@ls-stack/agent-eval 0.61.1 → 0.61.2

package/dist/index.d.mts

@@ -2061,9 +2061,9 @@ declare const traceAttributeDisplaySchema: z.ZodObject<{
     subtree: "subtree";
   }>>;
   mode: z.ZodOptional<z.ZodEnum<{
-    sum: "sum";
     all: "all";
     last: "last";
+    sum: "sum";
   }>>;
 }, z.core.$strip>;
 /**
@@ -2097,9 +2097,9 @@ declare const traceDisplayConfigSchema: z.ZodObject<{
       subtree: "subtree";
     }>>;
     mode: z.ZodOptional<z.ZodEnum<{
-      sum: "sum";
       all: "all";
       last: "last";
+      sum: "sum";
     }>>;
   }, z.core.$strip>>>;
 }, z.core.$strip>;
@@ -2137,9 +2137,9 @@ declare const traceAttributeDisplayInputSchema: z.ZodObject<{
     subtree: "subtree";
   }>>;
   mode: z.ZodOptional<z.ZodEnum<{
-    sum: "sum";
     all: "all";
     last: "last";
+    sum: "sum";
   }>>;
   transform: z.ZodOptional<z.ZodCustom<TraceAttributeTransform, TraceAttributeTransform>>;
 }, z.core.$strip>;
@@ -2175,9 +2175,9 @@ declare const traceDisplayInputConfigSchema: z.ZodObject<{
       subtree: "subtree";
     }>>;
     mode: z.ZodOptional<z.ZodEnum<{
-      sum: "sum";
       all: "all";
       last: "last";
+      sum: "sum";
     }>>;
     transform: z.ZodOptional<z.ZodCustom<TraceAttributeTransform, TraceAttributeTransform>>;
   }, z.core.$strip>>>;
@@ -2214,8 +2214,8 @@ declare const traceSpanSchema$1: z.ZodObject<{
   status: z.ZodEnum<{
     error: "error";
     running: "running";
-    cancelled: "cancelled";
     ok: "ok";
+    cancelled: "cancelled";
   }>;
   attributes: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
   error: z.ZodOptional<z.ZodObject<{
@@ -2260,10 +2260,10 @@ type EvalFreshnessStatus = z.infer<typeof evalFreshnessStatusSchema>;
  * `best` selects the highest finite value and `worst` selects the lowest.
  */
 declare const evalStatAggregateSchema: z.ZodEnum<{
+  sum: "sum";
   avg: "avg";
   min: "min";
   max: "max";
-  sum: "sum";
   best: "best";
   worst: "worst";
 }>;
@@ -2292,10 +2292,10 @@ declare const evalStatItemSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
   hideIfNoValue: z.ZodOptional<z.ZodBoolean>;
   kind: z.ZodLiteral<"duration">;
   aggregate: z.ZodOptional<z.ZodEnum<{
+    sum: "sum";
     avg: "avg";
     min: "min";
     max: "max";
-    sum: "sum";
     best: "best";
     worst: "worst";
   }>>;
@@ -2303,10 +2303,10 @@ declare const evalStatItemSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
   hideIfNoValue: z.ZodOptional<z.ZodBoolean>;
   kind: z.ZodLiteral<"cacheHits">;
   aggregate: z.ZodOptional<z.ZodEnum<{
+    sum: "sum";
     avg: "avg";
     min: "min";
     max: "max";
-    sum: "sum";
     best: "best";
     worst: "worst";
   }>>;
@@ -2316,10 +2316,10 @@ declare const evalStatItemSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
   key: z.ZodString;
   label: z.ZodOptional<z.ZodString>;
   aggregate: z.ZodEnum<{
+    sum: "sum";
     avg: "avg";
     min: "min";
     max: "max";
-    sum: "sum";
     best: "best";
     worst: "worst";
   }>;
@@ -2356,10 +2356,10 @@ declare const evalStatsConfigSchema: z.ZodArray<z.ZodDiscriminatedUnion<[z.ZodOb
   hideIfNoValue: z.ZodOptional<z.ZodBoolean>;
   kind: z.ZodLiteral<"duration">;
   aggregate: z.ZodOptional<z.ZodEnum<{
+    sum: "sum";
     avg: "avg";
     min: "min";
     max: "max";
-    sum: "sum";
     best: "best";
     worst: "worst";
   }>>;
@@ -2367,10 +2367,10 @@ declare const evalStatsConfigSchema: z.ZodArray<z.ZodDiscriminatedUnion<[z.ZodOb
   hideIfNoValue: z.ZodOptional<z.ZodBoolean>;
   kind: z.ZodLiteral<"cacheHits">;
   aggregate: z.ZodOptional<z.ZodEnum<{
+    sum: "sum";
     avg: "avg";
     min: "min";
     max: "max";
-    sum: "sum";
     best: "best";
     worst: "worst";
   }>>;
@@ -2380,10 +2380,10 @@ declare const evalStatsConfigSchema: z.ZodArray<z.ZodDiscriminatedUnion<[z.ZodOb
   key: z.ZodString;
   label: z.ZodOptional<z.ZodString>;
   aggregate: z.ZodEnum<{
+    sum: "sum";
     avg: "avg";
     min: "min";
     max: "max";
-    sum: "sum";
     best: "best";
     worst: "worst";
   }>;
@@ -2466,10 +2466,10 @@ declare const evalSummarySchema$1: z.ZodObject<{
   caseIds: z.ZodOptional<z.ZodArray<z.ZodString>>;
   lastRunStatus: z.ZodNullable<z.ZodEnum<{
     error: "error";
-    pass: "pass";
-    fail: "fail";
     running: "running";
     cancelled: "cancelled";
+    pass: "pass";
+    fail: "fail";
     unscored: "unscored";
   }>>;
   stats: z.ZodOptional<z.ZodArray<z.ZodDiscriminatedUnion<[z.ZodObject<{
@@ -2483,10 +2483,10 @@ declare const evalSummarySchema$1: z.ZodObject<{
     hideIfNoValue: z.ZodOptional<z.ZodBoolean>;
     kind: z.ZodLiteral<"duration">;
     aggregate: z.ZodOptional<z.ZodEnum<{
+      sum: "sum";
       avg: "avg";
       min: "min";
       max: "max";
-      sum: "sum";
       best: "best";
       worst: "worst";
     }>>;
@@ -2494,10 +2494,10 @@ declare const evalSummarySchema$1: z.ZodObject<{
     hideIfNoValue: z.ZodOptional<z.ZodBoolean>;
     kind: z.ZodLiteral<"cacheHits">;
     aggregate: z.ZodOptional<z.ZodEnum<{
+      sum: "sum";
       avg: "avg";
       min: "min";
       max: "max";
-      sum: "sum";
       best: "best";
       worst: "worst";
     }>>;
@@ -2507,10 +2507,10 @@ declare const evalSummarySchema$1: z.ZodObject<{
     key: z.ZodString;
     label: z.ZodOptional<z.ZodString>;
     aggregate: z.ZodEnum<{
+      sum: "sum";
       avg: "avg";
       min: "min";
       max: "max";
-      sum: "sum";
       best: "best";
       worst: "worst";
     }>;
@@ -2534,10 +2534,10 @@ declare const evalSummarySchema$1: z.ZodObject<{
     accent: z.ZodOptional<z.ZodBoolean>;
   }, z.core.$strip>], "kind">>>;
   defaultStatAggregate: z.ZodOptional<z.ZodEnum<{
+    sum: "sum";
     avg: "avg";
     min: "min";
     max: "max";
-    sum: "sum";
     best: "best";
     worst: "worst";
   }>>;
@@ -2560,9 +2560,9 @@ declare const evalSummarySchema$1: z.ZodObject<{
       color: z.ZodOptional<z.ZodEnum<{
         success: "success";
         error: "error";
+        warning: "warning";
         accent: "accent";
         accentDim: "accentDim";
-        warning: "warning";
         textMuted: "textMuted";
       }>>;
       axis: z.ZodOptional<z.ZodEnum<{
@@ -2573,10 +2573,10 @@ declare const evalSummarySchema$1: z.ZodObject<{
       source: z.ZodLiteral<"column">;
       key: z.ZodString;
       aggregate: z.ZodEnum<{
+        sum: "sum";
         avg: "avg";
         min: "min";
         max: "max";
-        sum: "sum";
         latest: "latest";
         passThresholdRate: "passThresholdRate";
       }>;
@@ -2584,9 +2584,9 @@ declare const evalSummarySchema$1: z.ZodObject<{
       color: z.ZodOptional<z.ZodEnum<{
         success: "success";
         error: "error";
+        warning: "warning";
         accent: "accent";
         accentDim: "accentDim";
-        warning: "warning";
         textMuted: "textMuted";
       }>>;
       axis: z.ZodOptional<z.ZodEnum<{
@@ -2615,10 +2615,10 @@ declare const evalSummarySchema$1: z.ZodObject<{
       source: z.ZodLiteral<"column">;
       key: z.ZodString;
       aggregate: z.ZodEnum<{
+        sum: "sum";
         avg: "avg";
         min: "min";
         max: "max";
-        sum: "sum";
         latest: "latest";
         passThresholdRate: "passThresholdRate";
       }>;
@@ -2715,10 +2715,10 @@ declare const caseRowSchema$1: z.ZodObject<{
   tags: z.ZodOptional<z.ZodArray<z.ZodString>>;
   status: z.ZodEnum<{
     error: "error";
-    pass: "pass";
-    fail: "fail";
     running: "running";
     cancelled: "cancelled";
+    pass: "pass";
+    fail: "fail";
     pending: "pending";
   }>;
   durationMs: z.ZodNullable<z.ZodNumber>;
@@ -2857,8 +2857,8 @@ declare const scoreTraceSchema: z.ZodObject<{
     status: z.ZodEnum<{
       error: "error";
       running: "running";
-      cancelled: "cancelled";
       ok: "ok";
+      cancelled: "cancelled";
     }>;
     attributes: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
     error: z.ZodOptional<z.ZodObject<{
@@ -2908,9 +2908,9 @@ declare const scoreTraceSchema: z.ZodObject<{
         subtree: "subtree";
       }>>;
       mode: z.ZodOptional<z.ZodEnum<{
-        sum: "sum";
         all: "all";
         last: "last";
+        sum: "sum";
       }>>;
     }, z.core.$strip>>>;
   }, z.core.$strip>;
@@ -2942,10 +2942,10 @@ declare const caseDetailSchema$1: z.ZodObject<{
   tags: z.ZodOptional<z.ZodArray<z.ZodString>>;
   status: z.ZodEnum<{
     error: "error";
-    pass: "pass";
-    fail: "fail";
     running: "running";
     cancelled: "cancelled";
+    pass: "pass";
+    fail: "fail";
     pending: "pending";
   }>;
   input: z.ZodUnknown;
@@ -2960,8 +2960,8 @@ declare const caseDetailSchema$1: z.ZodObject<{
     status: z.ZodEnum<{
       error: "error";
       running: "running";
-      cancelled: "cancelled";
       ok: "ok";
+      cancelled: "cancelled";
     }>;
     attributes: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
     error: z.ZodOptional<z.ZodObject<{
@@ -3011,9 +3011,9 @@ declare const caseDetailSchema$1: z.ZodObject<{
         subtree: "subtree";
       }>>;
       mode: z.ZodOptional<z.ZodEnum<{
-        sum: "sum";
         all: "all";
         last: "last";
+        sum: "sum";
       }>>;
     }, z.core.$strip>>>;
   }, z.core.$strip>;
@@ -3029,8 +3029,8 @@ declare const caseDetailSchema$1: z.ZodObject<{
       status: z.ZodEnum<{
         error: "error";
         running: "running";
-        cancelled: "cancelled";
         ok: "ok";
+        cancelled: "cancelled";
       }>;
       attributes: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
       error: z.ZodOptional<z.ZodObject<{
@@ -3080,9 +3080,9 @@ declare const caseDetailSchema$1: z.ZodObject<{
           subtree: "subtree";
         }>>;
         mode: z.ZodOptional<z.ZodEnum<{
-          sum: "sum";
           all: "all";
           last: "last";
+          sum: "sum";
         }>>;
       }, z.core.$strip>>>;
     }, z.core.$strip>;
@@ -3269,10 +3269,10 @@ declare const evalChartBuiltinMetricSchema: z.ZodEnum<{
 type EvalChartBuiltinMetric = z.infer<typeof evalChartBuiltinMetricSchema>;
 /** Reducer applied to a numeric column across all cases of a single run. */
 declare const evalChartAggregateSchema: z.ZodEnum<{
+  sum: "sum";
   avg: "avg";
   min: "min";
   max: "max";
-  sum: "sum";
   latest: "latest";
   passThresholdRate: "passThresholdRate";
 }>;
@@ -3285,9 +3285,9 @@ type EvalChartAggregate = z.infer<typeof evalChartAggregateSchema>;
 declare const evalChartColorSchema: z.ZodEnum<{
   success: "success";
   error: "error";
+  warning: "warning";
   accent: "accent";
   accentDim: "accentDim";
-  warning: "warning";
   textMuted: "textMuted";
 }>;
 /** Semantic color token resolved to a theme color by the web UI. */
@@ -3314,9 +3314,9 @@ declare const evalChartMetricSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
   color: z.ZodOptional<z.ZodEnum<{
     success: "success";
     error: "error";
+    warning: "warning";
     accent: "accent";
     accentDim: "accentDim";
-    warning: "warning";
     textMuted: "textMuted";
   }>>;
   axis: z.ZodOptional<z.ZodEnum<{
@@ -3327,10 +3327,10 @@ declare const evalChartMetricSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
   source: z.ZodLiteral<"column">;
   key: z.ZodString;
   aggregate: z.ZodEnum<{
+    sum: "sum";
     avg: "avg";
     min: "min";
     max: "max";
-    sum: "sum";
     latest: "latest";
     passThresholdRate: "passThresholdRate";
   }>;
@@ -3338,9 +3338,9 @@ declare const evalChartMetricSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
   color: z.ZodOptional<z.ZodEnum<{
     success: "success";
     error: "error";
+    warning: "warning";
     accent: "accent";
     accentDim: "accentDim";
-    warning: "warning";
     textMuted: "textMuted";
   }>>;
   axis: z.ZodOptional<z.ZodEnum<{
@@ -3362,10 +3362,10 @@ declare const evalChartTooltipExtraSchema: z.ZodDiscriminatedUnion<[z.ZodObject<
   source: z.ZodLiteral<"column">;
   key: z.ZodString;
   aggregate: z.ZodEnum<{
+    sum: "sum";
     avg: "avg";
     min: "min";
     max: "max";
-    sum: "sum";
     latest: "latest";
     passThresholdRate: "passThresholdRate";
   }>;
@@ -3397,9 +3397,9 @@ declare const evalChartConfigSchema: z.ZodObject<{
     color: z.ZodOptional<z.ZodEnum<{
       success: "success";
       error: "error";
+      warning: "warning";
       accent: "accent";
       accentDim: "accentDim";
-      warning: "warning";
       textMuted: "textMuted";
     }>>;
     axis: z.ZodOptional<z.ZodEnum<{
@@ -3410,10 +3410,10 @@ declare const evalChartConfigSchema: z.ZodObject<{
     source: z.ZodLiteral<"column">;
     key: z.ZodString;
     aggregate: z.ZodEnum<{
+      sum: "sum";
       avg: "avg";
       min: "min";
       max: "max";
-      sum: "sum";
       latest: "latest";
       passThresholdRate: "passThresholdRate";
     }>;
@@ -3421,9 +3421,9 @@ declare const evalChartConfigSchema: z.ZodObject<{
     color: z.ZodOptional<z.ZodEnum<{
       success: "success";
       error: "error";
+      warning: "warning";
       accent: "accent";
       accentDim: "accentDim";
-      warning: "warning";
       textMuted: "textMuted";
     }>>;
     axis: z.ZodOptional<z.ZodEnum<{
@@ -3452,10 +3452,10 @@ declare const evalChartConfigSchema: z.ZodObject<{
     source: z.ZodLiteral<"column">;
     key: z.ZodString;
     aggregate: z.ZodEnum<{
+      sum: "sum";
       avg: "avg";
       min: "min";
       max: "max";
-      sum: "sum";
       latest: "latest";
       passThresholdRate: "passThresholdRate";
     }>;
@@ -3487,9 +3487,9 @@ declare const evalChartsConfigSchema: z.ZodArray<z.ZodObject<{
     color: z.ZodOptional<z.ZodEnum<{
       success: "success";
       error: "error";
+      warning: "warning";
       accent: "accent";
       accentDim: "accentDim";
-      warning: "warning";
       textMuted: "textMuted";
     }>>;
     axis: z.ZodOptional<z.ZodEnum<{
@@ -3500,10 +3500,10 @@ declare const evalChartsConfigSchema: z.ZodArray<z.ZodObject<{
     source: z.ZodLiteral<"column">;
     key: z.ZodString;
     aggregate: z.ZodEnum<{
+      sum: "sum";
       avg: "avg";
       min: "min";
       max: "max";
-      sum: "sum";
       latest: "latest";
       passThresholdRate: "passThresholdRate";
     }>;
@@ -3511,9 +3511,9 @@ declare const evalChartsConfigSchema: z.ZodArray<z.ZodObject<{
     color: z.ZodOptional<z.ZodEnum<{
       success: "success";
       error: "error";
+      warning: "warning";
       accent: "accent";
       accentDim: "accentDim";
-      warning: "warning";
       textMuted: "textMuted";
     }>>;
     axis: z.ZodOptional<z.ZodEnum<{
@@ -3542,10 +3542,10 @@ declare const evalChartsConfigSchema: z.ZodArray<z.ZodObject<{
     source: z.ZodLiteral<"column">;
     key: z.ZodString;
     aggregate: z.ZodEnum<{
+      sum: "sum";
       avg: "avg";
       min: "min";
       max: "max";
-      sum: "sum";
       latest: "latest";
       passThresholdRate: "passThresholdRate";
     }>;
@@ -3573,8 +3573,8 @@ declare const runManifestSchema$1: z.ZodObject<{
   evalSourceFingerprints: z.ZodDefault<z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>>;
   target: z.ZodObject<{
     mode: z.ZodEnum<{
-      caseIds: "caseIds";
       all: "all";
+      caseIds: "caseIds";
       evalIds: "evalIds";
     }>;
     evalKeys: z.ZodOptional<z.ZodArray<z.ZodString>>;
@@ -4729,8 +4729,8 @@ declare const cacheRecordingSchema: z.ZodObject<{
   finalStatus: z.ZodOptional<z.ZodEnum<{
     error: "error";
     running: "running";
-    cancelled: "cancelled";
     ok: "ok";
+    cancelled: "cancelled";
   }>>;
   finalError: z.ZodOptional<z.ZodObject<{
     name: z.ZodOptional<z.ZodString>;
@@ -4830,8 +4830,8 @@ declare const cacheEntrySchema: z.ZodObject<{
     finalStatus: z.ZodOptional<z.ZodEnum<{
       error: "error";
       running: "running";
-      cancelled: "cancelled";
       ok: "ok";
+      cancelled: "cancelled";
     }>>;
     finalError: z.ZodOptional<z.ZodObject<{
       name: z.ZodOptional<z.ZodString>;
@@ -4948,8 +4948,8 @@ declare const cacheDebugKeyEntrySchema: z.ZodObject<{
       finalStatus: z.ZodOptional<z.ZodEnum<{
         error: "error";
         running: "running";
-        cancelled: "cancelled";
         ok: "ok";
+        cancelled: "cancelled";
       }>>;
       finalError: z.ZodOptional<z.ZodObject<{
         name: z.ZodOptional<z.ZodString>;
@@ -5055,8 +5055,8 @@ declare const cacheEntryWithDebugKeySchema$1: z.ZodObject<{
     finalStatus: z.ZodOptional<z.ZodEnum<{
       error: "error";
       running: "running";
-      cancelled: "cancelled";
       ok: "ok";
+      cancelled: "cancelled";
     }>>;
     finalError: z.ZodOptional<z.ZodObject<{
       name: z.ZodOptional<z.ZodString>;
@@ -5164,8 +5164,8 @@ declare const cacheEntryWithDebugKeySchema$1: z.ZodObject<{
         finalStatus: z.ZodOptional<z.ZodEnum<{
           error: "error";
           running: "running";
-          cancelled: "cancelled";
           ok: "ok";
+          cancelled: "cancelled";
         }>>;
         finalError: z.ZodOptional<z.ZodObject<{
           name: z.ZodOptional<z.ZodString>;
@@ -5271,8 +5271,8 @@ declare const cacheFileSchema: z.ZodObject<{
       finalStatus: z.ZodOptional<z.ZodEnum<{
         error: "error";
         running: "running";
-        cancelled: "cancelled";
         ok: "ok";
+        cancelled: "cancelled";
       }>>;
       finalError: z.ZodOptional<z.ZodObject<{
         name: z.ZodOptional<z.ZodString>;
@@ -5388,8 +5388,8 @@ declare const cacheDebugKeyFileSchema: z.ZodObject<{
         finalStatus: z.ZodOptional<z.ZodEnum<{
           error: "error";
           running: "running";
-          cancelled: "cancelled";
           ok: "ok";
+          cancelled: "cancelled";
         }>>;
         finalError: z.ZodOptional<z.ZodObject<{
           name: z.ZodOptional<z.ZodString>;
@@ -5573,8 +5573,8 @@ type ConfigReloadState = z.infer<typeof configReloadStateSchema$1>;
 declare const createRunRequestSchema$1: z.ZodObject<{
   target: z.ZodObject<{
     mode: z.ZodEnum<{
-      caseIds: "caseIds";
       all: "all";
+      caseIds: "caseIds";
       evalIds: "evalIds";
     }>;
     evalKeys: z.ZodOptional<z.ZodArray<z.ZodString>>;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ls-stack/agent-eval",
-  "version": "0.61.1",
+  "version": "0.61.2",
   "type": "module",
   "bin": {
     "agent-evals": "./dist/bin.mjs"

package/skills/agent-eval/SKILL.md

@@ -212,11 +212,11 @@ See `EvalScoreDef` / `EvalManualScoreDef` in the types for the full shape (forma
 - `tracingAssertions` is a single function that can be authored globally or locally on one eval when a finished-trace invariant should pass or fail the case without creating a fake score column. It receives the same `{ trace, input, case }` context as `deriveFromTracing`; call `evalAssert(...)` or `evalExpect(...)` inside it. Useful trace helpers include `trace.findSpan(name)`, `trace.findSpans(name)`, `trace.hasSpan(name)`, `trace.findSpansByKind(kind)`, `trace.findToolCallSpans()`, `trace.listToolCallSpanNames()`, `trace.hasToolCallSpan(name)`, `trace.getToolCallSpans(name)`, `trace.getToolCallSpanCount(toolName)`, `trace.hasToolCallSpanCount(toolName, expectedCalls)`, `trace.listSpanNames(kind?)`, `trace.listSpanNamesDfs(kind?)`, and `trace.flattenDfs()`. The tool-call helpers include both `kind: 'tool'` spans and imported execution spans recorded as `kind: 'tool_call'`. Tool-name checks and counts match the span `name` as well as GenAI/Mastra identity attributes such as `genAI["gen_ai.tool.name"]` and `mastra.entityName`; list helpers prefer those tool identity attributes when present. `getToolCallSpans(name)` returns one normalized object per matching call, including parsed `arguments`, parsed `result`, `description`, `toolType`, `attributes`, and the original `span`.
 - `traceDisplay` promotes selected span attributes into the trace tree and detail pane; it supports aggregation across subtrees (`scope`, `mode`) and user-defined `transform(...)` for derived views (e.g. currency conversion). See the `TraceDisplayInputConfig` type.
 - `llmCalls` (in `agent-evals.config.ts`) configures how LLM-call spans are summarized for review. Defaults to `kind: 'llm'` spans with `model`, `usage.*`, `latencyMs`, `input`, `output`, etc. read from conventional attribute paths. The default `steps` path reads an array from `span.attributes.steps`; if it is missing, direct child `model_step` spans are shown as that call's steps. Tool calls are aggregated from the configured `toolCalls` path plus step-level `toolCalls` on authored step arrays or direct `model_step` child spans, including Mastra's serialized `mastra.model_step.output` format, and child `tool_call` execution spans under each model step. `latencyMs` is time to first token; duration, total tokens, output tokens/sec, and USD costs are derived. Override `kinds` to broaden the filter, override `attributes.<field>` for non-default primitive span shapes, configure model-keyed `pricing` to derive USD costs from token counts, with nested `providers` entries for provider-specific rates, add `costCurrencies` to show converted cost columns in the expanded breakdown table only, add `derivedAttributes` to persist computed values back onto matching LLM spans before trace consumers run, and add entries to `metrics` to surface arbitrary user metrics (`format: 'string' | 'number' | 'duration' | 'json' | 'boolean'`, `placements: ['header' | 'body']`). `derivedAttributes` can be a keyed map for one-off fields or one callback that returns multiple path/value pairs. Derived keys are dot-paths under `span.attributes`; return `undefined` to skip one span or one returned key.
-- Default usage config derives missing eval outputs from matching LLM/API spans before `outputsSchema` and scores run: `apiCalls`, `costUsd`, `llmTurns`, `inputTokens`, `outputTokens`, `totalTokens`, `cachedInputTokens`, `cacheCreationInputTokens`, `reasoningTokens`, and `llmDurationMs`. Authored outputs and column overrides win. Default usage columns, stats, and charts use `hideIfNoValue: true`. Default LLM usage charts configure cost, input tokens, and output tokens separately and use `dedupeConsecutiveValues: true` to skip repeated adjacent chart values. `totalTokens` is input + output only; cache read/write tokens stay separate and affect `costUsd` at their own rates. `llmTurns` is the maximum per-call turn count in the case run, using configured steps when available and otherwise one turn per matched LLM call span. Derived base input cost uses `inputTokens - cachedInputTokens - cacheCreationInputTokens` so cache details are not double-counted. `cacheCreationInputTokens` is the total cache-write count; optional `cacheCreationInput1hTokens` only splits that total for 1-hour write pricing via `cacheCreationInput1hUsdPerMillion`. `llmDurationMs` sums elapsed matched LLM span durations; it is not time-to-first-token latency. Remove defaults globally or per eval with `removeDefaultConfig: true` or a key list such as `removeDefaultConfig: ['apiCalls', 'reasoningTokens']`.
+- Default usage config derives missing eval outputs from matching LLM/API spans before `outputsSchema` and scores run: `apiCalls`, `costUsd`, `llmTurns`, `inputTokens`, `outputTokens`, `totalTokens`, `cachedInputTokens`, `cacheCreationInputTokens`, `reasoningTokens`, and `llmDurationMs`. Authored outputs and column overrides win. The web UI fills in baseline run-health stats (`cases`, `passRate`, `duration`) and a pass-rate/duration history chart when an eval has not already authored equivalent run-health UI. If discovery metadata is missing but saved runs contain runtime columns such as `costUsd`, `inputTokens`, or `apiCalls`, the single-eval page can infer the standard usage stats and charts from those saved run values. Default usage columns, stats, and charts use `hideIfNoValue: true`. Default LLM usage charts configure cost, input tokens, and output tokens separately and use `dedupeConsecutiveValues: true` to skip repeated adjacent chart values. `totalTokens` is input + output only; cache read/write tokens stay separate and affect `costUsd` at their own rates. `llmTurns` is the maximum per-call turn count in the case run, using configured steps when available and otherwise one turn per matched LLM call span. Derived base input cost uses `inputTokens - cachedInputTokens - cacheCreationInputTokens` so cache details are not double-counted. `cacheCreationInputTokens` is the total cache-write count; optional `cacheCreationInput1hTokens` only splits that total for 1-hour write pricing via `cacheCreationInput1hUsdPerMillion`. `llmDurationMs` sums elapsed matched LLM span durations; it is not time-to-first-token latency. Remove defaults globally or per eval with `removeDefaultConfig: true` or a key list such as `removeDefaultConfig: ['apiCalls', 'reasoningTokens']`.
 - `apiCalls` (in `agent-evals.config.ts`) configures how API-call spans are summarized for review. Defaults to `kind: 'api'`, `'http'`, `'http.client'`, and `'fetch'` spans with `method`, `url`, `statusCode`, `request`, `routeAlias`, `response`, `requestBody`, `responseBody`, `headers`, `durationMs`, and `error` read from conventional attribute paths. Override `kinds` or `attributes.<field>` for external tracers. Set a per-span `routeAlias` attribute such as `/v3/tabs/:id` to group dynamic URL paths in API-call route labels and endpoint charts while preserving original URLs in row details. Add `derivedAttributes` as a keyed map or object-returning callback for computed persisted API span attributes, and add `metrics` with the same formats and placements as LLM-call metrics.
 - `runLogs` (in `agent-evals.config.ts`) controls case log capture. Use `runLogs: { captureConsole: false }` to keep console output in the terminal without persisting console calls to case details. Manual `evalLog(...)` calls are still captured. Captured log locations store the selected user-facing source frame and the full JavaScript stack so agents can inspect additional frames in persisted artifacts when diagnosing where a log came from.
 
-Stats rows and history charts can be authored via `stats` / `charts` on the eval definition. Global `stats` in `agent-evals.config.ts` combine with eval-level stats. Native stat kinds include `cases`, `passRate`, `duration`, and `cacheHits`; `cacheHits` shows Agent Eval operation-level cache hits over total cache operations (`hits/total`) from spans and `evalTracer.cache(...)` refs, not LLM provider prompt-cache read tokens such as `cachedInputTokens`. Cache-hit stats use a separate aggregate control and default to `sum`; `avg` is average per-case hit rate, and min/max/best/worst select cases by hit rate. `duration` aggregates per-case durations using the same modes as column stats. Usage stats and LLM usage charts are added by default unless removed with `removeDefaultConfig`. Column stats can override `format` and `numberFormat`, otherwise they inherit from the matching column. Duration and column stat aggregates support `avg`, `min`, `max`, `sum`, `best` (highest finite value), and `worst` (lowest finite value). Use `defaultStatAggregate` in `agent-evals.config.ts` to set the workspace-wide initial duration/column stat mode, or on an eval definition to override it for that eval. Number formats use `maxDecimalPlaces` to cap decimals and `minDecimalPlaces` to pad trailing zeroes. Without `maxDecimalPlaces`, the default cap is 3 decimal places. Stats and charts support `hideIfNoValue: true`. Charts support `dedupeConsecutiveValues: true` to omit consecutive points whose plotted metrics and tooltip extras match the previous kept point. Their shapes live in the types; no need to memorize the option set.
+Stats rows and history charts can be authored via `stats` / `charts` on the eval definition. Global `stats` in `agent-evals.config.ts` combine with eval-level stats. The web UI automatically supplies missing `cases`, `passRate`, and `duration` stats plus a pass-rate/duration history chart, including for a single completed run. Native stat kinds include `cases`, `passRate`, `duration`, and `cacheHits`; `cacheHits` shows Agent Eval operation-level cache hits over total cache operations (`hits/total`) from spans and `evalTracer.cache(...)` refs, not LLM provider prompt-cache read tokens such as `cachedInputTokens`. Cache-hit stats use a separate aggregate control and default to `sum`; `avg` is average per-case hit rate, and min/max/best/worst select cases by hit rate. `duration` aggregates per-case durations using the same modes as column stats. Usage stats and LLM usage charts are added by default unless removed with `removeDefaultConfig`. Column stats can override `format` and `numberFormat`, otherwise they inherit from the matching column. Duration and column stat aggregates support `avg`, `min`, `max`, `sum`, `best` (highest finite value), and `worst` (lowest finite value). Use `defaultStatAggregate` in `agent-evals.config.ts` to set the workspace-wide initial duration/column stat mode, or on an eval definition to override it for that eval. Number formats use `maxDecimalPlaces` to cap decimals and `minDecimalPlaces` to pad trailing zeroes. Without `maxDecimalPlaces`, the default cap is 3 decimal places. Stats and charts support `hideIfNoValue: true`. Charts support `dedupeConsecutiveValues: true` to omit consecutive points whose plotted metrics and tooltip extras match the previous kept point. Rendered charts with no plottable values show an unavailable state instead of a blank frame. Their shapes live in the types; no need to memorize the option set.
 
 ## Cached operations