAQ,CAAC;IAClB,CAAC;IACD,IAAI,GAAG,CAAC,KAAK,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC;QAC5C,OAAO,aAAa,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,EAAE,eAAe,CAAC,CAAC;IACzE,CAAC;IACD,uEAAuE;IACvE,wEAAwE;IACxE,MAAM,OAAO,GAAG,YAAY,IAAI,CAAC,CAAC;IAClC,OAAO,mBAAmB,CAAC,eAAe,CAAC;QACzC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE;QACnF,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;AACxF,CAAC;AAED;;;;;GAKG;AACH,SAAS,aAAa,CACpB,KAA4B,EAC5B,KAAa,EACb,eAA+C;IAE/C,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,IAAI,CAAC,CAAC;IACnC,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC;IACjC,MAAM,QAAQ,GAAG,mBAAmB,CAAC,eAAe,CAAC,CAAC;IACtD,OAAO;QACL,MAAM,EAAE,KAAK,CAAC,MAAM,IAAI,CAAC;QACzB,MAAM,EAAE,KAAK,CAAC,MAAM,IAAI,CAAC;QACzB,OAAO,EAAE,KAAK,CAAC,OAAO,IAAI,CAAC;QAC3B,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO;QAC/B,MAAM,EAAE,QAAQ,CAAC,CAAC,CAAC,MAAM,GAAG,OAAO,CAAC,CAAC,CAAC,MAAM;QAC5C,KAAK;KACN,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,0BAA0B,CACxC,GAAiB,EACjB,OAA2B;IAE3B,MAAM,UAAU,GAA4C,EAAE,CAAC;IAC/D,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE;QACvB,UAAU,CAAC,MAAM,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,iBAAiB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;IAErD,OAAO;QACL,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,KAAK,EAAE,SAAS,CAAC,KAAK;KACvB,CAAC;AACJ,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,qBAAqB,CACnC,GAAmF;IAEnF,IAAI,GAAG,CAAC,MAAM;QAAE,OAAO,GAAG,CAAC,MAAM,CAAC;IAClC,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;IACjD,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,SAAS,CAAC;QAAE,OAAO,SAAS,CAAC;IAChE,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,SAAS,CAAC;QACzC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,WAAW,IAAI,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC;QAAE,OAAO,SAAS,CAAC;IAC5F,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,WAAW,CAAC,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC;QAAE,OAAO,WAAW,CAAC;IAChH,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,eAAe,CAC7B,GAAmF;IAEnF,OAAO,qBAAqB,CAAC,GAAG,CAAC,KAAK,SAAS,CAAC;AAClD,CAAC"}

package/lib/dist/lib/testCases/define.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"define.d.ts","sourceRoot":"","sources":["../../../testCases/define.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EACV,YAAY,EACZ,WAAW,EACX,UAAU,EACV,MAAM,EAEN,cAAc,EACf,MAAM,YAAY,CAAC;AAiCpB,gBAAgB;AAChB,wBAAgB,yBAAyB,IAAI,IAAI,CAEhD;AAED,wBAAgB,aAAa,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAQpD;AAED;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAgB,IAAI,CAClB,IAAI,EAAE,MAAM,EACZ,IAAI,EAAE,CAAC,MAAM,EAAE,UAAU,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,GACjD,IAAI,CAAC;AACR,wBAAgB,IAAI,CAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,WAAW,EACpB,IAAI,EAAE,CAAC,MAAM,EAAE,UAAU,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,GACjD,IAAI,CAAC;AAuDR;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,IAAI,GAAG,IAAI,CAsB3D;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,YAAY,EAAE,CAGpE;AAED;;;;GAIG;AACH,wBAAgB,kBAAkB,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,cAAc,EAAE,CAGtE;AAED,wBAAgB,aAAa,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CASrD;AA4BD;;;;;;;;;;;;GAYG;AACH,wBAAgB,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAAmC;AAE9E;;;;;;;;;;;GAWG;AACH,wBAAgB,QAAQ,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAAkC;AAE5E;;;;;;;;;;;;;GAaG;AACH,wBAAgB,UAAU,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAAoC;AAEhF;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAAmC;AAS9E,yBAAiB,IAAI,CAAC;IAEb,MAAM,SAAS,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,IAA4C,CAAC;IAE9E,MAAM,QAAQ,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,IAA2C,CAAC;IAE5E,MAAM,UAAU,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,IAA6C,CAAC;IAEhF,MAAM,SAAS,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,IAA4C,CAAC;CACtF"}
+{"version":3,"file":"define.d.ts","sourceRoot":"","sources":["../../../testCases/define.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EACV,YAAY,EACZ,WAAW,EACX,UAAU,EACV,MAAM,EAEN,cAAc,EACf,MAAM,YAAY,CAAC;AAkFpB,gBAAgB;AAChB,wBAAgB,yBAAyB,IAAI,IAAI,CAEhD;AAED,wBAAgB,aAAa,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CASpD;AAED;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAgB,IAAI,CAClB,IAAI,EAAE,MAAM,EACZ,IAAI,EAAE,CAAC,MAAM,EAAE,UAAU,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,GACjD,IAAI,CAAC;AACR,wBAAgB,IAAI,CAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,WAAW,EACpB,IAAI,EAAE,CAAC,MAAM,EAAE,UAAU,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,GACjD,IAAI,CAAC;AAwDR;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,IAAI,GAAG,IAAI,CAuB3D;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,YAAY,EAAE,CAIpE;AAED;;;;GAIG;AACH,wBAAgB,kBAAkB,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,cAAc,EAAE,CAItE;AAED,wBAAgB,aAAa,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAUrD;AA6BD;;;;;;;;;;;;GAYG;AACH,wBAAgB,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAAmC;AAE9E;;;;;;;;;;;GAWG;AACH,wBAAgB,QAAQ,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAAkC;AAE5E;;;;;;;;;;;;;GAaG;AACH,wBAAgB,UAAU,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAAoC;AAEhF;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAAmC;AAS9E,yBAAiB,IAAI,CAAC;IAEb,MAAM,SAAS,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,IAA4C,CAAC;IAE9E,MAAM,QAAQ,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,IAA2C,CAAC;IAE5E,MAAM,UAAU,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,IAA6C,CAAC;IAEhF,MAAM,SAAS,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,IAA4C,CAAC;CACtF"}

package/lib/dist/lib/testCases/define.js

@@ -3,25 +3,65 @@
  * SPDX-License-Identifier: Apache-2.0
  */
 import { readEnv } from '../envCompat.js';
-const registries = new Map();
-// Hooks live in a parallel registry keyed by file path, exactly matching
-// the test() registry. The HookOrchestrator filters by describePath at
-// run time. Cross-file hooks never leak — the loader resets per file.
-const hookRegistries = new Map();
-let activeFile = null;
 const DEFAULT_KEY = '__default__';
-// Active describe stack (synchronous push/pop). Each entry is a describe
-// name; nested describes accumulate. When test() runs, it captures the
-// current stack and joins with ' > ' to derive the test's benchmarkPath.
-const describeStack = [];
-// Emit a one-time experimental-status warning the first time the SDK is
-// touched, unless the user opts out. This is intentionally noisy enough to be
-// noticed but only fires once per process so it doesn't pollute test output.
-let experimentalWarningEmitted = false;
+/**
+ * Registration state lives behind a `globalThis`-keyed symbol instead of
+ * plain module-level variables. Why: `.eval.js` files are `eval()`'d in a
+ * synthetic CJS context where `require('@opensearch-project/agent-health')`
+ * is intercepted by the loader and handed this same module's exports
+ * directly — but `.eval.ts` / `.eval.mjs` files are loaded via a plain
+ * native `import()`, which resolves `@opensearch-project/agent-health`
+ * through Node's normal module resolution (the package's `exports` map,
+ * e.g. `./lib/dist/lib/index.js`). When the host process itself runs from
+ * TypeScript *source* (dev CLI under tsx, or a project-local `node_modules`
+ * symlink pointing at this repo — both real setups), that resolution lands
+ * on a PHYSICALLY DIFFERENT file on disk than the one this module's own
+ * code imports internally. Two different files means two different ES
+ * module instances, each with its own module-level `Map` — so a `.eval.ts`
+ * file's `test()` calls landed in an orphaned registry the loader never
+ * read from, and `getRegisteredTests()` came back empty ("Module ... has no
+ * test cases").
+ *
+ * `Symbol.for(key)` returns the SAME symbol from the process-wide global
+ * symbol registry no matter which module instance asks for it (dist vs
+ * source, CJS vs ESM, symlinked vs installed) — so keying the shared state
+ * off `globalThis[Symbol.for(...)]` makes every instance of this module
+ * read and write the same underlying `Map`s, regardless of how many times /
+ * from where it gets loaded. This fixes the module-instance mismatch at the
+ * root instead of special-casing each way it can occur.
+ */
+const REGISTRY_KEY = Symbol.for('agent-health.test-registry.v1');
+function getSharedState() {
+    // Invariant this design relies on: eval files are loaded SEQUENTIALLY
+    // (setActiveFile -> synchronously evaluate one file's top-level test()/
+    // describe() calls -> move to the next), never concurrently, within a
+    // process. `activeFile` and `describeStack` are shared state now (not
+    // just per-module-instance as before this fix), so two files loading in
+    // parallel would race on them. Verified true for both real callers as of
+    // this fix: services/sourceResolver.ts's two loops and
+    // cli/commands/benchmark.ts's file-mode loader all `await
+    // loadTestCasesFromModule(...)` inside a plain `for` loop, never
+    // `Promise.all`. If a future caller parallelizes file loading, this
+    // invariant breaks and registration could scramble across files.
+    const g = globalThis;
+    let state = g[REGISTRY_KEY];
+    if (!state) {
+        state = {
+            registries: new Map(),
+            hookRegistries: new Map(),
+            activeFile: null,
+            describeStack: [],
+            experimentalWarningEmitted: false,
+        };
+        g[REGISTRY_KEY] = state;
+    }
+    return state;
+}
 function emitExperimentalWarningOnce() {
-    if (experimentalWarningEmitted)
+    const state = getSharedState();
+    if (state.experimentalWarningEmitted)
         return;
-    experimentalWarningEmitted = true;
+    state.experimentalWarningEmitted = true;
     if (readEnv('AH_SUPPRESS_EXPERIMENTAL', 'AGENT_HEALTH_SUPPRESS_EXPERIMENTAL') === '1')
         return;
     // eslint-disable-next-line no-console
@@ -32,15 +72,16 @@ function emitExperimentalWarningOnce() {
 }
 /** @internal */
 export function _resetExperimentalWarning() {
-    experimentalWarningEmitted = false;
+    getSharedState().experimentalWarningEmitted = false;
 }
 export function setActiveFile(filePath) {
-    activeFile = filePath;
-    if (!registries.has(filePath)) {
-        registries.set(filePath, []);
+    const state = getSharedState();
+    state.activeFile = filePath;
+    if (!state.registries.has(filePath)) {
+        state.registries.set(filePath, []);
     }
-    if (!hookRegistries.has(filePath)) {
-        hookRegistries.set(filePath, []);
+    if (!state.hookRegistries.has(filePath)) {
+        state.hookRegistries.set(filePath, []);
     }
 }
 export function test(name, optionsOrBody, maybeBody) {
@@ -62,17 +103,18 @@ export function test(name, optionsOrBody, maybeBody) {
     if (typeof evaluate !== 'function') {
         throw new Error(`test("${name}") requires a body function`);
     }
-    const key = activeFile ?? DEFAULT_KEY;
-    if (!registries.has(key)) {
-        registries.set(key, []);
+    const state = getSharedState();
+    const key = state.activeFile ?? DEFAULT_KEY;
+    if (!state.registries.has(key)) {
+        state.registries.set(key, []);
     }
-    const registry = registries.get(key);
+    const registry = state.registries.get(key);
     // Within-file uniqueness guard: a name+benchmarkPath pair must be unique.
     // The same name in two different describe blocks is allowed because they
     // map to different benchmarks.
-    const benchmarkPath = describeStack.length > 0 ? describeStack.join(' > ') : undefined;
+    const benchmarkPath = state.describeStack.length > 0 ? state.describeStack.join(' > ') : undefined;
     if (registry.some(t => t.name === name && t.benchmarkPath === benchmarkPath)) {
-        const fileLabel = activeFile ? ` in ${activeFile}` : '';
+        const fileLabel = state.activeFile ? ` in ${state.activeFile}` : '';
         const groupLabel = benchmarkPath ? ` (in describe "${benchmarkPath}")` : '';
         throw new Error(`Duplicate test name "${name}"${groupLabel}${fileLabel}. ` +
             `Test names must be unique within their describe block. ` +
@@ -82,7 +124,7 @@ export function test(name, optionsOrBody, maybeBody) {
         name,
         options,
         evaluate,
-        sourceFile: activeFile ?? undefined,
+        sourceFile: state.activeFile ?? undefined,
         benchmarkPath,
     });
 }
@@ -110,7 +152,8 @@ export function describe(name, fn) {
     if (typeof fn !== 'function') {
         throw new Error(`describe("${name}") requires a body function`);
     }
-    describeStack.push(name);
+    const state = getSharedState();
+    state.describeStack.push(name);
     try {
         const result = fn();
         if (result && typeof result.then === 'function') {
@@ -121,13 +164,14 @@ export function describe(name, fn) {
         }
     }
     finally {
-        describeStack.pop();
+        state.describeStack.pop();
     }
 }
 export function getRegisteredTests(filePath) {
+    const state = getSharedState();
     if (filePath)
-        return [...(registries.get(filePath) ?? [])];
-    return [...registries.values()].flatMap(r => [...r]);
+        return [...(state.registries.get(filePath) ?? [])];
+    return [...state.registries.values()].flatMap(r => [...r]);
 }
 /**
  * Return all hooks registered for the given file (or every file when no
@@ -135,20 +179,22 @@ export function getRegisteredTests(filePath) {
  * effect on the registry.
  */
 export function getRegisteredHooks(filePath) {
+    const state = getSharedState();
     if (filePath)
-        return [...(hookRegistries.get(filePath) ?? [])];
-    return [...hookRegistries.values()].flatMap(r => [...r]);
+        return [...(state.hookRegistries.get(filePath) ?? [])];
+    return [...state.hookRegistries.values()].flatMap(r => [...r]);
 }
 export function clearRegistry(filePath) {
+    const state = getSharedState();
     if (filePath) {
-        registries.delete(filePath);
-        hookRegistries.delete(filePath);
+        state.registries.delete(filePath);
+        state.hookRegistries.delete(filePath);
     }
     else {
-        registries.clear();
-        hookRegistries.clear();
+        state.registries.clear();
+        state.hookRegistries.clear();
     }
-    activeFile = null;
+    state.activeFile = null;
 }
 /**
  * Internal helper used by the four public hook registrars below.
@@ -163,15 +209,16 @@ function registerHook(kind, fn) {
     if (typeof fn !== 'function') {
         throw new Error(`${kind}() requires a function as its first argument`);
     }
-    const key = activeFile ?? DEFAULT_KEY;
-    if (!hookRegistries.has(key)) {
-        hookRegistries.set(key, []);
+    const state = getSharedState();
+    const key = state.activeFile ?? DEFAULT_KEY;
+    if (!state.hookRegistries.has(key)) {
+        state.hookRegistries.set(key, []);
     }
-    const describePath = describeStack.length > 0 ? describeStack.join(' > ') : undefined;
-    hookRegistries.get(key).push({
+    const describePath = state.describeStack.length > 0 ? state.describeStack.join(' > ') : undefined;
+    state.hookRegistries.get(key).push({
         kind,
         fn,
-        sourceFile: activeFile ?? undefined,
+        sourceFile: state.activeFile ?? undefined,
         describePath,
     });
 }

package/lib/dist/lib/testCases/define.js.map

@@ -1 +1 @@
-{"version":3,"file":"define.js","sourceRoot":"","sources":["../../../testCases/define.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAUH,OAAO,EAAE,OAAO,EAAE,MAAM,iBAAiB,CAAC;AAE1C,MAAM,UAAU,GAAG,IAAI,GAAG,EAA0B,CAAC;AACrD,yEAAyE;AACzE,uEAAuE;AACvE,sEAAsE;AACtE,MAAM,cAAc,GAAG,IAAI,GAAG,EAA4B,CAAC;AAC3D,IAAI,UAAU,GAAkB,IAAI,CAAC;AACrC,MAAM,WAAW,GAAG,aAAa,CAAC;AAElC,yEAAyE;AACzE,uEAAuE;AACvE,yEAAyE;AACzE,MAAM,aAAa,GAAa,EAAE,CAAC;AAEnC,wEAAwE;AACxE,8EAA8E;AAC9E,6EAA6E;AAC7E,IAAI,0BAA0B,GAAG,KAAK,CAAC;AACvC,SAAS,2BAA2B;IAClC,IAAI,0BAA0B;QAAE,OAAO;IACvC,0BAA0B,GAAG,IAAI,CAAC;IAClC,IAAI,OAAO,CAAC,0BAA0B,EAAE,oCAAoC,CAAC,KAAK,GAAG;QAAE,OAAO;IAC9F,sCAAsC;IACtC,OAAO,CAAC,IAAI,CACV,sEAAsE;QACtE,gEAAgE;QAChE,wEAAwE;QACxE,2DAA2D,CAC5D,CAAC;AACJ,CAAC;AAED,gBAAgB;AAChB,MAAM,UAAU,yBAAyB;IACvC,0BAA0B,GAAG,KAAK,CAAC;AACrC,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,QAAgB;IAC5C,UAAU,GAAG,QAAQ,CAAC;IACtB,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC9B,UAAU,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;IAC/B,CAAC;IACD,IAAI,CAAC,cAAc,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QAClC,cAAc,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;IACnC,CAAC;AACH,CAAC;AA6BD,MAAM,UAAU,IAAI,CAClB,IAAY,EACZ,aAA2E,EAC3E,SAAwD;IAExD,2BAA2B,EAAE,CAAC;IAE9B,2CAA2C;IAC3C,IAAI,OAAoB,CAAC;IACzB,IAAI,QAAsD,CAAC;IAC3D,IAAI,OAAO,aAAa,KAAK,UAAU,EAAE,CAAC;QACxC,OAAO,GAAG,EAAE,CAAC;QACb,QAAQ,GAAG,aAAa,CAAC;IAC3B,CAAC;SAAM,CAAC;QACN,OAAO,GAAG,aAAa,IAAI,EAAE,CAAC;QAC9B,QAAQ,GAAG,SAAyD,CAAC;IACvE,CAAC;IAED,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;IACjE,CAAC;IACD,IAAI,OAAO,QAAQ,KAAK,UAAU,EAAE,CAAC;QACnC,MAAM,IAAI,KAAK,CAAC,SAAS,IAAI,6BAA6B,CAAC,CAAC;IAC9D,CAAC;IAED,MAAM,GAAG,GAAG,UAAU,IAAI,WAAW,CAAC;IACtC,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;QACzB,UAAU,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;IAC1B,CAAC;IACD,MAAM,QAAQ,GAAG,UAAU,CAAC,GAAG,CAAC,GAAG,CAAE,CAAC;IAEtC,0EAA0E;IAC1E,yEAAyE;IACzE,+BAA+B;IAC/B,MAAM,aAAa,GAAG,aAAa,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvF,IAAI,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,IAAI,CAAC,CAAC,aAAa,KAAK,aAAa,CAAC,EAAE,CAAC;QAC7E,MAAM,SAAS,GAAG,UAAU,CAAC,CAAC,CAAC,OAAO,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACxD,MAAM,UAAU,GAAG,aAAa,CAAC,CAAC,CAAC,kBAAkB,aAAa,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5E,MAAM,IAAI,KAAK,CACb,wBAAwB,IAAI,IAAI,UAAU,GAAG,SAAS,IAAI;YAC1D,yDAAyD;YACzD,+DAA+D,CAChE,CAAC;IACJ,CAAC;IAED,QAAQ,CAAC,IAAI,CAAC;QACZ,IAAI;QACJ,OAAO;QACP,QAAQ;QACR,UAAU,EAAE,UAAU,IAAI,SAAS;QACnC,aAAa;KACd,CAAC,CAAC;AACL,CAAC;AAED;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,QAAQ,CAAC,IAAY,EAAE,EAAc;IACnD,2BAA2B,EAAE,CAAC;IAC9B,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,iDAAiD,CAAC,CAAC;IACrE,CAAC;IACD,IAAI,OAAO,EAAE,KAAK,UAAU,EAAE,CAAC;QAC7B,MAAM,IAAI,KAAK,CAAC,aAAa,IAAI,6BAA6B,CAAC,CAAC;IAClE,CAAC;IACD,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACzB,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,EAAE,EAAa,CAAC;QAC/B,IAAI,MAAM,IAAI,OAAQ,MAAc,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;YACzD,kEAAkE;YAClE,+DAA+D;YAC/D,MAAM,IAAI,KAAK,CACb,aAAa,IAAI,8BAA8B;gBAC/C,qEAAqE,CACtE,CAAC;QACJ,CAAC;IACH,CAAC;YAAS,CAAC;QACT,aAAa,CAAC,GAAG,EAAE,CAAC;IACtB,CAAC;AACH,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,QAAiB;IAClD,IAAI,QAAQ;QAAE,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IAC3D,OAAO,CAAC,GAAG,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;AACvD,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,kBAAkB,CAAC,QAAiB;IAClD,IAAI,QAAQ;QAAE,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IAC/D,OAAO,CAAC,GAAG,cAAc,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;AAC3D,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,QAAiB;IAC7C,IAAI,QAAQ,EAAE,CAAC;QACb,UAAU,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QAC5B,cAAc,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IAClC,CAAC;SAAM,CAAC;QACN,UAAU,CAAC,KAAK,EAAE,CAAC;QACnB,cAAc,CAAC,KAAK,EAAE,CAAC;IACzB,CAAC;IACD,UAAU,GAAG,IAAI,CAAC;AACpB,CAAC;AAED;;;;;;;GAOG;AACH,SAAS,YAAY,CAAC,IAAc,EAAE,EAAU;IAC9C,2BAA2B,EAAE,CAAC;IAC9B,IAAI,OAAO,EAAE,KAAK,UAAU,EAAE,CAAC;QAC7B,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,8CAA8C,CAAC,CAAC;IACzE,CAAC;IACD,MAAM,GAAG,GAAG,UAAU,IAAI,WAAW,CAAC;IACtC,IAAI,CAAC,cAAc,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;QAC7B,cAAc,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;IAC9B,CAAC;IACD,MAAM,YAAY,GAAG,aAAa,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACtF,cAAc,CAAC,GAAG,CAAC,GAAG,CAAE,CAAC,IAAI,CAAC;QAC5B,IAAI;QACJ,EAAE;QACF,UAAU,EAAE,UAAU,IAAI,SAAS;QACnC,YAAY;KACb,CAAC,CAAC;AACL,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,SAAS,CAAC,EAAU,IAAU,YAAY,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;AAE9E;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,QAAQ,CAAC,EAAU,IAAU,YAAY,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;AAE5E;;;;;;;;;;;;;GAaG;AACH,MAAM,UAAU,UAAU,CAAC,EAAU,IAAU,YAAY,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;AAEhF;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,SAAS,CAAC,EAAU,IAAU,YAAY,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;AAE9E,qDAAqD;AACrD,EAAE;AACF,2EAA2E;AAC3E,0EAA0E;AAC1E,2EAA2E;AAC3E,4CAA4C;AAC5C,2DAA2D;AAC3D,WAAiB,IAAI;IACnB,wDAAwD;IAC3C,cAAS,GAAyB,CAAC,EAAE,EAAE,EAAE,CAAC,YAAY,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;IACrF,wDAAwD;IAC3C,aAAQ,GAAyB,CAAC,EAAE,EAAE,EAAE,CAAC,YAAY,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IACnF,wDAAwD;IAC3C,eAAU,GAAyB,CAAC,EAAE,EAAE,EAAE,CAAC,YAAY,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC;IACvF,wDAAwD;IAC3C,cAAS,GAAyB,CAAC,EAAE,EAAE,EAAE,CAAC,YAAY,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;AACvF,CAAC,EATgB,IAAI,KAAJ,IAAI,QASpB"}
+{"version":3,"file":"define.js","sourceRoot":"","sources":["../../../testCases/define.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAUH,OAAO,EAAE,OAAO,EAAE,MAAM,iBAAiB,CAAC;AAE1C,MAAM,WAAW,GAAG,aAAa,CAAC;AAElC;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,MAAM,YAAY,GAAG,MAAM,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;AAUjE,SAAS,cAAc;IACrB,sEAAsE;IACtE,wEAAwE;IACxE,sEAAsE;IACtE,sEAAsE;IACtE,wEAAwE;IACxE,yEAAyE;IACzE,uDAAuD;IACvD,0DAA0D;IAC1D,iEAAiE;IACjE,oEAAoE;IACpE,iEAAiE;IACjE,MAAM,CAAC,GAAG,UAA4D,CAAC;IACvE,IAAI,KAAK,GAAG,CAAC,CAAC,YAAY,CAAC,CAAC;IAC5B,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,KAAK,GAAG;YACN,UAAU,EAAE,IAAI,GAAG,EAA0B;YAC7C,cAAc,EAAE,IAAI,GAAG,EAA4B;YACnD,UAAU,EAAE,IAAI;YAChB,aAAa,EAAE,EAAE;YACjB,0BAA0B,EAAE,KAAK;SAClC,CAAC;QACF,CAAC,CAAC,YAAY,CAAC,GAAG,KAAK,CAAC;IAC1B,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,2BAA2B;IAClC,MAAM,KAAK,GAAG,cAAc,EAAE,CAAC;IAC/B,IAAI,KAAK,CAAC,0BAA0B;QAAE,OAAO;IAC7C,KAAK,CAAC,0BAA0B,GAAG,IAAI,CAAC;IACxC,IAAI,OAAO,CAAC,0BAA0B,EAAE,oCAAoC,CAAC,KAAK,GAAG;QAAE,OAAO;IAC9F,sCAAsC;IACtC,OAAO,CAAC,IAAI,CACV,sEAAsE;QACtE,gEAAgE;QAChE,wEAAwE;QACxE,2DAA2D,CAC5D,CAAC;AACJ,CAAC;AAED,gBAAgB;AAChB,MAAM,UAAU,yBAAyB;IACvC,cAAc,EAAE,CAAC,0BAA0B,GAAG,KAAK,CAAC;AACtD,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,QAAgB;IAC5C,MAAM,KAAK,GAAG,cAAc,EAAE,CAAC;IAC/B,KAAK,CAAC,UAAU,GAAG,QAAQ,CAAC;IAC5B,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;IACrC,CAAC;IACD,IAAI,CAAC,KAAK,CAAC,cAAc,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QACxC,KAAK,CAAC,cAAc,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;IACzC,CAAC;AACH,CAAC;AA6BD,MAAM,UAAU,IAAI,CAClB,IAAY,EACZ,aAA2E,EAC3E,SAAwD;IAExD,2BAA2B,EAAE,CAAC;IAE9B,2CAA2C;IAC3C,IAAI,OAAoB,CAAC;IACzB,IAAI,QAAsD,CAAC;IAC3D,IAAI,OAAO,aAAa,KAAK,UAAU,EAAE,CAAC;QACxC,OAAO,GAAG,EAAE,CAAC;QACb,QAAQ,GAAG,aAAa,CAAC;IAC3B,CAAC;SAAM,CAAC;QACN,OAAO,GAAG,aAAa,IAAI,EAAE,CAAC;QAC9B,QAAQ,GAAG,SAAyD,CAAC;IACvE,CAAC;IAED,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;IACjE,CAAC;IACD,IAAI,OAAO,QAAQ,KAAK,UAAU,EAAE,CAAC;QACnC,MAAM,IAAI,KAAK,CAAC,SAAS,IAAI,6BAA6B,CAAC,CAAC;IAC9D,CAAC;IAED,MAAM,KAAK,GAAG,cAAc,EAAE,CAAC;IAC/B,MAAM,GAAG,GAAG,KAAK,CAAC,UAAU,IAAI,WAAW,CAAC;IAC5C,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;QAC/B,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;IAChC,CAAC;IACD,MAAM,QAAQ,GAAG,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,GAAG,CAAE,CAAC;IAE5C,0EAA0E;IAC1E,yEAAyE;IACzE,+BAA+B;IAC/B,MAAM,aAAa,GAAG,KAAK,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACnG,IAAI,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,IAAI,CAAC,CAAC,aAAa,KAAK,aAAa,CAAC,EAAE,CAAC;QAC7E,MAAM,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,OAAO,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpE,MAAM,UAAU,GAAG,aAAa,CAAC,CAAC,CAAC,kBAAkB,aAAa,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5E,MAAM,IAAI,KAAK,CACb,wBAAwB,IAAI,IAAI,UAAU,GAAG,SAAS,IAAI;YAC1D,yDAAyD;YACzD,+DAA+D,CAChE,CAAC;IACJ,CAAC;IAED,QAAQ,CAAC,IAAI,CAAC;QACZ,IAAI;QACJ,OAAO;QACP,QAAQ;QACR,UAAU,EAAE,KAAK,CAAC,UAAU,IAAI,SAAS;QACzC,aAAa;KACd,CAAC,CAAC;AACL,CAAC;AAED;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,QAAQ,CAAC,IAAY,EAAE,EAAc;IACnD,2BAA2B,EAAE,CAAC;IAC9B,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,iDAAiD,CAAC,CAAC;IACrE,CAAC;IACD,IAAI,OAAO,EAAE,KAAK,UAAU,EAAE,CAAC;QAC7B,MAAM,IAAI,KAAK,CAAC,aAAa,IAAI,6BAA6B,CAAC,CAAC;IAClE,CAAC;IACD,MAAM,KAAK,GAAG,cAAc,EAAE,CAAC;IAC/B,KAAK,CAAC,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC/B,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,EAAE,EAAa,CAAC;QAC/B,IAAI,MAAM,IAAI,OAAQ,MAAc,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;YACzD,kEAAkE;YAClE,+DAA+D;YAC/D,MAAM,IAAI,KAAK,CACb,aAAa,IAAI,8BAA8B;gBAC/C,qEAAqE,CACtE,CAAC;QACJ,CAAC;IACH,CAAC;YAAS,CAAC;QACT,KAAK,CAAC,aAAa,CAAC,GAAG,EAAE,CAAC;IAC5B,CAAC;AACH,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,QAAiB;IAClD,MAAM,KAAK,GAAG,cAAc,EAAE,CAAC;IAC/B,IAAI,QAAQ;QAAE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IACjE,OAAO,CAAC,GAAG,KAAK,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;AAC7D,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,kBAAkB,CAAC,QAAiB;IAClD,MAAM,KAAK,GAAG,cAAc,EAAE,CAAC;IAC/B,IAAI,QAAQ;QAAE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,cAAc,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IACrE,OAAO,CAAC,GAAG,KAAK,CAAC,cAAc,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;AACjE,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,QAAiB;IAC7C,MAAM,KAAK,GAAG,cAAc,EAAE,CAAC;IAC/B,IAAI,QAAQ,EAAE,CAAC;QACb,KAAK,CAAC,UAAU,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QAClC,KAAK,CAAC,cAAc,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IACxC,CAAC;SAAM,CAAC;QACN,KAAK,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC;QACzB,KAAK,CAAC,cAAc,CAAC,KAAK,EAAE,CAAC;IAC/B,CAAC;IACD,KAAK,CAAC,UAAU,GAAG,IAAI,CAAC;AAC1B,CAAC;AAED;;;;;;;GAOG;AACH,SAAS,YAAY,CAAC,IAAc,EAAE,EAAU;IAC9C,2BAA2B,EAAE,CAAC;IAC9B,IAAI,OAAO,EAAE,KAAK,UAAU,EAAE,CAAC;QAC7B,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,8CAA8C,CAAC,CAAC;IACzE,CAAC;IACD,MAAM,KAAK,GAAG,cAAc,EAAE,CAAC;IAC/B,MAAM,GAAG,GAAG,KAAK,CAAC,UAAU,IAAI,WAAW,CAAC;IAC5C,IAAI,CAAC,KAAK,CAAC,cAAc,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;QACnC,KAAK,CAAC,cAAc,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;IACpC,CAAC;IACD,MAAM,YAAY,GAAG,KAAK,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAClG,KAAK,CAAC,cAAc,CAAC,GAAG,CAAC,GAAG,CAAE,CAAC,IAAI,CAAC;QAClC,IAAI;QACJ,EAAE;QACF,UAAU,EAAE,KAAK,CAAC,UAAU,IAAI,SAAS;QACzC,YAAY;KACb,CAAC,CAAC;AACL,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,SAAS,CAAC,EAAU,IAAU,YAAY,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;AAE9E;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,QAAQ,CAAC,EAAU,IAAU,YAAY,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;AAE5E;;;;;;;;;;;;;GAaG;AACH,MAAM,UAAU,UAAU,CAAC,EAAU,IAAU,YAAY,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;AAEhF;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,SAAS,CAAC,EAAU,IAAU,YAAY,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;AAE9E,qDAAqD;AACrD,EAAE;AACF,2EAA2E;AAC3E,0EAA0E;AAC1E,2EAA2E;AAC3E,4CAA4C;AAC5C,2DAA2D;AAC3D,WAAiB,IAAI;IACnB,wDAAwD;IAC3C,cAAS,GAAyB,CAAC,EAAE,EAAE,EAAE,CAAC,YAAY,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;IACrF,wDAAwD;IAC3C,aAAQ,GAAyB,CAAC,EAAE,EAAE,EAAE,CAAC,YAAY,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IACnF,wDAAwD;IAC3C,eAAU,GAAyB,CAAC,EAAE,EAAE,EAAE,CAAC,YAAY,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC;IACvF,wDAAwD;IAC3C,cAAS,GAAyB,CAAC,EAAE,EAAE,EAAE,CAAC,YAAY,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;AACvF,CAAC,EATgB,IAAI,KAAJ,IAAI,QASpB"}

package/lib/dist/lib/testCases/judge.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"judge.d.ts","sourceRoot":"","sources":["../../../testCases/judge.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoCG;AAEH,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,SAAS,CAAC;AAK9C,8EAA8E;AAC9E,MAAM,MAAM,SAAS,GAAG,MAAM,GAAG,SAAS,CAAC;AAuB3C,8EAA8E;AAC9E,wBAAgB,uBAAuB,IAAI,IAAI,CAE9C;AAED;;;;;;;;GAQG;AACH,MAAM,WAAW,OAAO;IACtB,cAAc,EAAE,QAAQ,GAAG,QAAQ,CAAC;IACpC,kDAAkD;IAClD,QAAQ,EAAE,MAAM,CAAC;IACjB,kEAAkE;IAClE,KAAK,EAAE,MAAM,CAAC;IACd,iCAAiC;IACjC,SAAS,EAAE,MAAM,CAAC;IAClB,kDAAkD;IAClD,IAAI,EAAE,OAAO,CAAC;IACd,iEAAiE;IACjE,IAAI,EAAE,SAAS,CAAC;IAChB,8DAA8D;IAC9D,OAAO,EAAE,OAAO,CAAC;IACjB,iEAAiE;IACjE,OAAO,EAAE,OAAO,CAAC;IACjB,uDAAuD;IACvD,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;OAIG;IACH,OAAO,IAAI,OAAO,CAAC;CACpB;AAED,uEAAuE;AACvE,MAAM,MAAM,YAAY,GAAG,OAAO,CAAC;AAEnC;;;;GAIG;AACH,MAAM,WAAW,YAAY;IAC3B;;;OAGG;IACH,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;;;;OAMG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;;OAIG;IACH,IAAI,CAAC,EAAE,OAAO,CAAC;CAChB;AAeD,2EAA2E;AAC3E,wBAAgB,eAAe,IAAI,IAAI,CAEtC;AA0CD,wBAAgB,cAAc,IAAI,OAAO,CAExC;AAED,wBAAgB,cAAc,IAAI,IAAI,CAErC;AAED,UAAU,UAAU;IAClB,UAAU,CAAC,EAAE,cAAc,EAAE,CAAC;IAC9B,aAAa,CAAC,EAAE,MAAM,MAAM,CAAC;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AA8BD;;;GAGG;AACH,MAAM,WAAW,OAAO;IACtB,CACE,kBAAkB,EAAE,UAAU,GAAG,cAAc,EAAE,EACjD,aAAa,EAAE,MAAM,GAAG,MAAM,EAAE,EAChC,OAAO,CAAC,EAAE,YAAY,GACrB,OAAO,CAAC,OAAO,CAAC,CAAC;IACpB;;;OAGG;IACH,OAAO,CACL,kBAAkB,EAAE,UAAU,GAAG,cAAc,EAAE,EACjD,aAAa,EAAE,MAAM,GAAG,MAAM,EAAE,EAChC,OAAO,CAAC,EAAE,YAAY,GACrB,OAAO,CAAC,OAAO,CAAC,CAAC;CACrB;AAyND;;;;;;;GAOG;AACH,eAAO,MAAM,KAAK,EAAE,OAanB,CAAC;AAEF;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,SAAS,CAAC,QAAQ,CAAC,EAAE;IACnC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,IAAI,CAAC,EAAE,OAAO,CAAC;CAChB,GAAG,OAAO,CA+BV"}
+{"version":3,"file":"judge.d.ts","sourceRoot":"","sources":["../../../testCases/judge.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoCG;AAEH,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,SAAS,CAAC;AAK9C,8EAA8E;AAC9E,MAAM,MAAM,SAAS,GAAG,MAAM,GAAG,SAAS,CAAC;AAuB3C,8EAA8E;AAC9E,wBAAgB,uBAAuB,IAAI,IAAI,CAE9C;AAED;;;;;;;;GAQG;AACH,MAAM,WAAW,OAAO;IACtB,cAAc,EAAE,QAAQ,GAAG,QAAQ,CAAC;IACpC,kDAAkD;IAClD,QAAQ,EAAE,MAAM,CAAC;IACjB,kEAAkE;IAClE,KAAK,EAAE,MAAM,CAAC;IACd,iCAAiC;IACjC,SAAS,EAAE,MAAM,CAAC;IAClB,kDAAkD;IAClD,IAAI,EAAE,OAAO,CAAC;IACd,iEAAiE;IACjE,IAAI,EAAE,SAAS,CAAC;IAChB,8DAA8D;IAC9D,OAAO,EAAE,OAAO,CAAC;IACjB,iEAAiE;IACjE,OAAO,EAAE,OAAO,CAAC;IACjB,uDAAuD;IACvD,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;OAIG;IACH,OAAO,IAAI,OAAO,CAAC;CACpB;AAED,uEAAuE;AACvE,MAAM,MAAM,YAAY,GAAG,OAAO,CAAC;AAEnC;;;;GAIG;AACH,MAAM,WAAW,YAAY;IAC3B;;;OAGG;IACH,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;;;;OAMG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;;OAIG;IACH,IAAI,CAAC,EAAE,OAAO,CAAC;CAChB;AAmBD,2EAA2E;AAC3E,wBAAgB,eAAe,IAAI,IAAI,CAEtC;AA0CD,wBAAgB,cAAc,IAAI,OAAO,CAExC;AAED,wBAAgB,cAAc,IAAI,IAAI,CAErC;AAED,UAAU,UAAU;IAClB,UAAU,CAAC,EAAE,cAAc,EAAE,CAAC;IAC9B,aAAa,CAAC,EAAE,MAAM,MAAM,CAAC;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AA8BD;;;GAGG;AACH,MAAM,WAAW,OAAO;IACtB,CACE,kBAAkB,EAAE,UAAU,GAAG,cAAc,EAAE,EACjD,aAAa,EAAE,MAAM,GAAG,MAAM,EAAE,EAChC,OAAO,CAAC,EAAE,YAAY,GACrB,OAAO,CAAC,OAAO,CAAC,CAAC;IACpB;;;OAGG;IACH,OAAO,CACL,kBAAkB,EAAE,UAAU,GAAG,cAAc,EAAE,EACjD,aAAa,EAAE,MAAM,GAAG,MAAM,EAAE,EAChC,OAAO,CAAC,EAAE,YAAY,GACrB,OAAO,CAAC,OAAO,CAAC,CAAC;CACrB;AA4OD;;;;;;;GAOG;AACH,eAAO,MAAM,KAAK,EAAE,OAanB,CAAC;AAEF;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,SAAS,CAAC,QAAQ,CAAC,EAAE;IACnC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,IAAI,CAAC,EAAE,OAAO,CAAC;CAChB,GAAG,OAAO,CA+BV"}

package/lib/dist/lib/testCases/judge.js

@@ -144,10 +144,19 @@ async function runJudge(resultOrTrajectory, claimOrClaims, options, role) {
     // Record the success path identically whether the verdict came fresh from
     // the endpoint or from the in-process cache.
     const finalize = (raw, durationMs) => {
-        const accuracy = raw.metrics?.accuracy ?? 0;
+        // Headline score precedence: legacy `accuracy` metric → server-computed
+        // weighted overall (custom evaluators declare their own dimensions and
+        // rarely emit `accuracy`) → none. Never fabricate a 0: pre-fix this
+        // defaulted to 0, so every custom-evaluator verdict rendered as
+        // "score 0%" — even passing ones.
+        const headline = typeof raw.metrics?.accuracy === 'number'
+            ? raw.metrics.accuracy
+            : typeof raw.overallScore === 'number'
+                ? raw.overallScore
+                : undefined;
         const verdict = makeVerdict({
             passFailStatus: raw.passFailStatus ?? 'failed',
-            accuracy,
+            accuracy: headline ?? 0,
             reasoning: raw.llmJudgeReasoning ?? '',
             role,
             skipped: false,
@@ -159,10 +168,16 @@ async function runJudge(resultOrTrajectory, claimOrClaims, options, role) {
             method: 'llm-judge',
             role,
             durationMs,
-            score: verdict.score,
+            // Only record a score when the judge actually produced a headline
+            // number — a fabricated 0 renders as a misleading "score 0%".
+            ...(headline !== undefined ? { score: headline / 100 } : {}),
             reasoning: verdict.reasoning,
             model: options?.model,
-            errorMessage: verdict.passFailStatus === 'failed' ? verdict.reasoning : undefined,
+            // NOTE: no `errorMessage` mirror. It used to copy `reasoning` verbatim
+            // on failure, which persisted the same multi-KB string twice and made
+            // the UI render it twice (once red as "error", once as "reasoning").
+            // `reasoning` is the single source of truth; consumers that need a
+            // failure detail fall back to it (see expect.toPass()).
             // Preserve the rest of the judge payload — these were silently
             // dropped before, which made SDK `judge()` calls strictly less
             // informative than the legacy auto-judge path. See MatcherResult.
@@ -172,6 +187,9 @@ async function runJudge(resultOrTrajectory, claimOrClaims, options, role) {
             ...(raw.metrics && typeof raw.metrics === 'object'
                 ? { judgeMetrics: { ...raw.metrics } }
                 : {}),
+            ...(raw.extraFields && typeof raw.extraFields === 'object' && Object.keys(raw.extraFields).length > 0
+                ? { judgeExtraFields: raw.extraFields }
+                : {}),
         });
         return verdict;
     };