@opensearch-project/agent-health 0.5.2 → 0.7.0

package/lib/dist/lib/agentTrends.d.ts

@@ -0,0 +1,210 @@
+/**
+ * Agent Trends Band — Aggregation Utilities
+ *
+ * Builds the per-run, per-agent data model behind the landing-page
+ * "Agent trends" band: v3 replaces the all-agents overlay chart (one
+ * ECharts line+scatter series per agent, sharing a 10-color palette) with
+ * a small-multiples sparkline-table — one row per agent, sorted by latest
+ * score or biggest drop, each row's sparkline + a single-agent focused
+ * detail chart built from the SAME gap-broken series shape.
+ *
+ * Deliberately kept UI-framework-free so it's cheaply unit testable and
+ * so it can be reused by the row list, the drawer, and the detail chart.
+ *
+ * Data-source note: accuracy/pass counts come straight from
+ * `Benchmark.runs[]` (already loaded by the Dashboard for the runs list —
+ * no extra fetch). Cost/tokens are trace-derived and only available for
+ * runs whose reports were resolved in the caller's `metricsMap` (see
+ * services/metrics.ts#fetchBatchMetrics); a run reports `costUsd`/`tokens`
+ * only when EVERY one of its reports resolved a match — a partial sum
+ * would understate the run's true cost while looking complete, which is
+ * worse than admitting "unknown" — so partially- or fully-unmatched runs
+ * get `null` rather than a fabricated or partial total, and callers render
+ * an honest "—" instead.
+ */
+import type { Benchmark, BenchmarkRun, EvaluationReport } from '../types/index.js';
+export type TrendMetricKey = 'accuracy' | 'cost' | 'tokens';
+export type TrendsTimeRange = '7d' | '30d' | '90d';
+export type TrendSortMode = 'latest' | 'biggestDrop';
+export interface RunMetricsLookup {
+    costUsd: number;
+    tokens: number;
+}
+/** One dot on the trend chart: a single BenchmarkRun for one agent. */
+export interface AgentRunPoint {
+    runDocId: string;
+    benchmarkId: string;
+    benchmarkName: string;
+    agentKey: string;
+    agentName: string;
+    modelId: string;
+    createdAt: string;
+    timestamp: number;
+    passed: number;
+    failed: number;
+    total: number;
+    /**
+     * 0-100 pass rate over the evaluable set, or `null` when the run had
+     * ZERO evaluable test cases (every case errored — no judge verdict at
+     * all). Owner-reported bug: this used to default to `0`, which is
+     * visually indistinguishable from a genuine "the agent passed nothing"
+     * result — a run with no signal must never render as a real data point.
+     */
+    accuracyPct: number | null;
+    costUsd: number | null;
+    tokens: number | null;
+}
+/**
+ * Deterministic, visually distinct color palette for the chart lines / chip
+ * left-borders. Assigned by sorted agentKey so colors stay stable across
+ * re-renders and time-range/benchmark filtering (not by first-seen order,
+ * which would shuffle as the visible run set changes).
+ *
+ * v3 note: the small-multiples redesign no longer overlays every agent's
+ * line on one chart, so per-agent hue no longer needs to scale to N agents
+ * (the old 10-color palette repeated after the 10th agent — "indistinguishable
+ * colors" in the owner's feedback with 14 agents in scope). Each row/detail
+ * chart is already labeled by name, so a single, theme-aware accent color
+ * plus semantic (up/down) delta coloring is used everywhere instead — see
+ * `TREND_LINE_COLOR` in AgentTrendSparkline/AgentTrendDetailChart.
+ */
+/**
+ * Compute passed/failed/total/accuracyPct for a run, preferring the
+ * denormalized `run.stats` (fast path) and falling back to bucketing
+ * `run.results` directly (same canonical logic used by the runs list) —
+ * no report fetch required either way.
+ */
+export declare function getRunAccuracy(run: BenchmarkRun): {
+    passed: number;
+    failed: number;
+    total: number;
+    accuracyPct: number | null;
+};
+export declare function timeRangeToSinceMs(range: TrendsTimeRange, nowMs?: number): number;
+export interface BuildAgentRunPointsOptions {
+    benchmarkId?: string | null;
+    sinceMs?: number | null;
+    agentDisplayName?: (agentKey: string) => string;
+}
+/**
+ * Build one AgentRunPoint per BenchmarkRun (skipping runs with zero test
+ * cases — e.g. still-provisioning or cancelled-before-start runs), sorted
+ * ascending by time.
+ */
+export declare function buildAgentRunPoints(benchmarks: Benchmark[], reports: EvaluationReport[], metricsMap: Map<string, RunMetricsLookup>, options?: BuildAgentRunPointsOptions): AgentRunPoint[];
+/** Group already time-sorted points by agentKey, preserving relative order. */
+export declare function groupPointsByAgent(points: AgentRunPoint[]): Map<string, AgentRunPoint[]>;
+export declare function metricValue(point: AgentRunPoint, metric: TrendMetricKey): number | null;
+export declare function formatMetricValue(metric: TrendMetricKey, value: number): string;
+/** One entry in a gap-broken trend series. `value: null` marks a synthetic break, never a real run. */
+export interface TrendSeriesEntry {
+    timestamp: number;
+    value: number | null;
+    point: AgentRunPoint | null;
+}
+/**
+ * Build one agent's gap-broken series for `metric`: points whose metric
+ * value is null (no evaluable accuracy, or no trace-resolved cost/tokens)
+ * are dropped — never plotted as a fake zero. A synthetic `{ value: null }`
+ * marker is inserted at the midpoint between any two consecutive PLOTTED
+ * points whose time gap exceeds `gapBreakMs` (default 7 days).
+ *
+ * Why: a chart line drawn between two data-array entries connects them
+ * with a straight (or smoothed) segment regardless of how far apart their
+ * x-values are — so two runs 25 days apart look, visually, like a
+ * continuous trend across those 25 days ("sparse runs produce misleading
+ * long interpolated lines" — owner feedback). Inserting a null-valued
+ * entry between them gives recharts' `<Line connectNulls={false}>` (the
+ * default) an explicit place to break the path, while the surrounding
+ * dead space still renders proportionally on a time-scaled x-axis.
+ */
+export declare function buildGapBrokenSeries(agentPoints: AgentRunPoint[], // already time-sorted, single agent
+metric: TrendMetricKey, gapBreakMs?: number): TrendSeriesEntry[];
+/** Latest-vs-previous delta for one agent's metric series (not week-over-week — the immediately prior run). */
+export interface TrendDelta {
+    latestValue: number | null;
+    previousValue: number | null;
+    /** latestValue - previousValue; null unless BOTH resolved to a real value. */
+    delta: number | null;
+}
+export declare function computeLatestDelta(agentPoints: AgentRunPoint[], metric: TrendMetricKey): TrendDelta;
+export declare function formatDelta(metric: TrendMetricKey, delta: number | null): string;
+/** One row in the sparkline-table: one agent, its gap-broken series for the current metric, latest value + delta. */
+export interface AgentTrendRow {
+    agentKey: string;
+    agentName: string;
+    runCount: number;
+    latestRunAt: string | null;
+    latestRunDocId: string | null;
+    latestBenchmarkId: string | null;
+    /** All of this agent's points, time-sorted (unfiltered by metric) — the detail chart re-derives per-metric series from this on focus. */
+    points: AgentRunPoint[];
+    /** Gap-broken series for the CURRENTLY selected metric — what the row's sparkline draws. */
+    series: TrendSeriesEntry[];
+    latestValue: number | null;
+    previousValue: number | null;
+    delta: number | null;
+}
+/**
+ * Group `points` by agent and shape each into a sparkline-table row for
+ * the given metric. One row per agent that has at least one point in
+ * scope — a single run is a perfectly valid row (a lone dot, `delta: null`
+ * rendered as "n/a"); there is no "not enough data" floor at the row
+ * level (unlike the old all-agents chart, which needed >=2 points across
+ * ALL agents just to avoid an empty canvas).
+ */
+export declare function buildAgentTrendRows(points: AgentRunPoint[], metric: TrendMetricKey, gapBreakMs?: number): AgentTrendRow[];
+/**
+ * Sort rows by latest value (highest first; rows with no value for the
+ * current metric sort last) or by biggest drop (most negative delta
+ * first; rows with no delta sort last, ordered among themselves by latest
+ * value so a still-informative "no signal yet" row isn't buried below a
+ * literal `undefined`-sorts-arbitrarily bucket).
+ */
+export declare function sortAgentTrendRows(rows: AgentTrendRow[], sortMode: TrendSortMode): AgentTrendRow[];
+/** Default benchmark selector value: the benchmark with the most recent run. */
+export declare function getMostRecentlyActiveBenchmarkId(benchmarks: Benchmark[]): string | null;
+/** One plottable dot: a run + its resolved metric value (metric-null points are never included). */
+export interface DotPlotEntry {
+    point: AgentRunPoint;
+    value: number;
+}
+/** One row of the ranked dot plot: one agent, its latest dot (solid) + earlier dots (faded). */
+export interface BenchmarkDotPlotRow {
+    agentKey: string;
+    agentName: string;
+    /** null when this agent has no run with a resolved value for the current metric (e.g. no trace-derived cost yet). */
+    latest: DotPlotEntry | null;
+    /** Earlier runs with a resolved value, time-ascending, EXCLUDING `latest`. */
+    history: DotPlotEntry[];
+    /** 0-based rank after `rankDotPlotRows` (0 = top/best); -1 until ranked. */
+    rank: number;
+}
+/**
+ * Group `points` (already scoped to one benchmark by the caller) by agent
+ * and split each agent's valued runs into `latest` (most recent with a
+ * resolved value) + `history` (earlier ones). An agent with zero valued
+ * runs for this metric still gets a row (`latest: null`) so it's visible
+ * in the list rather than silently disappearing when switching metrics.
+ */
+export declare function buildBenchmarkDotPlotRows(points: AgentRunPoint[], metric: TrendMetricKey): BenchmarkDotPlotRow[];
+/**
+ * Rank rows by latest value, best (highest) first — rows with no value for
+ * the current metric sort last (still listed, just unranked among
+ * themselves by name for a stable order). Stamps `rank` (0 = top) on the
+ * returned copies; does not mutate the input.
+ */
+export declare function rankDotPlotRows(rows: BenchmarkDotPlotRow[]): BenchmarkDotPlotRow[];
+/**
+ * Shared X-domain for a dot plot: accuracy is always anchored to the true
+ * [0, 100] scale (auto-scaling to the observed min/max would visually
+ * exaggerate small real-world differences, e.g. an 88-95% cluster would
+ * stretch to fill the whole width). Cost/tokens are zero-anchored (0 is a
+ * meaningful reference point for both) up to the observed max plus a
+ * little headroom so the top dot isn't flush against the edge. Returns
+ * `null` when there is nothing to plot at all.
+ */
+export declare function metricDomain(rows: BenchmarkDotPlotRow[], metric: TrendMetricKey): [number, number] | null;
+/** Map a value into a 0-100 position within `domain`, clamped — usable directly as a CSS `left`/`width` percentage. */
+export declare function valueToPercent(value: number, domain: [number, number]): number;
+//# sourceMappingURL=agentTrends.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"agentTrends.d.ts","sourceRoot":"","sources":["../../agentTrends.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAEH,OAAO,KAAK,EAAE,SAAS,EAAE,YAAY,EAAE,gBAAgB,EAAE,MAAM,SAAS,CAAC;AAIzE,MAAM,MAAM,cAAc,GAAG,UAAU,GAAG,MAAM,GAAG,QAAQ,CAAC;AAE5D,MAAM,MAAM,eAAe,GAAG,IAAI,GAAG,KAAK,GAAG,KAAK,CAAC;AAEnD,MAAM,MAAM,aAAa,GAAG,QAAQ,GAAG,aAAa,CAAC;AAErD,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,uEAAuE;AACvE,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd;;;;;;OAMG;IACH,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;CACvB;AAID;;;;;;;;;;;;;GAaG;AAEH;;;;;GAKG;AACH,wBAAgB,cAAc,CAAC,GAAG,EAAE,YAAY,GAAG;IACjD,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;CAC5B,CAUA;AAED,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,eAAe,EAAE,KAAK,GAAE,MAAmB,GAAG,MAAM,CAG7F;AAED,MAAM,WAAW,0BAA0B;IACzC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,gBAAgB,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,MAAM,CAAC;CACjD;AAED;;;;GAIG;AACH,wBAAgB,mBAAmB,CACjC,UAAU,EAAE,SAAS,EAAE,EACvB,OAAO,EAAE,gBAAgB,EAAE,EAC3B,UAAU,EAAE,GAAG,CAAC,MAAM,EAAE,gBAAgB,CAAC,EACzC,OAAO,GAAE,0BAA+B,GACvC,aAAa,EAAE,CAgEjB;AAED,+EAA+E;AAC/E,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,aAAa,EAAE,GAAG,GAAG,CAAC,MAAM,EAAE,aAAa,EAAE,CAAC,CAQxF;AAED,wBAAgB,WAAW,CAAC,KAAK,EAAE,aAAa,EAAE,MAAM,EAAE,cAAc,GAAG,MAAM,GAAG,IAAI,CAWvF;AAED,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,cAAc,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,CAW/E;AAED,uGAAuG;AACvG,MAAM,WAAW,gBAAgB;IAC/B,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,KAAK,EAAE,aAAa,GAAG,IAAI,CAAC;CAC7B;AAED;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,oBAAoB,CAClC,WAAW,EAAE,aAAa,EAAE,EAAE,oCAAoC;AAClE,MAAM,EAAE,cAAc,EACtB,UAAU,GAAE,MAA6B,GACxC,gBAAgB,EAAE,CAiBpB;AAED,+GAA+G;AAC/G,MAAM,WAAW,UAAU;IACzB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,8EAA8E;IAC9E,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;CACtB;AAED,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,aAAa,EAAE,EAAE,MAAM,EAAE,cAAc,GAAG,UAAU,CASnG;AAED,wBAAgB,WAAW,CAAC,MAAM,EAAE,cAAc,EAAE,KAAK,EAAE,MAAM,GAAG,IAAI,GAAG,MAAM,CAShF;AAED,qHAAqH;AACrH,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,yIAAyI;IACzI,MAAM,EAAE,aAAa,EAAE,CAAC;IACxB,4FAA4F;IAC5F,MAAM,EAAE,gBAAgB,EAAE,CAAC;IAC3B,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;CACtB;AAED;;;;;;;GAOG;AACH,wBAAgB,mBAAmB,CACjC,MAAM,EAAE,aAAa,EAAE,EACvB,MAAM,EAAE,cAAc,EACtB,UAAU,GAAE,MAA6B,GACxC,aAAa,EAAE,CAsBjB;AAED;;;;;;GAMG;AACH,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,aAAa,EAAE,EAAE,QAAQ,EAAE,aAAa,GAAG,aAAa,EAAE,CAkBlG;AAED,gFAAgF;AAChF,wBAAgB,gCAAgC,CAAC,UAAU,EAAE,SAAS,EAAE,GAAG,MAAM,GAAG,IAAI,CAavF;AAWD,oGAAoG;AACpG,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,aAAa,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,gGAAgG;AAChG,MAAM,WAAW,mBAAmB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,qHAAqH;IACrH,MAAM,EAAE,YAAY,GAAG,IAAI,CAAC;IAC5B,8EAA8E;IAC9E,OAAO,EAAE,YAAY,EAAE,CAAC;IACxB,4EAA4E;IAC5E,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;GAMG;AACH,wBAAgB,yBAAyB,CACvC,MAAM,EAAE,aAAa,EAAE,EACvB,MAAM,EAAE,cAAc,GACrB,mBAAmB,EAAE,CAkBvB;AAED;;;;;GAKG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,mBAAmB,EAAE,GAAG,mBAAmB,EAAE,CAQlF;AAED;;;;;;;;GAQG;AACH,wBAAgB,YAAY,CAAC,IAAI,EAAE,mBAAmB,EAAE,EAAE,MAAM,EAAE,cAAc,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,IAAI,CAUzG;AAED,uHAAuH;AACvH,wBAAgB,cAAc,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,MAAM,CAK9E"}

package/lib/dist/lib/agentTrends.js

@@ -0,0 +1,360 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+import { bucketRunResults } from '../lib/runStats.js';
+import { formatCost, formatTokens } from '../services/metrics.js';
+const GAP_BREAK_MS_DEFAULT = 7 * 24 * 60 * 60 * 1000; // 7 days
+/**
+ * Deterministic, visually distinct color palette for the chart lines / chip
+ * left-borders. Assigned by sorted agentKey so colors stay stable across
+ * re-renders and time-range/benchmark filtering (not by first-seen order,
+ * which would shuffle as the visible run set changes).
+ *
+ * v3 note: the small-multiples redesign no longer overlays every agent's
+ * line on one chart, so per-agent hue no longer needs to scale to N agents
+ * (the old 10-color palette repeated after the 10th agent — "indistinguishable
+ * colors" in the owner's feedback with 14 agents in scope). Each row/detail
+ * chart is already labeled by name, so a single, theme-aware accent color
+ * plus semantic (up/down) delta coloring is used everywhere instead — see
+ * `TREND_LINE_COLOR` in AgentTrendSparkline/AgentTrendDetailChart.
+ */
+/**
+ * Compute passed/failed/total/accuracyPct for a run, preferring the
+ * denormalized `run.stats` (fast path) and falling back to bucketing
+ * `run.results` directly (same canonical logic used by the runs list) —
+ * no report fetch required either way.
+ */
+export function getRunAccuracy(run) {
+    if (run.stats && run.stats.total > 0) {
+        const evaluable = Math.max(0, run.stats.total - (run.stats.errored || 0));
+        const accuracyPct = evaluable > 0 ? (run.stats.passed / evaluable) * 100 : null;
+        return { passed: run.stats.passed, failed: run.stats.failed, total: run.stats.total, accuracyPct };
+    }
+    const bucketed = bucketRunResults(run.results);
+    const evaluable = Math.max(0, bucketed.total - bucketed.errored);
+    const accuracyPct = evaluable > 0 ? (bucketed.passed / evaluable) * 100 : null;
+    return { passed: bucketed.passed, failed: bucketed.failed, total: bucketed.total, accuracyPct };
+}
+export function timeRangeToSinceMs(range, nowMs = Date.now()) {
+    const days = range === '7d' ? 7 : range === '90d' ? 90 : 30;
+    return nowMs - days * 24 * 60 * 60 * 1000;
+}
+/**
+ * Build one AgentRunPoint per BenchmarkRun (skipping runs with zero test
+ * cases — e.g. still-provisioning or cancelled-before-start runs), sorted
+ * ascending by time.
+ */
+export function buildAgentRunPoints(benchmarks, reports, metricsMap, options = {}) {
+    const displayName = options.agentDisplayName ?? ((key) => key);
+    // Pre-index reports by the BenchmarkRun id they belong to
+    // (EvaluationReport.experimentRunId === BenchmarkRun.id).
+    const reportsByRunDocId = new Map();
+    for (const report of reports) {
+        if (!report.experimentRunId)
+            continue;
+        const existing = reportsByRunDocId.get(report.experimentRunId);
+        if (existing)
+            existing.push(report);
+        else
+            reportsByRunDocId.set(report.experimentRunId, [report]);
+    }
+    const points = [];
+    for (const benchmark of benchmarks) {
+        if (options.benchmarkId && benchmark.id !== options.benchmarkId)
+            continue;
+        for (const run of benchmark.runs || []) {
+            const timestamp = new Date(run.createdAt).getTime();
+            if (Number.isNaN(timestamp))
+                continue;
+            if (options.sinceMs != null && timestamp < options.sinceMs)
+                continue;
+            const { passed, failed, total, accuracyPct } = getRunAccuracy(run);
+            if (total === 0)
+                continue; // nothing to plot for an empty run
+            const runReports = reportsByRunDocId.get(run.id) || [];
+            let matched = 0;
+            let costSum = 0;
+            let tokenSum = 0;
+            for (const report of runReports) {
+                const lookup = report.runId ? metricsMap.get(report.runId) : undefined;
+                if (!lookup)
+                    continue;
+                costSum += lookup.costUsd;
+                tokenSum += lookup.tokens;
+                matched++;
+            }
+            // Only expose a run-level cost/token total when EVERY report for this
+            // run resolved trace metrics. A partial sum (some test cases matched,
+            // some didn't) would silently understate the run's true cost/tokens
+            // while looking like a complete total — worse than admitting "unknown".
+            const isComplete = runReports.length > 0 && matched === runReports.length;
+            points.push({
+                runDocId: run.id,
+                benchmarkId: benchmark.id,
+                benchmarkName: benchmark.name,
+                agentKey: run.agentKey || 'unknown',
+                agentName: displayName(run.agentKey || 'unknown'),
+                modelId: run.modelId || '',
+                createdAt: run.createdAt,
+                timestamp,
+                passed,
+                failed,
+                total,
+                accuracyPct,
+                costUsd: isComplete ? costSum : null,
+                tokens: isComplete ? tokenSum : null,
+            });
+        }
+    }
+    points.sort((a, b) => a.timestamp - b.timestamp);
+    return points;
+}
+/** Group already time-sorted points by agentKey, preserving relative order. */
+export function groupPointsByAgent(points) {
+    const map = new Map();
+    for (const point of points) {
+        const existing = map.get(point.agentKey);
+        if (existing)
+            existing.push(point);
+        else
+            map.set(point.agentKey, [point]);
+    }
+    return map;
+}
+export function metricValue(point, metric) {
+    switch (metric) {
+        case 'accuracy':
+            return point.accuracyPct;
+        case 'cost':
+            return point.costUsd;
+        case 'tokens':
+            return point.tokens;
+        default:
+            return null;
+    }
+}
+export function formatMetricValue(metric, value) {
+    switch (metric) {
+        case 'accuracy':
+            return `${value.toFixed(1)}%`;
+        case 'cost':
+            return formatCost(value);
+        case 'tokens':
+            return formatTokens(value);
+        default:
+            return String(value);
+    }
+}
+/**
+ * Build one agent's gap-broken series for `metric`: points whose metric
+ * value is null (no evaluable accuracy, or no trace-resolved cost/tokens)
+ * are dropped — never plotted as a fake zero. A synthetic `{ value: null }`
+ * marker is inserted at the midpoint between any two consecutive PLOTTED
+ * points whose time gap exceeds `gapBreakMs` (default 7 days).
+ *
+ * Why: a chart line drawn between two data-array entries connects them
+ * with a straight (or smoothed) segment regardless of how far apart their
+ * x-values are — so two runs 25 days apart look, visually, like a
+ * continuous trend across those 25 days ("sparse runs produce misleading
+ * long interpolated lines" — owner feedback). Inserting a null-valued
+ * entry between them gives recharts' `<Line connectNulls={false}>` (the
+ * default) an explicit place to break the path, while the surrounding
+ * dead space still renders proportionally on a time-scaled x-axis.
+ */
+export function buildGapBrokenSeries(agentPoints, // already time-sorted, single agent
+metric, gapBreakMs = GAP_BREAK_MS_DEFAULT) {
+    const valued = agentPoints
+        .map(p => ({ p, v: metricValue(p, metric) }))
+        .filter((d) => d.v != null);
+    const out = [];
+    for (let i = 0; i < valued.length; i++) {
+        const { p, v } = valued[i];
+        if (i > 0) {
+            const prev = valued[i - 1].p;
+            if (p.timestamp - prev.timestamp > gapBreakMs) {
+                out.push({ timestamp: (prev.timestamp + p.timestamp) / 2, value: null, point: null });
+            }
+        }
+        out.push({ timestamp: p.timestamp, value: v, point: p });
+    }
+    return out;
+}
+export function computeLatestDelta(agentPoints, metric) {
+    const validValues = agentPoints
+        .map(p => metricValue(p, metric))
+        .filter((v) => v != null);
+    const n = validValues.length;
+    const latestValue = n >= 1 ? validValues[n - 1] : null;
+    const previousValue = n >= 2 ? validValues[n - 2] : null;
+    const delta = latestValue != null && previousValue != null ? latestValue - previousValue : null;
+    return { latestValue, previousValue, delta };
+}
+export function formatDelta(metric, delta) {
+    if (delta == null)
+        return 'n/a';
+    const sign = delta > 0 ? '+' : delta < 0 ? '-' : '±';
+    if (metric === 'accuracy') {
+        const rounded = Math.round(delta * 10) / 10;
+        if (Math.abs(rounded) < 0.05)
+            return '±0pp';
+        return `${rounded > 0 ? '+' : ''}${rounded}pp`;
+    }
+    return `${sign}${formatMetricValue(metric, Math.abs(delta))}`;
+}
+/**
+ * Group `points` by agent and shape each into a sparkline-table row for
+ * the given metric. One row per agent that has at least one point in
+ * scope — a single run is a perfectly valid row (a lone dot, `delta: null`
+ * rendered as "n/a"); there is no "not enough data" floor at the row
+ * level (unlike the old all-agents chart, which needed >=2 points across
+ * ALL agents just to avoid an empty canvas).
+ */
+export function buildAgentTrendRows(points, metric, gapBreakMs = GAP_BREAK_MS_DEFAULT) {
+    const grouped = groupPointsByAgent(points);
+    const rows = [];
+    for (const [agentKey, agentPoints] of grouped) {
+        if (agentPoints.length === 0)
+            continue;
+        const latest = agentPoints[agentPoints.length - 1];
+        const { latestValue, previousValue, delta } = computeLatestDelta(agentPoints, metric);
+        rows.push({
+            agentKey,
+            agentName: latest.agentName,
+            runCount: agentPoints.length,
+            latestRunAt: latest.createdAt,
+            latestRunDocId: latest.runDocId,
+            latestBenchmarkId: latest.benchmarkId,
+            points: agentPoints,
+            series: buildGapBrokenSeries(agentPoints, metric, gapBreakMs),
+            latestValue,
+            previousValue,
+            delta,
+        });
+    }
+    return rows;
+}
+/**
+ * Sort rows by latest value (highest first; rows with no value for the
+ * current metric sort last) or by biggest drop (most negative delta
+ * first; rows with no delta sort last, ordered among themselves by latest
+ * value so a still-informative "no signal yet" row isn't buried below a
+ * literal `undefined`-sorts-arbitrarily bucket).
+ */
+export function sortAgentTrendRows(rows, sortMode) {
+    const sorted = [...rows];
+    if (sortMode === 'biggestDrop') {
+        sorted.sort((a, b) => {
+            if (a.delta == null && b.delta == null)
+                return (b.latestValue ?? -Infinity) - (a.latestValue ?? -Infinity);
+            if (a.delta == null)
+                return 1;
+            if (b.delta == null)
+                return -1;
+            return a.delta - b.delta; // ascending: most negative (biggest drop) first
+        });
+    }
+    else {
+        sorted.sort((a, b) => {
+            if (a.latestValue == null && b.latestValue == null)
+                return 0;
+            if (a.latestValue == null)
+                return 1;
+            if (b.latestValue == null)
+                return -1;
+            return b.latestValue - a.latestValue; // descending: highest first
+        });
+    }
+    return sorted;
+}
+/** Default benchmark selector value: the benchmark with the most recent run. */
+export function getMostRecentlyActiveBenchmarkId(benchmarks) {
+    let bestId = null;
+    let bestTimestamp = -Infinity;
+    for (const benchmark of benchmarks) {
+        for (const run of benchmark.runs || []) {
+            const ts = new Date(run.createdAt).getTime();
+            if (!Number.isNaN(ts) && ts > bestTimestamp) {
+                bestTimestamp = ts;
+                bestId = benchmark.id;
+            }
+        }
+    }
+    return bestId;
+}
+/**
+ * Group `points` (already scoped to one benchmark by the caller) by agent
+ * and split each agent's valued runs into `latest` (most recent with a
+ * resolved value) + `history` (earlier ones). An agent with zero valued
+ * runs for this metric still gets a row (`latest: null`) so it's visible
+ * in the list rather than silently disappearing when switching metrics.
+ */
+export function buildBenchmarkDotPlotRows(points, metric) {
+    const grouped = groupPointsByAgent(points);
+    const rows = [];
+    for (const [agentKey, agentPoints] of grouped) {
+        const valued = agentPoints
+            .map(p => ({ point: p, value: metricValue(p, metric) }))
+            .filter((d) => d.value != null);
+        const agentName = agentPoints[0].agentName;
+        if (valued.length === 0) {
+            rows.push({ agentKey, agentName, latest: null, history: [], rank: -1 });
+            continue;
+        }
+        const latest = valued[valued.length - 1];
+        const history = valued.slice(0, -1);
+        rows.push({ agentKey, agentName, latest, history, rank: -1 });
+    }
+    return rows;
+}
+/**
+ * Rank rows by latest value, best (highest) first — rows with no value for
+ * the current metric sort last (still listed, just unranked among
+ * themselves by name for a stable order). Stamps `rank` (0 = top) on the
+ * returned copies; does not mutate the input.
+ */
+export function rankDotPlotRows(rows) {
+    const sorted = [...rows].sort((a, b) => {
+        if (a.latest == null && b.latest == null)
+            return a.agentName.localeCompare(b.agentName);
+        if (a.latest == null)
+            return 1;
+        if (b.latest == null)
+            return -1;
+        return b.latest.value - a.latest.value;
+    });
+    return sorted.map((row, index) => ({ ...row, rank: index }));
+}
+/**
+ * Shared X-domain for a dot plot: accuracy is always anchored to the true
+ * [0, 100] scale (auto-scaling to the observed min/max would visually
+ * exaggerate small real-world differences, e.g. an 88-95% cluster would
+ * stretch to fill the whole width). Cost/tokens are zero-anchored (0 is a
+ * meaningful reference point for both) up to the observed max plus a
+ * little headroom so the top dot isn't flush against the edge. Returns
+ * `null` when there is nothing to plot at all.
+ */
+export function metricDomain(rows, metric) {
+    const values = [];
+    for (const row of rows) {
+        if (row.latest)
+            values.push(row.latest.value);
+        for (const h of row.history)
+            values.push(h.value);
+    }
+    if (values.length === 0)
+        return null;
+    if (metric === 'accuracy')
+        return [0, 100];
+    const max = Math.max(...values);
+    return [0, max <= 0 ? 1 : max * 1.08];
+}
+/** Map a value into a 0-100 position within `domain`, clamped — usable directly as a CSS `left`/`width` percentage. */
+export function valueToPercent(value, domain) {
+    const [min, max] = domain;
+    if (max === min)
+        return 50;
+    const pct = ((value - min) / (max - min)) * 100;
+    return Math.max(0, Math.min(100, pct));
+}
+//# sourceMappingURL=agentTrends.js.map

package/lib/dist/lib/agentTrends.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agentTrends.js","sourceRoot":"","sources":["../../agentTrends.ts"],"names":[],"mappings":"AAAA;;;GAGG;AA4BH,OAAO,EAAE,gBAAgB,EAAE,MAAM,gBAAgB,CAAC;AAClD,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAsC9D,MAAM,oBAAoB,GAAG,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC,SAAS;AAE/D;;;;;;;;;;;;;GAaG;AAEH;;;;;GAKG;AACH,MAAM,UAAU,cAAc,CAAC,GAAiB;IAM9C,IAAI,GAAG,CAAC,KAAK,IAAI,GAAG,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC;QACrC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC;QAC1E,MAAM,WAAW,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,GAAG,SAAS,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC;QAChF,OAAO,EAAE,MAAM,EAAE,GAAG,CAAC,KAAK,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,CAAC,KAAK,CAAC,MAAM,EAAE,KAAK,EAAE,GAAG,CAAC,KAAK,CAAC,KAAK,EAAE,WAAW,EAAE,CAAC;IACrG,CAAC;IACD,MAAM,QAAQ,GAAG,gBAAgB,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IAC/C,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,CAAC,KAAK,GAAG,QAAQ,CAAC,OAAO,CAAC,CAAC;IACjE,MAAM,WAAW,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,GAAG,SAAS,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC;IAC/E,OAAO,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,WAAW,EAAE,CAAC;AAClG,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,KAAsB,EAAE,QAAgB,IAAI,CAAC,GAAG,EAAE;IACnF,MAAM,IAAI,GAAG,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,KAAK,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IAC5D,OAAO,KAAK,GAAG,IAAI,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;AAC5C,CAAC;AAQD;;;;GAIG;AACH,MAAM,UAAU,mBAAmB,CACjC,UAAuB,EACvB,OAA2B,EAC3B,UAAyC,EACzC,UAAsC,EAAE;IAExC,MAAM,WAAW,GAAG,OAAO,CAAC,gBAAgB,IAAI,CAAC,CAAC,GAAW,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC;IAEvE,0DAA0D;IAC1D,0DAA0D;IAC1D,MAAM,iBAAiB,GAAG,IAAI,GAAG,EAA8B,CAAC;IAChE,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,CAAC,MAAM,CAAC,eAAe;YAAE,SAAS;QACtC,MAAM,QAAQ,GAAG,iBAAiB,CAAC,GAAG,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC;QAC/D,IAAI,QAAQ;YAAE,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;;YAC/B,iBAAiB,CAAC,GAAG,CAAC,MAAM,CAAC,eAAe,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;IAC/D,CAAC;IAED,MAAM,MAAM,GAAoB,EAAE,CAAC;IAEnC,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACnC,IAAI,OAAO,CAAC,WAAW,IAAI,SAAS,CAAC,EAAE,KAAK,OAAO,CAAC,WAAW;YAAE,SAAS;QAE1E,KAAK,MAAM,GAAG,IAAI,SAAS,CAAC,IAAI,IAAI,EAAE,EAAE,CAAC;YACvC,MAAM,SAAS,GAAG,IAAI,IAAI,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,OAAO,EAAE,CAAC;YACpD,IAAI,MAAM,CAAC,KAAK,CAAC,SAAS,CAAC;gBAAE,SAAS;YACtC,IAAI,OAAO,CAAC,OAAO,IAAI,IAAI,IAAI,SAAS,GAAG,OAAO,CAAC,OAAO;gBAAE,SAAS;YAErE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,GAAG,cAAc,CAAC,GAAG,CAAC,CAAC;YACnE,IAAI,KAAK,KAAK,CAAC;gBAAE,SAAS,CAAC,mCAAmC;YAE9D,MAAM,UAAU,GAAG,iBAAiB,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC;YACvD,IAAI,OAAO,GAAG,CAAC,CAAC;YAChB,IAAI,OAAO,GAAG,CAAC,CAAC;YAChB,IAAI,QAAQ,GAAG,CAAC,CAAC;YACjB,KAAK,MAAM,MAAM,IAAI,UAAU,EAAE,CAAC;gBAChC,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;gBACvE,IAAI,CAAC,MAAM;oBAAE,SAAS;gBACtB,OAAO,IAAI,MAAM,CAAC,OAAO,CAAC;gBAC1B,QAAQ,IAAI,MAAM,CAAC,MAAM,CAAC;gBAC1B,OAAO,EAAE,CAAC;YACZ,CAAC;YACD,sEAAsE;YACtE,sEAAsE;YACtE,oEAAoE;YACpE,wEAAwE;YACxE,MAAM,UAAU,GAAG,UAAU,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,KAAK,UAAU,CAAC,MAAM,CAAC;YAE1E,MAAM,CAAC,IAAI,CAAC;gBACV,QAAQ,EAAE,GAAG,CAAC,EAAE;gBAChB,WAAW,EAAE,SAAS,CAAC,EAAE;gBACzB,aAAa,EAAE,SAAS,CAAC,IAAI;gBAC7B,QAAQ,EAAE,GAAG,CAAC,QAAQ,IAAI,SAAS;gBACnC,SAAS,EAAE,WAAW,CAAC,GAAG,CAAC,QAAQ,IAAI,SAAS,CAAC;gBACjD,OAAO,EAAE,GAAG,CAAC,OAAO,IAAI,EAAE;gBAC1B,SAAS,EAAE,GAAG,CAAC,SAAS;gBACxB,SAAS;gBACT,MAAM;gBACN,MAAM;gBACN,KAAK;gBACL,WAAW;gBACX,OAAO,EAAE,UAAU,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;gBACpC,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI;aACrC,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,SAAS,CAAC,CAAC;IACjD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,+EAA+E;AAC/E,MAAM,UAAU,kBAAkB,CAAC,MAAuB;IACxD,MAAM,GAAG,GAAG,IAAI,GAAG,EAA2B,CAAC;IAC/C,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC3B,MAAM,QAAQ,GAAG,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QACzC,IAAI,QAAQ;YAAE,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;;YAC9B,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC;IACxC,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED,MAAM,UAAU,WAAW,CAAC,KAAoB,EAAE,MAAsB;IACtE,QAAQ,MAAM,EAAE,CAAC;QACf,KAAK,UAAU;YACb,OAAO,KAAK,CAAC,WAAW,CAAC;QAC3B,KAAK,MAAM;YACT,OAAO,KAAK,CAAC,OAAO,CAAC;QACvB,KAAK,QAAQ;YACX,OAAO,KAAK,CAAC,MAAM,CAAC;QACtB;YACE,OAAO,IAAI,CAAC;IAChB,CAAC;AACH,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,MAAsB,EAAE,KAAa;IACrE,QAAQ,MAAM,EAAE,CAAC;QACf,KAAK,UAAU;YACb,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;QAChC,KAAK,MAAM;YACT,OAAO,UAAU,CAAC,KAAK,CAAC,CAAC;QAC3B,KAAK,QAAQ;YACX,OAAO,YAAY,CAAC,KAAK,CAAC,CAAC;QAC7B;YACE,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC;IACzB,CAAC;AACH,CAAC;AASD;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,oBAAoB,CAClC,WAA4B,EAAE,oCAAoC;AAClE,MAAsB,EACtB,aAAqB,oBAAoB;IAEzC,MAAM,MAAM,GAAG,WAAW;SACvB,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,WAAW,CAAC,CAAC,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC;SAC5C,MAAM,CAAC,CAAC,CAAC,EAAwC,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC;IAEpE,MAAM,GAAG,GAAuB,EAAE,CAAC;IACnC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACvC,MAAM,EAAE,CAAC,EAAE,CAAC,EAAE,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;QAC3B,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC;YACV,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;YAC7B,IAAI,CAAC,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,GAAG,UAAU,EAAE,CAAC;gBAC9C,GAAG,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,CAAC,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;YACxF,CAAC;QACH,CAAC;QACD,GAAG,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,SAAS,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC;IAC3D,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAUD,MAAM,UAAU,kBAAkB,CAAC,WAA4B,EAAE,MAAsB;IACrF,MAAM,WAAW,GAAG,WAAW;SAC5B,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,WAAW,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC;SAChC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC;IACzC,MAAM,CAAC,GAAG,WAAW,CAAC,MAAM,CAAC;IAC7B,MAAM,WAAW,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACvD,MAAM,aAAa,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACzD,MAAM,KAAK,GAAG,WAAW,IAAI,IAAI,IAAI,aAAa,IAAI,IAAI,CAAC,CAAC,CAAC,WAAW,GAAG,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE,KAAK,EAAE,CAAC;AAC/C,CAAC;AAED,MAAM,UAAU,WAAW,CAAC,MAAsB,EAAE,KAAoB;IACtE,IAAI,KAAK,IAAI,IAAI;QAAE,OAAO,KAAK,CAAC;IAChC,MAAM,IAAI,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;IACrD,IAAI,MAAM,KAAK,UAAU,EAAE,CAAC;QAC1B,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,EAAE,CAAC,GAAG,EAAE,CAAC;QAC5C,IAAI,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,IAAI;YAAE,OAAO,MAAM,CAAC;QAC5C,OAAO,GAAG,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,OAAO,IAAI,CAAC;IACjD,CAAC;IACD,OAAO,GAAG,IAAI,GAAG,iBAAiB,CAAC,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC;AAChE,CAAC;AAmBD;;;;;;;GAOG;AACH,MAAM,UAAU,mBAAmB,CACjC,MAAuB,EACvB,MAAsB,EACtB,aAAqB,oBAAoB;IAEzC,MAAM,OAAO,GAAG,kBAAkB,CAAC,MAAM,CAAC,CAAC;IAC3C,MAAM,IAAI,GAAoB,EAAE,CAAC;IACjC,KAAK,MAAM,CAAC,QAAQ,EAAE,WAAW,CAAC,IAAI,OAAO,EAAE,CAAC;QAC9C,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC;YAAE,SAAS;QACvC,MAAM,MAAM,GAAG,WAAW,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACnD,MAAM,EAAE,WAAW,EAAE,aAAa,EAAE,KAAK,EAAE,GAAG,kBAAkB,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC;QACtF,IAAI,CAAC,IAAI,CAAC;YACR,QAAQ;YACR,SAAS,EAAE,MAAM,CAAC,SAAS;YAC3B,QAAQ,EAAE,WAAW,CAAC,MAAM;YAC5B,WAAW,EAAE,MAAM,CAAC,SAAS;YAC7B,cAAc,EAAE,MAAM,CAAC,QAAQ;YAC/B,iBAAiB,EAAE,MAAM,CAAC,WAAW;YACrC,MAAM,EAAE,WAAW;YACnB,MAAM,EAAE,oBAAoB,CAAC,WAAW,EAAE,MAAM,EAAE,UAAU,CAAC;YAC7D,WAAW;YACX,aAAa;YACb,KAAK;SACN,CAAC,CAAC;IACL,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,kBAAkB,CAAC,IAAqB,EAAE,QAAuB;IAC/E,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,CAAC,CAAC;IACzB,IAAI,QAAQ,KAAK,aAAa,EAAE,CAAC;QAC/B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;YACnB,IAAI,CAAC,CAAC,KAAK,IAAI,IAAI,IAAI,CAAC,CAAC,KAAK,IAAI,IAAI;gBAAE,OAAO,CAAC,CAAC,CAAC,WAAW,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,WAAW,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC3G,IAAI,CAAC,CAAC,KAAK,IAAI,IAAI;gBAAE,OAAO,CAAC,CAAC;YAC9B,IAAI,CAAC,CAAC,KAAK,IAAI,IAAI;gBAAE,OAAO,CAAC,CAAC,CAAC;YAC/B,OAAO,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,gDAAgD;QAC5E,CAAC,CAAC,CAAC;IACL,CAAC;SAAM,CAAC;QACN,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;YACnB,IAAI,CAAC,CAAC,WAAW,IAAI,IAAI,IAAI,CAAC,CAAC,WAAW,IAAI,IAAI;gBAAE,OAAO,CAAC,CAAC;YAC7D,IAAI,CAAC,CAAC,WAAW,IAAI,IAAI;gBAAE,OAAO,CAAC,CAAC;YACpC,IAAI,CAAC,CAAC,WAAW,IAAI,IAAI;gBAAE,OAAO,CAAC,CAAC,CAAC;YACrC,OAAO,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,WAAW,CAAC,CAAC,4BAA4B;QACpE,CAAC,CAAC,CAAC;IACL,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,gFAAgF;AAChF,MAAM,UAAU,gCAAgC,CAAC,UAAuB;IACtE,IAAI,MAAM,GAAkB,IAAI,CAAC;IACjC,IAAI,aAAa,GAAG,CAAC,QAAQ,CAAC;IAC9B,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACnC,KAAK,MAAM,GAAG,IAAI,SAAS,CAAC,IAAI,IAAI,EAAE,EAAE,CAAC;YACvC,MAAM,EAAE,GAAG,IAAI,IAAI,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,OAAO,EAAE,CAAC;YAC7C,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,EAAE,GAAG,aAAa,EAAE,CAAC;gBAC5C,aAAa,GAAG,EAAE,CAAC;gBACnB,MAAM,GAAG,SAAS,CAAC,EAAE,CAAC;YACxB,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AA6BD;;;;;;GAMG;AACH,MAAM,UAAU,yBAAyB,CACvC,MAAuB,EACvB,MAAsB;IAEtB,MAAM,OAAO,GAAG,kBAAkB,CAAC,MAAM,CAAC,CAAC;IAC3C,MAAM,IAAI,GAA0B,EAAE,CAAC;IACvC,KAAK,MAAM,CAAC,QAAQ,EAAE,WAAW,CAAC,IAAI,OAAO,EAAE,CAAC;QAC9C,MAAM,MAAM,GAAG,WAAW;aACvB,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,WAAW,CAAC,CAAC,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC;aACvD,MAAM,CAAC,CAAC,CAAC,EAAqB,EAAE,CAAC,CAAC,CAAC,KAAK,IAAI,IAAI,CAAC,CAAC;QAErD,MAAM,SAAS,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QAC3C,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACxB,IAAI,CAAC,IAAI,CAAC,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC;YACxE,SAAS;QACX,CAAC;QACD,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACzC,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QACpC,IAAI,CAAC,IAAI,CAAC,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,eAAe,CAAC,IAA2B;IACzD,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACrC,IAAI,CAAC,CAAC,MAAM,IAAI,IAAI,IAAI,CAAC,CAAC,MAAM,IAAI,IAAI;YAAE,OAAO,CAAC,CAAC,SAAS,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;QACxF,IAAI,CAAC,CAAC,MAAM,IAAI,IAAI;YAAE,OAAO,CAAC,CAAC;QAC/B,IAAI,CAAC,CAAC,MAAM,IAAI,IAAI;YAAE,OAAO,CAAC,CAAC,CAAC;QAChC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;IACzC,CAAC,CAAC,CAAC;IACH,OAAO,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,GAAG,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;AAC/D,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,YAAY,CAAC,IAA2B,EAAE,MAAsB;IAC9E,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,IAAI,GAAG,CAAC,MAAM;YAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAC9C,KAAK,MAAM,CAAC,IAAI,GAAG,CAAC,OAAO;YAAE,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC;IACpD,CAAC;IACD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACrC,IAAI,MAAM,KAAK,UAAU;QAAE,OAAO,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IAC3C,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC;IAChC,OAAO,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,IAAI,CAAC,CAAC;AACxC,CAAC;AAED,uHAAuH;AACvH,MAAM,UAAU,cAAc,CAAC,KAAa,EAAE,MAAwB;IACpE,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,GAAG,MAAM,CAAC;IAC1B,IAAI,GAAG,KAAK,GAAG;QAAE,OAAO,EAAE,CAAC;IAC3B,MAAM,GAAG,GAAG,CAAC,CAAC,KAAK,GAAG,GAAG,CAAC,GAAG,CAAC,GAAG,GAAG,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC;IAChD,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AACzC,CAAC"}

package/lib/dist/lib/bedrockCompat.d.ts

@@ -0,0 +1,27 @@
+/** Whether the given Bedrock model still accepts an explicit `temperature`. */
+export declare function modelSupportsTemperature(modelId: string): boolean;
+/**
+ * Build a Converse `inferenceConfig`, including `temperature` only when the
+ * target model accepts it. Callers pass their preferred temperature; for
+ * models that have deprecated the parameter it is silently omitted (the
+ * model's own default applies).
+ */
+export declare function buildInferenceConfig(modelId: string, opts: {
+    maxTokens: number;
+    temperature?: number;
+}): {
+    maxTokens: number;
+    temperature?: number;
+};
+/** Inference-profile prefix appropriate for the given (or current) AWS region. */
+export declare function regionInferencePrefix(region?: string): string;
+/**
+ * Re-home a cross-region inference-profile id to the active AWS region.
+ *
+ * `us.anthropic.claude-...` → `eu.anthropic.claude-...` when running in an
+ * EU region, etc. Ids that carry no regional prefix (bare model ids,
+ * `global.` profiles, non-Bedrock ids like `mock://` or `gpt-4o`) are
+ * returned unchanged.
+ */
+export declare function resolveRegionAwareModelId(modelId: string, region?: string): string;
+//# sourceMappingURL=bedrockCompat.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"bedrockCompat.d.ts","sourceRoot":"","sources":["../../bedrockCompat.ts"],"names":[],"mappings":"AAmCA,+EAA+E;AAC/E,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAGjE;AAED;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAClC,OAAO,EAAE,MAAM,EACf,IAAI,EAAE;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,GAChD;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,CAM7C;AAED,kFAAkF;AAClF,wBAAgB,qBAAqB,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,MAAM,CAU7D;AAID;;;;;;;GAOG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,MAAM,GAAG,MAAM,CASlF"}