@opensearch-project/agent-health 0.5.2 → 0.7.0

package/lib/dist/lib/runActions.d.ts

@@ -0,0 +1,120 @@
+/**
+ * Pure, isomorphic predicates for the run-lifecycle action matrix (delete /
+ * cancel / re-run / retry-judgement) shared by every run surface (runs list,
+ * benchmark runs list, run detail/report page, inspector header) AND by the
+ * server routes that enforce the same rules server-side. No storage/IO here
+ * — callers pass in the run document they already have.
+ *
+ * Action matrix (see AGENTS.md / PR description for the full writeup):
+ *   - Delete: any run, any status. Always available (existing endpoint).
+ *   - Cancel: only while `status === 'running'`.
+ *   - Re-run: only top-level EvaluationRun docs (docType === 'evaluation-run').
+ *     Legacy benchmark-embedded BenchmarkRun rows don't support the
+ *     provenance-tracked rerun endpoint (pre-existing constraint — see
+ *     RunConfigDialog / EvalRunsPage).
+ *   - Retry judgement: only EvaluationRun docs, only when the run is
+ *     terminal (not running) AND it has at least one test case whose agent
+ *     execution completed but the judge produced NO verdict (a judge-failed
+ *     / "errored" case — trace timeout, judge 400, "evaluator could not
+ *     run" — as opposed to an agent-failed one — retrying the judge on a
+ *     case the agent itself never finished has nothing to re-grade). Same
+ *     predicate the retry-judgement pipeline itself selects on
+ *     (services/evaluation/retryJudgement.ts `isJudgeFailedCase`, keyed on
+ *     the report's `metricsStatus: 'error'`, which the runner mirrors onto
+ *     the run's results map as a `completed` result with no
+ *     `passFailStatus`) and that `lib/runStats` buckets as `errored`.
+ */
+import type { BenchmarkRun, EvaluationRun } from '../types/index.js';
+/** Minimal shape both BenchmarkRun and EvaluationRun satisfy for these checks. */
+export type RunLike = Pick<BenchmarkRun | EvaluationRun, 'status' | 'results'> & {
+    docType?: string;
+};
+/**
+ * True when `run` is a top-level EvaluationRun document (created via
+ * `POST /api/storage/evaluation-runs`), as opposed to a legacy
+ * benchmark-embedded BenchmarkRun (`benchmark.runs[]`). The two share a lot
+ * of shape but only EvaluationRun docs carry `docType: 'evaluation-run'` and
+ * support the rerun/retry-judgement endpoints.
+ *
+ * Null-tolerant wrapper over the typed predicate in `types/index.ts` (the
+ * single source of truth for the docType discriminator) — kept so callers
+ * holding a possibly-null run don't need their own guard.
+ */
+export declare function isEvaluationRun(run: RunLike | null | undefined): run is EvaluationRun;
+/** True while the run has an in-progress executor that a Cancel action could stop. */
+export declare function isRunRunning(run: RunLike | null | undefined): boolean;
+/** True once a run has reached any terminal state (not running/pending). */
+export declare function isRunTerminal(run: RunLike | null | undefined): boolean;
+/**
+ * Count test cases where the AGENT finished (`status === 'completed'`) but
+ * the JUDGE produced no verdict (`passFailStatus` neither 'passed' nor
+ * 'failed') — the "errored" bucket of `lib/runStats` `bucketRunResults`
+ * (issue #242) and exactly the set `POST .../retry-judgement` (default
+ * `scope=errored`) will re-judge. Deliberately excludes:
+ *   - `status !== 'completed'` (agent-failed/cancelled/pending cases — no
+ *     trajectory to re-judge, or nothing ran).
+ *   - a real 'failed' verdict — the judge DID run and graded the case; that
+ *     is a legitimate result, not a judge failure (re-grading it is
+ *     `scope=all`, opt-in from the inspector's dedicated button).
+ *
+ * `passFailStatus` isn't declared on `EvaluationRun['results']`'s static
+ * type (a pre-existing gap — evaluationRunner.ts writes it via an `as any`
+ * spread) so this reads it defensively.
+ */
+export declare function countJudgeFailed(run: RunLike | null | undefined): number;
+export interface RunActionVisibility {
+    /** Delete is always available for any run in any status. */
+    canDelete: boolean;
+    /** Cancel is available only while the run is actively running. */
+    canCancel: boolean;
+    /** Re-run is available only for top-level EvaluationRun docs. */
+    canRerun: boolean;
+    /** Reason to show (e.g. as a disabled-item tooltip) when canRerun is false. */
+    rerunDisabledReason?: string;
+    /** Retry judgement: EvaluationRun, terminal, with >0 judge-failed cases. */
+    canRetryJudgement: boolean;
+    /** Reason to show when canRetryJudgement is false. */
+    retryJudgementDisabledReason?: string;
+    /** Number of judge-failed test cases (0 when not applicable/unknown). */
+    judgeFailedCount: number;
+}
+/**
+ * Minimum time a run must have been persisted before a Cancel request with
+ * no in-memory executor token is allowed to take the "zombie" fallback path
+ * (mark cancelled directly — see getRunActionVisibility callers in the
+ * cancel routes). Guards the narrow window right after a run is created:
+ * the doc is persisted (and therefore visible to a concurrent Cancel
+ * request) strictly before its executor registers its cancellation token,
+ * so a Cancel that lands in that gap would otherwise mark a run "cancelled"
+ * moments before its own executor starts making progress on it. A brand-new
+ * run is also the case the fallback is LEAST useful for — "zombie" (no
+ * executor anywhere) is far more plausible once a run has been running for
+ * a while than in its first couple of seconds.
+ */
+export declare const ZOMBIE_CANCEL_MIN_AGE_MS = 5000;
+/**
+ * True once a run is old enough that a tokenless Cancel request can safely
+ * assume its executor (if any) would already have registered a
+ * cancellation token — i.e. it's safe to treat "no token" as "no executor"
+ * rather than "executor hasn't started yet".
+ *
+ * NOTE — known limitation, not fixed by this check: cancellation tokens are
+ * tracked in an in-memory `Map` scoped to ONE server process. In a
+ * multi-process/clustered deployment, a Cancel request routed to a
+ * DIFFERENT process than the one executing the run will always find no
+ * token there, regardless of run age, and this zombie fallback will mark
+ * the run cancelled in storage even though it's alive and progressing on
+ * another process. This mirrors a pre-existing, documented constraint of
+ * this codebase's run-execution model (see AGENTS.md's "orphan-run
+ * recovery" notes: "active is tracked per-process"); fixing it for real
+ * needs the same heartbeat-based ownership (`run.heartbeatAt`) that doc
+ * already calls out as the eventual replacement. Out of scope here.
+ */
+export declare function isOldEnoughForZombieCancel(createdAt: string | undefined, now?: number): boolean;
+/**
+ * Compute the full action-visibility matrix for one run. Pure function —
+ * safe to call from both React components and server-side route validation
+ * so the two never drift.
+ */
+export declare function getRunActionVisibility(run: RunLike | null | undefined): RunActionVisibility;
+//# sourceMappingURL=runActions.d.ts.map

package/lib/dist/lib/runActions.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runActions.d.ts","sourceRoot":"","sources":["../../runActions.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAG3D,kFAAkF;AAClF,MAAM,MAAM,OAAO,GAAG,IAAI,CAAC,YAAY,GAAG,aAAa,EAAE,QAAQ,GAAG,SAAS,CAAC,GAAG;IAC/E,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF;;;;;;;;;;GAUG;AACH,wBAAgB,eAAe,CAAC,GAAG,EAAE,OAAO,GAAG,IAAI,GAAG,SAAS,GAAG,GAAG,IAAI,aAAa,CAErF;AAED,sFAAsF;AACtF,wBAAgB,YAAY,CAAC,GAAG,EAAE,OAAO,GAAG,IAAI,GAAG,SAAS,GAAG,OAAO,CAErE;AAED,4EAA4E;AAC5E,wBAAgB,aAAa,CAAC,GAAG,EAAE,OAAO,GAAG,IAAI,GAAG,SAAS,GAAG,OAAO,CAEtE;AAED;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,gBAAgB,CAAC,GAAG,EAAE,OAAO,GAAG,IAAI,GAAG,SAAS,GAAG,MAAM,CAUxE;AAED,MAAM,WAAW,mBAAmB;IAClC,4DAA4D;IAC5D,SAAS,EAAE,OAAO,CAAC;IACnB,kEAAkE;IAClE,SAAS,EAAE,OAAO,CAAC;IACnB,iEAAiE;IACjE,QAAQ,EAAE,OAAO,CAAC;IAClB,+EAA+E;IAC/E,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,4EAA4E;IAC5E,iBAAiB,EAAE,OAAO,CAAC;IAC3B,sDAAsD;IACtD,4BAA4B,CAAC,EAAE,MAAM,CAAC;IACtC,yEAAyE;IACzE,gBAAgB,EAAE,MAAM,CAAC;CAC1B;AAOD;;;;;;;;;;;;GAYG;AACH,eAAO,MAAM,wBAAwB,OAAO,CAAC;AAE7C;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAgB,0BAA0B,CAAC,SAAS,EAAE,MAAM,GAAG,SAAS,EAAE,GAAG,GAAE,MAAmB,GAAG,OAAO,CAI3G;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,GAAG,EAAE,OAAO,GAAG,IAAI,GAAG,SAAS,GAAG,mBAAmB,CAuB3F"}

package/lib/dist/lib/runActions.js

@@ -0,0 +1,130 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+import { isEvaluationRun as isEvaluationRunDoc } from '../types/index.js';
+/**
+ * True when `run` is a top-level EvaluationRun document (created via
+ * `POST /api/storage/evaluation-runs`), as opposed to a legacy
+ * benchmark-embedded BenchmarkRun (`benchmark.runs[]`). The two share a lot
+ * of shape but only EvaluationRun docs carry `docType: 'evaluation-run'` and
+ * support the rerun/retry-judgement endpoints.
+ *
+ * Null-tolerant wrapper over the typed predicate in `types/index.ts` (the
+ * single source of truth for the docType discriminator) — kept so callers
+ * holding a possibly-null run don't need their own guard.
+ */
+export function isEvaluationRun(run) {
+    return !!run && isEvaluationRunDoc(run);
+}
+/** True while the run has an in-progress executor that a Cancel action could stop. */
+export function isRunRunning(run) {
+    return run?.status === 'running';
+}
+/** True once a run has reached any terminal state (not running/pending). */
+export function isRunTerminal(run) {
+    return !!run && (run.status === 'completed' || run.status === 'failed' || run.status === 'cancelled');
+}
+/**
+ * Count test cases where the AGENT finished (`status === 'completed'`) but
+ * the JUDGE produced no verdict (`passFailStatus` neither 'passed' nor
+ * 'failed') — the "errored" bucket of `lib/runStats` `bucketRunResults`
+ * (issue #242) and exactly the set `POST .../retry-judgement` (default
+ * `scope=errored`) will re-judge. Deliberately excludes:
+ *   - `status !== 'completed'` (agent-failed/cancelled/pending cases — no
+ *     trajectory to re-judge, or nothing ran).
+ *   - a real 'failed' verdict — the judge DID run and graded the case; that
+ *     is a legitimate result, not a judge failure (re-grading it is
+ *     `scope=all`, opt-in from the inspector's dedicated button).
+ *
+ * `passFailStatus` isn't declared on `EvaluationRun['results']`'s static
+ * type (a pre-existing gap — evaluationRunner.ts writes it via an `as any`
+ * spread) so this reads it defensively.
+ */
+export function countJudgeFailed(run) {
+    if (!run?.results)
+        return 0;
+    let count = 0;
+    for (const r of Object.values(run.results)) {
+        const result = r;
+        if (result.status !== 'completed')
+            continue;
+        if (result.passFailStatus === 'passed' || result.passFailStatus === 'failed')
+            continue;
+        count++;
+    }
+    return count;
+}
+const RERUN_NOT_SUPPORTED_REASON = "Re-run isn't available for legacy benchmark-embedded runs";
+const RETRY_JUDGEMENT_NOT_SUPPORTED_REASON = "Retry judgement isn't available for legacy benchmark-embedded runs";
+const RETRY_JUDGEMENT_STILL_RUNNING_REASON = 'Retry judgement is only available once the run finishes';
+const RETRY_JUDGEMENT_NONE_FAILED_REASON = 'No judge-failed test cases to retry';
+/**
+ * Minimum time a run must have been persisted before a Cancel request with
+ * no in-memory executor token is allowed to take the "zombie" fallback path
+ * (mark cancelled directly — see getRunActionVisibility callers in the
+ * cancel routes). Guards the narrow window right after a run is created:
+ * the doc is persisted (and therefore visible to a concurrent Cancel
+ * request) strictly before its executor registers its cancellation token,
+ * so a Cancel that lands in that gap would otherwise mark a run "cancelled"
+ * moments before its own executor starts making progress on it. A brand-new
+ * run is also the case the fallback is LEAST useful for — "zombie" (no
+ * executor anywhere) is far more plausible once a run has been running for
+ * a while than in its first couple of seconds.
+ */
+export const ZOMBIE_CANCEL_MIN_AGE_MS = 5000;
+/**
+ * True once a run is old enough that a tokenless Cancel request can safely
+ * assume its executor (if any) would already have registered a
+ * cancellation token — i.e. it's safe to treat "no token" as "no executor"
+ * rather than "executor hasn't started yet".
+ *
+ * NOTE — known limitation, not fixed by this check: cancellation tokens are
+ * tracked in an in-memory `Map` scoped to ONE server process. In a
+ * multi-process/clustered deployment, a Cancel request routed to a
+ * DIFFERENT process than the one executing the run will always find no
+ * token there, regardless of run age, and this zombie fallback will mark
+ * the run cancelled in storage even though it's alive and progressing on
+ * another process. This mirrors a pre-existing, documented constraint of
+ * this codebase's run-execution model (see AGENTS.md's "orphan-run
+ * recovery" notes: "active is tracked per-process"); fixing it for real
+ * needs the same heartbeat-based ownership (`run.heartbeatAt`) that doc
+ * already calls out as the eventual replacement. Out of scope here.
+ */
+export function isOldEnoughForZombieCancel(createdAt, now = Date.now()) {
+    const created = createdAt ? Date.parse(createdAt) : NaN;
+    if (Number.isNaN(created))
+        return true; // no timestamp to compare against — don't block on it
+    return now - created >= ZOMBIE_CANCEL_MIN_AGE_MS;
+}
+/**
+ * Compute the full action-visibility matrix for one run. Pure function —
+ * safe to call from both React components and server-side route validation
+ * so the two never drift.
+ */
+export function getRunActionVisibility(run) {
+    const evalRun = isEvaluationRun(run);
+    const running = isRunRunning(run);
+    const terminal = isRunTerminal(run);
+    const judgeFailedCount = evalRun ? countJudgeFailed(run) : 0;
+    const canRetryJudgement = evalRun && terminal && judgeFailedCount > 0;
+    let retryJudgementDisabledReason;
+    if (!canRetryJudgement) {
+        if (!evalRun)
+            retryJudgementDisabledReason = RETRY_JUDGEMENT_NOT_SUPPORTED_REASON;
+        else if (!terminal)
+            retryJudgementDisabledReason = RETRY_JUDGEMENT_STILL_RUNNING_REASON;
+        else
+            retryJudgementDisabledReason = RETRY_JUDGEMENT_NONE_FAILED_REASON;
+    }
+    return {
+        canDelete: true,
+        canCancel: running,
+        canRerun: evalRun,
+        rerunDisabledReason: evalRun ? undefined : RERUN_NOT_SUPPORTED_REASON,
+        canRetryJudgement,
+        retryJudgementDisabledReason,
+        judgeFailedCount,
+    };
+}
+//# sourceMappingURL=runActions.js.map

package/lib/dist/lib/runActions.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"runActions.js","sourceRoot":"","sources":["../../runActions.ts"],"names":[],"mappings":"AAAA;;;GAGG;AA8BH,OAAO,EAAE,eAAe,IAAI,kBAAkB,EAAE,MAAM,SAAS,CAAC;AAOhE;;;;;;;;;;GAUG;AACH,MAAM,UAAU,eAAe,CAAC,GAA+B;IAC7D,OAAO,CAAC,CAAC,GAAG,IAAI,kBAAkB,CAAC,GAAmC,CAAC,CAAC;AAC1E,CAAC;AAED,sFAAsF;AACtF,MAAM,UAAU,YAAY,CAAC,GAA+B;IAC1D,OAAO,GAAG,EAAE,MAAM,KAAK,SAAS,CAAC;AACnC,CAAC;AAED,4EAA4E;AAC5E,MAAM,UAAU,aAAa,CAAC,GAA+B;IAC3D,OAAO,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,KAAK,WAAW,IAAI,GAAG,CAAC,MAAM,KAAK,QAAQ,IAAI,GAAG,CAAC,MAAM,KAAK,WAAW,CAAC,CAAC;AACxG,CAAC;AAED;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,gBAAgB,CAAC,GAA+B;IAC9D,IAAI,CAAC,GAAG,EAAE,OAAO;QAAE,OAAO,CAAC,CAAC;IAC5B,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;QAC3C,MAAM,MAAM,GAAG,CAAwD,CAAC;QACxE,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW;YAAE,SAAS;QAC5C,IAAI,MAAM,CAAC,cAAc,KAAK,QAAQ,IAAI,MAAM,CAAC,cAAc,KAAK,QAAQ;YAAE,SAAS;QACvF,KAAK,EAAE,CAAC;IACV,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAmBD,MAAM,0BAA0B,GAAG,2DAA2D,CAAC;AAC/F,MAAM,oCAAoC,GAAG,oEAAoE,CAAC;AAClH,MAAM,oCAAoC,GAAG,yDAAyD,CAAC;AACvG,MAAM,kCAAkC,GAAG,qCAAqC,CAAC;AAEjF;;;;;;;;;;;;GAYG;AACH,MAAM,CAAC,MAAM,wBAAwB,GAAG,IAAI,CAAC;AAE7C;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,UAAU,0BAA0B,CAAC,SAA6B,EAAE,MAAc,IAAI,CAAC,GAAG,EAAE;IAChG,MAAM,OAAO,GAAG,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;IACxD,IAAI,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC;QAAE,OAAO,IAAI,CAAC,CAAC,sDAAsD;IAC9F,OAAO,GAAG,GAAG,OAAO,IAAI,wBAAwB,CAAC;AACnD,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,sBAAsB,CAAC,GAA+B;IACpE,MAAM,OAAO,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC;IACrC,MAAM,OAAO,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC;IAClC,MAAM,QAAQ,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC;IACpC,MAAM,gBAAgB,GAAG,OAAO,CAAC,CAAC,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAE7D,MAAM,iBAAiB,GAAG,OAAO,IAAI,QAAQ,IAAI,gBAAgB,GAAG,CAAC,CAAC;IACtE,IAAI,4BAAgD,CAAC;IACrD,IAAI,CAAC,iBAAiB,EAAE,CAAC;QACvB,IAAI,CAAC,OAAO;YAAE,4BAA4B,GAAG,oCAAoC,CAAC;aAC7E,IAAI,CAAC,QAAQ;YAAE,4BAA4B,GAAG,oCAAoC,CAAC;;YACnF,4BAA4B,GAAG,kCAAkC,CAAC;IACzE,CAAC;IAED,OAAO;QACL,SAAS,EAAE,IAAI;QACf,SAAS,EAAE,OAAO;QAClB,QAAQ,EAAE,OAAO;QACjB,mBAAmB,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,0BAA0B;QACrE,iBAAiB;QACjB,4BAA4B;QAC5B,gBAAgB;KACjB,CAAC;AACJ,CAAC"}

package/lib/dist/lib/runInsights.d.ts

@@ -0,0 +1,86 @@
+/**
+ * Deterministic (no-LLM) aggregation helpers for the run-report "insights"
+ * pane (RunInsightsPane.tsx), shown on the bare
+ * `/benchmarks/:benchmarkId/runs/:runId` route when no test case is
+ * selected — see owner feedback on goyamegh/run-report-redesign: "if no
+ * test case is selected, the right side can show an aggregated view ...
+ * why did the failing tests fail — something that is complete info."
+ *
+ * Everything here is a pure function over already-fetched data (report
+ * summaries + test-case categories). No network calls, no LLM calls — v1
+ * is explicitly deterministic-only per the product ask.
+ */
+export interface CategoryStatusRow {
+    category: string;
+    /** Any ResultStatus value; only 'passed' / 'failed' / 'errored' are counted distinctly, everything else falls into the bar's `total` only (pending/running cases). */
+    status: string;
+}
+export interface CategoryBar {
+    category: string;
+    passed: number;
+    failed: number;
+    errored: number;
+    total: number;
+}
+/**
+ * Group rows by test-case category and tally pass/fail/errored/total.
+ * Deterministic order: largest category first, ties broken alphabetically.
+ */
+export declare function computeCategoryBars(rows: CategoryStatusRow[]): CategoryBar[];
+/**
+ * Normalize a judge-reasoning string down to its first sentence,
+ * lowercased, punctuation-stripped, whitespace-collapsed. Used both as the
+ * clustering input and as the theme's stable `key`.
+ */
+export declare function normalizeReasoningKey(reasoning: string): string;
+export interface FailureThemeInput {
+    testCaseId: string;
+    reasoning: string;
+}
+export interface FailureTheme {
+    /** Stable cluster key — the normalized first sentence of the theme's representative case. */
+    key: string;
+    count: number;
+    /** Trimmed, human-readable first sentence sampled from the theme's most common exact phrasing. */
+    sampleSnippet: string;
+    testCaseIds: string[];
+}
+/**
+ * Cluster failing test cases into "why they failed" themes using a
+ * deterministic, LLM-free heuristic: normalize each case's judge-reasoning
+ * first sentence, then union-find cases that share at least one contiguous
+ * N-word shingle. This is robust to minor paraphrasing (a judge saying
+ * "unable to retrieve" vs "failed to retrieve" the same underlying tool
+ * connectivity failure) while still keeping genuinely distinct failure
+ * modes (e.g. "missing required facts" vs "MCP server unavailable")
+ * separate, because they share no contiguous phrase.
+ *
+ * Verified against a real production run (418-verify, 64 failing cases):
+ * 57 of 64 connectivity-flavored reasonings collapse into ONE dominant
+ * theme; the remaining 7 ("Required facts evaluation: ...", a genuinely
+ * different failure shape) form a second, correctly separate theme.
+ *
+ * Output order: largest theme first, ties broken by the theme's
+ * lowest-sorting testCaseId (deterministic, no dependency on input order).
+ */
+export declare function clusterFailureThemes(items: FailureThemeInput[], shingleSize?: number): FailureTheme[];
+/**
+ * "Based on N of M failing cases" note shown under the theme list when the
+ * reasoning fetch was capped (RunInsightsPane caps at the first 100 failing
+ * cases). Returns null when nothing was capped (fetchedCount >= totalCount).
+ */
+export declare function formatCappedNote(fetchedCount: number, totalFailingCount: number): string | null;
+export interface RankedCase {
+    testCaseId: string;
+    value: number;
+}
+/**
+ * Top-N cases by a numeric value (duration, cost, ...), descending.
+ * Cases with a null/undefined/NaN value are excluded. Deterministic tie
+ * break: lower testCaseId first.
+ */
+export declare function pickTopN(cases: {
+    testCaseId: string;
+    value: number | null | undefined;
+}[], n: number): RankedCase[];
+//# sourceMappingURL=runInsights.d.ts.map

package/lib/dist/lib/runInsights.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runInsights.d.ts","sourceRoot":"","sources":["../../runInsights.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;GAWG;AAIH,MAAM,WAAW,iBAAiB;IAChC,QAAQ,EAAE,MAAM,CAAC;IACjB,sKAAsK;IACtK,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,WAAW;IAC1B,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;CACf;AAID;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,iBAAiB,EAAE,GAAG,WAAW,EAAE,CAiB5E;AAID;;;;GAIG;AACH,wBAAgB,qBAAqB,CAAC,SAAS,EAAE,MAAM,GAAG,MAAM,CAM/D;AAqBD,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,YAAY;IAC3B,6FAA6F;IAC7F,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,kGAAkG;IAClG,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,MAAM,EAAE,CAAC;CACvB;AAOD;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,iBAAiB,EAAE,EAC1B,WAAW,GAAE,MAA6B,GACzC,YAAY,EAAE,CAwEhB;AAED;;;;GAIG;AACH,wBAAgB,gBAAgB,CAAC,YAAY,EAAE,MAAM,EAAE,iBAAiB,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAG/F;AAID,MAAM,WAAW,UAAU;IACzB,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;CACf;AAED;;;;GAIG;AACH,wBAAgB,QAAQ,CAAC,KAAK,EAAE;IAAE,UAAU,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAA;CAAE,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,UAAU,EAAE,CAKnH"}

package/lib/dist/lib/runInsights.js

@@ -0,0 +1,185 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+const UNCATEGORIZED = 'Uncategorized';
+/**
+ * Group rows by test-case category and tally pass/fail/errored/total.
+ * Deterministic order: largest category first, ties broken alphabetically.
+ */
+export function computeCategoryBars(rows) {
+    const map = new Map();
+    for (const row of rows) {
+        const category = (row.category || '').trim() || UNCATEGORIZED;
+        let bar = map.get(category);
+        if (!bar) {
+            bar = { category, passed: 0, failed: 0, errored: 0, total: 0 };
+            map.set(category, bar);
+        }
+        bar.total++;
+        if (row.status === 'passed')
+            bar.passed++;
+        else if (row.status === 'errored')
+            bar.errored++;
+        else if (row.status === 'failed')
+            bar.failed++;
+    }
+    return Array.from(map.values()).sort((a, b) => b.total - a.total || a.category.localeCompare(b.category));
+}
+// ── Failure-theme clustering ───────────────────────────────────────────────
+/**
+ * Normalize a judge-reasoning string down to its first sentence,
+ * lowercased, punctuation-stripped, whitespace-collapsed. Used both as the
+ * clustering input and as the theme's stable `key`.
+ */
+export function normalizeReasoningKey(reasoning) {
+    const trimmed = (reasoning || '').trim();
+    if (!trimmed)
+        return '';
+    const sentenceMatch = trimmed.match(/^[^.!?\n]+[.!?]?/);
+    const firstSentence = (sentenceMatch ? sentenceMatch[0] : trimmed).toLowerCase();
+    return firstSentence.replace(/[^a-z0-9\s]/g, ' ').replace(/\s+/g, ' ').trim();
+}
+/** Returns the literal (non-lowercased) first sentence, trimmed — used as the human-facing sample snippet. */
+function literalFirstSentence(reasoning) {
+    const trimmed = (reasoning || '').trim();
+    if (!trimmed)
+        return '';
+    const sentenceMatch = trimmed.match(/^[^.!?\n]+[.!?]?/);
+    return (sentenceMatch ? sentenceMatch[0] : trimmed).trim();
+}
+function shingles(normalized, size) {
+    const words = normalized.split(' ').filter(Boolean);
+    if (words.length === 0)
+        return [];
+    if (words.length <= size)
+        return [words.join(' ')];
+    const out = [];
+    for (let i = 0; i <= words.length - size; i++) {
+        out.push(words.slice(i, i + size).join(' '));
+    }
+    return out;
+}
+/** Contiguous-word window size used for near-duplicate matching. Paraphrases of the same failure ("agent was unable to retrieve..." vs "agent failed to retrieve...") reliably share at least one 6-word run even when their leading words differ. */
+const DEFAULT_SHINGLE_SIZE = 6;
+/** Reasoning first-sentences shorter than this many words don't carry enough signal to safely dedupe against unrelated failures — left as their own singleton theme rather than risk clustering unrelated short reasons together (e.g. "The agent failed to retrieve the required information."). */
+const MIN_WORDS_FOR_SHINGLING = 3;
+/**
+ * Cluster failing test cases into "why they failed" themes using a
+ * deterministic, LLM-free heuristic: normalize each case's judge-reasoning
+ * first sentence, then union-find cases that share at least one contiguous
+ * N-word shingle. This is robust to minor paraphrasing (a judge saying
+ * "unable to retrieve" vs "failed to retrieve" the same underlying tool
+ * connectivity failure) while still keeping genuinely distinct failure
+ * modes (e.g. "missing required facts" vs "MCP server unavailable")
+ * separate, because they share no contiguous phrase.
+ *
+ * Verified against a real production run (418-verify, 64 failing cases):
+ * 57 of 64 connectivity-flavored reasonings collapse into ONE dominant
+ * theme; the remaining 7 ("Required facts evaluation: ...", a genuinely
+ * different failure shape) form a second, correctly separate theme.
+ *
+ * Output order: largest theme first, ties broken by the theme's
+ * lowest-sorting testCaseId (deterministic, no dependency on input order).
+ */
+export function clusterFailureThemes(items, shingleSize = DEFAULT_SHINGLE_SIZE) {
+    const n = items.length;
+    if (n === 0)
+        return [];
+    const parent = Array.from({ length: n }, (_, i) => i);
+    function find(i) {
+        while (parent[i] !== i) {
+            parent[i] = parent[parent[i]];
+            i = parent[i];
+        }
+        return i;
+    }
+    function union(a, b) {
+        const ra = find(a);
+        const rb = find(b);
+        if (ra !== rb)
+            parent[Math.max(ra, rb)] = Math.min(ra, rb);
+    }
+    const normalized = items.map(it => normalizeReasoningKey(it.reasoning));
+    const shingleBuckets = new Map(); // shingle -> first index seen
+    normalized.forEach((norm, idx) => {
+        const words = norm.split(' ').filter(Boolean);
+        if (words.length < MIN_WORDS_FOR_SHINGLING)
+            return;
+        for (const sh of shingles(norm, shingleSize)) {
+            const firstIdx = shingleBuckets.get(sh);
+            if (firstIdx === undefined)
+                shingleBuckets.set(sh, idx);
+            else
+                union(firstIdx, idx);
+        }
+    });
+    const groups = new Map();
+    for (let i = 0; i < n; i++) {
+        const root = find(i);
+        const arr = groups.get(root);
+        if (arr)
+            arr.push(i);
+        else
+            groups.set(root, [i]);
+    }
+    const themes = [];
+    for (const idxs of groups.values()) {
+        // Representative snippet: the most frequent literal first-sentence
+        // phrasing within the cluster (ties broken by first occurrence) - so a
+        // theme with 4 identical sentences + a handful of near-duplicate
+        // paraphrases surfaces the exact sentence, not an arbitrary member.
+        const counts = new Map();
+        for (const idx of idxs) {
+            const key = normalized[idx];
+            const snippet = literalFirstSentence(items[idx].reasoning);
+            const existing = counts.get(key);
+            if (existing)
+                existing.count++;
+            else
+                counts.set(key, { count: 1, firstIdx: idx, snippet });
+        }
+        let best = null;
+        for (const v of counts.values()) {
+            if (!best || v.count > best.count || (v.count === best.count && v.firstIdx < best.firstIdx))
+                best = v;
+        }
+        const testCaseIds = idxs.map(idx => items[idx].testCaseId);
+        themes.push({
+            key: normalized[idxs[0]] || `theme-${idxs[0]}`,
+            count: idxs.length,
+            sampleSnippet: best?.snippet || '',
+            testCaseIds,
+        });
+    }
+    themes.sort((a, b) => {
+        if (b.count !== a.count)
+            return b.count - a.count;
+        const aMin = [...a.testCaseIds].sort()[0] || '';
+        const bMin = [...b.testCaseIds].sort()[0] || '';
+        return aMin < bMin ? -1 : aMin > bMin ? 1 : 0;
+    });
+    return themes;
+}
+/**
+ * "Based on N of M failing cases" note shown under the theme list when the
+ * reasoning fetch was capped (RunInsightsPane caps at the first 100 failing
+ * cases). Returns null when nothing was capped (fetchedCount >= totalCount).
+ */
+export function formatCappedNote(fetchedCount, totalFailingCount) {
+    if (totalFailingCount <= fetchedCount)
+        return null;
+    return `Based on ${fetchedCount} of ${totalFailingCount} failing cases`;
+}
+/**
+ * Top-N cases by a numeric value (duration, cost, ...), descending.
+ * Cases with a null/undefined/NaN value are excluded. Deterministic tie
+ * break: lower testCaseId first.
+ */
+export function pickTopN(cases, n) {
+    return cases
+        .filter((c) => typeof c.value === 'number' && Number.isFinite(c.value))
+        .sort((a, b) => (b.value !== a.value ? b.value - a.value : a.testCaseId.localeCompare(b.testCaseId)))
+        .slice(0, n);
+}
+//# sourceMappingURL=runInsights.js.map

package/lib/dist/lib/runInsights.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"runInsights.js","sourceRoot":"","sources":["../../runInsights.ts"],"names":[],"mappings":"AAAA;;;GAGG;AA+BH,MAAM,aAAa,GAAG,eAAe,CAAC;AAEtC;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,IAAyB;IAC3D,MAAM,GAAG,GAAG,IAAI,GAAG,EAAuB,CAAC;IAC3C,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,QAAQ,GAAG,CAAC,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,IAAI,aAAa,CAAC;QAC9D,IAAI,GAAG,GAAG,GAAG,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;QAC5B,IAAI,CAAC,GAAG,EAAE,CAAC;YACT,GAAG,GAAG,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;YAC/D,GAAG,CAAC,GAAG,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QACzB,CAAC;QACD,GAAG,CAAC,KAAK,EAAE,CAAC;QACZ,IAAI,GAAG,CAAC,MAAM,KAAK,QAAQ;YAAE,GAAG,CAAC,MAAM,EAAE,CAAC;aACrC,IAAI,GAAG,CAAC,MAAM,KAAK,SAAS;YAAE,GAAG,CAAC,OAAO,EAAE,CAAC;aAC5C,IAAI,GAAG,CAAC,MAAM,KAAK,QAAQ;YAAE,GAAG,CAAC,MAAM,EAAE,CAAC;IACjD,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC,IAAI,CAClC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,QAAQ,CAAC,aAAa,CAAC,CAAC,CAAC,QAAQ,CAAC,CACpE,CAAC;AACJ,CAAC;AAED,8EAA8E;AAE9E;;;;GAIG;AACH,MAAM,UAAU,qBAAqB,CAAC,SAAiB;IACrD,MAAM,OAAO,GAAG,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACzC,IAAI,CAAC,OAAO;QAAE,OAAO,EAAE,CAAC;IACxB,MAAM,aAAa,GAAG,OAAO,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC;IACxD,MAAM,aAAa,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,WAAW,EAAE,CAAC;IACjF,OAAO,aAAa,CAAC,OAAO,CAAC,cAAc,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;AAChF,CAAC;AAED,8GAA8G;AAC9G,SAAS,oBAAoB,CAAC,SAAiB;IAC7C,MAAM,OAAO,GAAG,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACzC,IAAI,CAAC,OAAO;QAAE,OAAO,EAAE,CAAC;IACxB,MAAM,aAAa,GAAG,OAAO,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC;IACxD,OAAO,CAAC,aAAa,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC;AAC7D,CAAC;AAED,SAAS,QAAQ,CAAC,UAAkB,EAAE,IAAY;IAChD,MAAM,KAAK,GAAG,UAAU,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IACpD,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAClC,IAAI,KAAK,CAAC,MAAM,IAAI,IAAI;QAAE,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;IACnD,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,KAAK,CAAC,MAAM,GAAG,IAAI,EAAE,CAAC,EAAE,EAAE,CAAC;QAC9C,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;IAC/C,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAgBD,sPAAsP;AACtP,MAAM,oBAAoB,GAAG,CAAC,CAAC;AAC/B,qSAAqS;AACrS,MAAM,uBAAuB,GAAG,CAAC,CAAC;AAElC;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,UAAU,oBAAoB,CAClC,KAA0B,EAC1B,cAAsB,oBAAoB;IAE1C,MAAM,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC;IACvB,IAAI,CAAC,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAEvB,MAAM,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;IACtD,SAAS,IAAI,CAAC,CAAS;QACrB,OAAO,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC;YACvB,MAAM,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;YAC9B,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;QAChB,CAAC;QACD,OAAO,CAAC,CAAC;IACX,CAAC;IACD,SAAS,KAAK,CAAC,CAAS,EAAE,CAAS;QACjC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QACnB,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QACnB,IAAI,EAAE,KAAK,EAAE;YAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;IAC7D,CAAC;IAED,MAAM,UAAU,GAAG,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,qBAAqB,CAAC,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;IACxE,MAAM,cAAc,GAAG,IAAI,GAAG,EAAkB,CAAC,CAAC,8BAA8B;IAChF,UAAU,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,GAAG,EAAE,EAAE;QAC/B,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QAC9C,IAAI,KAAK,CAAC,MAAM,GAAG,uBAAuB;YAAE,OAAO;QACnD,KAAK,MAAM,EAAE,IAAI,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAC,EAAE,CAAC;YAC7C,MAAM,QAAQ,GAAG,cAAc,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YACxC,IAAI,QAAQ,KAAK,SAAS;gBAAE,cAAc,CAAC,GAAG,CAAC,EAAE,EAAE,GAAG,CAAC,CAAC;;gBACnD,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QAC5B,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,MAAM,MAAM,GAAG,IAAI,GAAG,EAAoB,CAAC;IAC3C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,MAAM,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QACrB,MAAM,GAAG,GAAG,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAC7B,IAAI,GAAG;YAAE,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;;YAChB,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC7B,CAAC;IAED,MAAM,MAAM,GAAmB,EAAE,CAAC;IAClC,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,MAAM,EAAE,EAAE,CAAC;QACnC,mEAAmE;QACnE,uEAAuE;QACvE,iEAAiE;QACjE,oEAAoE;QACpE,MAAM,MAAM,GAAG,IAAI,GAAG,EAAgE,CAAC;QACvF,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;YACvB,MAAM,GAAG,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC;YAC5B,MAAM,OAAO,GAAG,oBAAoB,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,SAAS,CAAC,CAAC;YAC3D,MAAM,QAAQ,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YACjC,IAAI,QAAQ;gBAAE,QAAQ,CAAC,KAAK,EAAE,CAAC;;gBAC1B,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,QAAQ,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,CAAC;QAC7D,CAAC;QACD,IAAI,IAAI,GAAgE,IAAI,CAAC;QAC7E,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,MAAM,EAAE,EAAE,CAAC;YAChC,IAAI,CAAC,IAAI,IAAI,CAAC,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC,CAAC,KAAK,KAAK,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC;gBAAE,IAAI,GAAG,CAAC,CAAC;QACxG,CAAC;QACD,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,UAAU,CAAC,CAAC;QAC3D,MAAM,CAAC,IAAI,CAAC;YACV,GAAG,EAAE,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,SAAS,IAAI,CAAC,CAAC,CAAC,EAAE;YAC9C,KAAK,EAAE,IAAI,CAAC,MAAM;YAClB,aAAa,EAAE,IAAI,EAAE,OAAO,IAAI,EAAE;YAClC,WAAW;SACZ,CAAC,CAAC;IACL,CAAC;IAED,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACnB,IAAI,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,KAAK;YAAE,OAAO,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC;QAClD,MAAM,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC,WAAW,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAChD,MAAM,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC,WAAW,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAChD,OAAO,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAChD,CAAC,CAAC,CAAC;IACH,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAAC,YAAoB,EAAE,iBAAyB;IAC9E,IAAI,iBAAiB,IAAI,YAAY;QAAE,OAAO,IAAI,CAAC;IACnD,OAAO,YAAY,YAAY,OAAO,iBAAiB,gBAAgB,CAAC;AAC1E,CAAC;AASD;;;;GAIG;AACH,MAAM,UAAU,QAAQ,CAAC,KAAiE,EAAE,CAAS;IACnG,OAAO,KAAK;SACT,MAAM,CAAC,CAAC,CAAC,EAAmB,EAAE,CAAC,OAAO,CAAC,CAAC,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC;SACvF,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,aAAa,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC;SACpG,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;AACjB,CAAC"}

package/lib/dist/lib/runName.d.ts

@@ -0,0 +1,29 @@
+/**
+ * Validation for renaming an evaluation run (PATCH /api/storage/evaluation-runs/:id
+ * `{ name }`). This is the server route's authoritative check.
+ *
+ * Rename is intentionally narrow: it only ever touches `name`. No version bump,
+ * no stats recompute — this file has no opinion on anything but the string.
+ */
+/** Renamed evaluation runs are capped at this length (arbitrary but generous —
+ *  long enough for any real run name, short enough to keep list/table UIs sane). */
+export declare const MAX_RUN_NAME_LENGTH = 200;
+export type RunNameValidation = {
+    ok: true;
+    value: string;
+} | {
+    ok: false;
+    error: string;
+};
+/**
+ * Validate + normalize a candidate evaluation-run name.
+ *
+ * - Must be a string.
+ * - Trimmed value must be non-empty.
+ * - Trimmed value must be at most {@link MAX_RUN_NAME_LENGTH} characters.
+ *
+ * Returns the trimmed value on success so callers persist a normalized name
+ * (no leading/trailing whitespace) rather than the raw input.
+ */
+export declare function validateRunNameUpdate(name: unknown): RunNameValidation;
+//# sourceMappingURL=runName.d.ts.map

package/lib/dist/lib/runName.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runName.d.ts","sourceRoot":"","sources":["../../runName.ts"],"names":[],"mappings":"AAKA;;;;;;GAMG;AAEH;oFACoF;AACpF,eAAO,MAAM,mBAAmB,MAAM,CAAC;AAEvC,MAAM,MAAM,iBAAiB,GACzB;IAAE,EAAE,EAAE,IAAI,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAC3B;IAAE,EAAE,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC;AAEjC;;;;;;;;;GASG;AACH,wBAAgB,qBAAqB,CAAC,IAAI,EAAE,OAAO,GAAG,iBAAiB,CAYtE"}

package/lib/dist/lib/runName.js

@@ -0,0 +1,38 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+/**
+ * Validation for renaming an evaluation run (PATCH /api/storage/evaluation-runs/:id
+ * `{ name }`). This is the server route's authoritative check.
+ *
+ * Rename is intentionally narrow: it only ever touches `name`. No version bump,
+ * no stats recompute — this file has no opinion on anything but the string.
+ */
+/** Renamed evaluation runs are capped at this length (arbitrary but generous —
+ *  long enough for any real run name, short enough to keep list/table UIs sane). */
+export const MAX_RUN_NAME_LENGTH = 200;
+/**
+ * Validate + normalize a candidate evaluation-run name.
+ *
+ * - Must be a string.
+ * - Trimmed value must be non-empty.
+ * - Trimmed value must be at most {@link MAX_RUN_NAME_LENGTH} characters.
+ *
+ * Returns the trimmed value on success so callers persist a normalized name
+ * (no leading/trailing whitespace) rather than the raw input.
+ */
+export function validateRunNameUpdate(name) {
+    if (typeof name !== 'string') {
+        return { ok: false, error: 'name must be a string' };
+    }
+    const trimmed = name.trim();
+    if (!trimmed) {
+        return { ok: false, error: 'name must not be empty' };
+    }
+    if (trimmed.length > MAX_RUN_NAME_LENGTH) {
+        return { ok: false, error: `name must be ${MAX_RUN_NAME_LENGTH} characters or fewer` };
+    }
+    return { ok: true, value: trimmed };
+}
+//# sourceMappingURL=runName.js.map

package/lib/dist/lib/runName.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"runName.js","sourceRoot":"","sources":["../../runName.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH;;;;;;GAMG;AAEH;oFACoF;AACpF,MAAM,CAAC,MAAM,mBAAmB,GAAG,GAAG,CAAC;AAMvC;;;;;;;;;GASG;AACH,MAAM,UAAU,qBAAqB,CAAC,IAAa;IACjD,IAAI,OAAO,IAAI,KAAK,QAAQ,EAAE,CAAC;QAC7B,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,uBAAuB,EAAE,CAAC;IACvD,CAAC;IACD,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,IAAI,CAAC,OAAO,EAAE,CAAC;QACb,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,wBAAwB,EAAE,CAAC;IACxD,CAAC;IACD,IAAI,OAAO,CAAC,MAAM,GAAG,mBAAmB,EAAE,CAAC;QACzC,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,gBAAgB,mBAAmB,sBAAsB,EAAE,CAAC;IACzF,CAAC;IACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;AACtC,CAAC"}

package/lib/dist/lib/runReportPath.d.ts

@@ -0,0 +1,16 @@
+/**
+ * Canonical run-report path for a run surfaced on the compare page.
+ *
+ * Benchmark runs (embedded in a benchmark's `runs[]`) resolve at
+ * `/evaluations/benchmarks/:benchmarkId/runs/:runId`; everything else —
+ * ad-hoc / SDK eval-runs, or runs whose `benchmarkId` is merely a label — goes
+ * to the bare `/evaluations/runs/:runId` route (the benchmark route would
+ * 404/redirect for those, and the bare route 404s for benchmark run ids).
+ *
+ * Shared by the scoreboard's run-name link, its "Open run" icon, and the
+ * per-case table's run headers so the three can never disagree. Callers pass
+ * the benchmarkId from `ComparisonPage`'s `runBenchmarkIdById` map (which is
+ * `undefined` unless the run is a real benchmark member).
+ */
+export declare const runReportPath: (runId: string, benchmarkId?: string) => string;
+//# sourceMappingURL=runReportPath.d.ts.map

package/lib/dist/lib/runReportPath.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runReportPath.d.ts","sourceRoot":"","sources":["../../runReportPath.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;;;GAaG;AACH,eAAO,MAAM,aAAa,GAAI,OAAO,MAAM,EAAE,cAAc,MAAM,KAAG,MAGd,CAAC"}