npm - @ls-stack/agent-eval - Versions diffs - 0.56.0 → 0.56.1 - Mend

@ls-stack/agent-eval 0.56.0 → 0.56.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (16) hide show

package/dist/{app-Bpe6Monh.mjs → app-Db_x-Rit.mjs} +4 -4
package/dist/apps/web/dist/assets/{index-CfSiAVmi.js → index-bB8IBDp1.js} +27 -27
package/dist/apps/web/dist/index.html +1 -1
package/dist/bin.mjs +1 -1
package/dist/caseChild.mjs +1 -1
package/dist/{cli-DQO2Fpt2.mjs → cli-Ck0mqxd-.mjs} +4 -4
package/dist/index.d.mts +20 -20
package/dist/index.mjs +3 -3
package/dist/runChild.mjs +2 -2
package/dist/{runExecution-6lrtj48K.mjs → runExecution-BH7DlMXl.mjs} +5 -1
package/dist/{runOrchestration-BYaN2mzS.mjs → runOrchestration-C1Ex9QI-.mjs} +1 -1
package/dist/{runner-DYlwuAT3.mjs → runner-B3hEOT_I.mjs} +2 -2
package/dist/{runner-C3CiS2o7.mjs → runner-DbVYcapC.mjs} +1 -1
package/dist/{src-DCGrFAmO.mjs → src-B3iq-tuv.mjs} +2 -2
package/package.json +3 -3
package/skills/agent-eval/SKILL.md +3 -1

package/dist/apps/web/dist/index.html CHANGED Viewed

@@ -25,7 +25,7 @@
       href="https://fonts.googleapis.com/css2?family=Geist:wght@400;500;600&family=JetBrains+Mono:wght@400;500&display=swap"
       rel="stylesheet"
     />
-    <script type="module" crossorigin src="/assets/index-CfSiAVmi.js"></script>
+    <script type="module" crossorigin src="/assets/index-bB8IBDp1.js"></script>
     <link rel="stylesheet" crossorigin href="/assets/index-Xa_7PteQ.css">
   </head>
   <body>

package/dist/bin.mjs CHANGED Viewed

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { t as runCli } from "./cli-DQO2Fpt2.mjs";
+import { t as runCli } from "./cli-Ck0mqxd-.mjs";
 import { spawn } from "node:child_process";
 //#region src/bin.ts
 const moduleMocksFlag = "--experimental-test-module-mocks";

package/dist/caseChild.mjs CHANGED Viewed

@@ -1,4 +1,4 @@
-import { I as configureEvalRunLogs, Pt as runWithEvalRegistry, St as resolveLlmCallsConfig, _ as createBufferedCacheStore, a as isCaseChildParentMessage, d as loadEvalModule, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, q as runInEvalRuntimeScope, r as runCase, v as createFsCacheStore, xt as resolveApiCallsConfig } from "./runExecution-6lrtj48K.mjs";
+import { I as configureEvalRunLogs, Pt as runWithEvalRegistry, St as resolveLlmCallsConfig, _ as createBufferedCacheStore, a as isCaseChildParentMessage, d as loadEvalModule, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, q as runInEvalRuntimeScope, r as runCase, v as createFsCacheStore, xt as resolveApiCallsConfig } from "./runExecution-BH7DlMXl.mjs";
 //#region ../runner/src/caseChild.ts
 let fatalErrorReported = false;
 let disconnectExpected = false;

package/dist/{cli-DQO2Fpt2.mjs → cli-Ck0mqxd-.mjs} RENAMED Viewed

@@ -1,5 +1,5 @@
-import { Ct as buildEvalKey, Nt as getEvalRegistry, St as resolveLlmCallsConfig, bt as runSummarySchema, c as resolveArtifactPath, ct as applyDerivedCallAttributes, dt as getEvalDisplayStatus, f as resolveEvalDefaultConfig, ft as deriveScopedSummaryFromCases, gt as matchesTagsFilter, h as normalizeScoreDef, m as buildDeclaredColumnDefs, o as stripTerminalControlCodes, p as loadConfig, s as resolveTracePresentation, ut as getEvalTitle, v as createFsCacheStore, wt as getCaseRowCaseKey, xt as resolveApiCallsConfig } from "./runExecution-6lrtj48K.mjs";
-import { C as validateCharts, S as parseEvalDiscovery, _ as runTouchesEval, a as validateTagsFilters, b as deriveEvalFreshness, c as getLatestRunInfos, d as nextShortIdFromSnapshots, f as persistCaseDetail, g as recomputePersistedCaseStatus, h as recomputeEvalStatusesInRuns, i as resolveEvalTags, l as loadPersistedRunSnapshot, m as persistRunState, n as getTargetEvalKeys, o as generateRunId, p as deleteTemporaryRuns, s as getLastRunStatuses, u as loadPersistedRunSnapshots, v as buildManualInputDescriptor, x as loadIsolatedEvalRegistry, y as parseManualInputValues } from "./runOrchestration-BYaN2mzS.mjs";
+import { Ct as buildEvalKey, Nt as getEvalRegistry, St as resolveLlmCallsConfig, bt as runSummarySchema, c as resolveArtifactPath, ct as applyDerivedCallAttributes, dt as getEvalDisplayStatus, f as resolveEvalDefaultConfig, ft as deriveScopedSummaryFromCases, gt as matchesTagsFilter, h as normalizeScoreDef, m as buildDeclaredColumnDefs, o as stripTerminalControlCodes, p as loadConfig, s as resolveTracePresentation, ut as getEvalTitle, v as createFsCacheStore, wt as getCaseRowCaseKey, xt as resolveApiCallsConfig } from "./runExecution-BH7DlMXl.mjs";
+import { C as validateCharts, S as parseEvalDiscovery, _ as runTouchesEval, a as validateTagsFilters, b as deriveEvalFreshness, c as getLatestRunInfos, d as nextShortIdFromSnapshots, f as persistCaseDetail, g as recomputePersistedCaseStatus, h as recomputeEvalStatusesInRuns, i as resolveEvalTags, l as loadPersistedRunSnapshot, m as persistRunState, n as getTargetEvalKeys, o as generateRunId, p as deleteTemporaryRuns, s as getLastRunStatuses, u as loadPersistedRunSnapshots, v as buildManualInputDescriptor, x as loadIsolatedEvalRegistry, y as parseManualInputValues } from "./runOrchestration-C1Ex9QI-.mjs";
 import { copyFile, mkdir, readFile, rm, writeFile } from "node:fs/promises";
 import { basename, dirname, extname, isAbsolute, join, relative, resolve, sep } from "node:path";
 import { createHash, randomUUID } from "node:crypto";
@@ -2171,8 +2171,8 @@ async function commandApp(args) {
 	const { serve } = await import("@hono/node-server");
 	const bundledWebDist = resolve(currentDir, "apps/web/dist");
 	if (existsSync(bundledWebDist)) process.env.AGENT_EVALS_WEB_DIST = bundledWebDist;
-	const appModule = await import("./app-Bpe6Monh.mjs");
-	const runnerModule = await import("./runner-C3CiS2o7.mjs");
+	const appModule = await import("./app-Db_x-Rit.mjs");
+	const runnerModule = await import("./runner-DbVYcapC.mjs");
 	if (!isHonoAppModule(appModule)) throw new Error("Server app module is invalid");
 	if (!isServerRunnerModule(runnerModule)) throw new Error("Server runner module is invalid");
 	await runnerModule.initRunner();

package/dist/index.d.mts CHANGED Viewed

@@ -2660,9 +2660,9 @@ declare const caseRowSchema$1: z$1.ZodObject<{
     error: "error";
     running: "running";
     cancelled: "cancelled";
+    pending: "pending";
     pass: "pass";
     fail: "fail";
-    pending: "pending";
   }>;
   durationMs: z$1.ZodNullable<z$1.ZodNumber>;
   cacheHits: z$1.ZodOptional<z$1.ZodNumber>;
@@ -2860,10 +2860,10 @@ declare const scoreTraceSchema: z$1.ZodObject<{
     namespace: z$1.ZodString;
     key: z$1.ZodString;
     status: z$1.ZodEnum<{
+      bypass: "bypass";
+      refresh: "refresh";
       hit: "hit";
       miss: "miss";
-      refresh: "refresh";
-      bypass: "bypass";
     }>;
     read: z$1.ZodOptional<z$1.ZodBoolean>;
     stored: z$1.ZodOptional<z$1.ZodBoolean>;
@@ -2884,9 +2884,9 @@ declare const caseDetailSchema$1: z$1.ZodObject<{
     error: "error";
     running: "running";
     cancelled: "cancelled";
+    pending: "pending";
     pass: "pass";
     fail: "fail";
-    pending: "pending";
   }>;
   input: z$1.ZodUnknown;
   trace: z$1.ZodArray<z$1.ZodObject<{
@@ -3032,10 +3032,10 @@ declare const caseDetailSchema$1: z$1.ZodObject<{
       namespace: z$1.ZodString;
       key: z$1.ZodString;
       status: z$1.ZodEnum<{
+        bypass: "bypass";
+        refresh: "refresh";
         hit: "hit";
         miss: "miss";
-        refresh: "refresh";
-        bypass: "bypass";
       }>;
       read: z$1.ZodOptional<z$1.ZodBoolean>;
       stored: z$1.ZodOptional<z$1.ZodBoolean>;
@@ -3138,10 +3138,10 @@ declare const caseDetailSchema$1: z$1.ZodObject<{
     namespace: z$1.ZodString;
     key: z$1.ZodString;
     status: z$1.ZodEnum<{
+      bypass: "bypass";
+      refresh: "refresh";
       hit: "hit";
       miss: "miss";
-      refresh: "refresh";
-      bypass: "bypass";
     }>;
     read: z$1.ZodOptional<z$1.ZodBoolean>;
     stored: z$1.ZodOptional<z$1.ZodBoolean>;
@@ -3499,8 +3499,8 @@ declare const runManifestSchema$1: z$1.ZodObject<{
   target: z$1.ZodObject<{
     mode: z$1.ZodEnum<{
       all: "all";
-      caseIds: "caseIds";
       evalIds: "evalIds";
+      caseIds: "caseIds";
     }>;
     evalKeys: z$1.ZodOptional<z$1.ZodArray<z$1.ZodString>>;
     files: z$1.ZodOptional<z$1.ZodArray<z$1.ZodString>>;
@@ -3514,9 +3514,9 @@ declare const runManifestSchema$1: z$1.ZodObject<{
     median: "median";
   }>>>;
   cacheMode: z$1.ZodOptional<z$1.ZodEnum<{
-    refresh: "refresh";
-    bypass: "bypass";
     use: "use";
+    bypass: "bypass";
+    refresh: "refresh";
   }>>;
 }, z$1.core.$strip>;
 /** Persisted lifecycle metadata for a single eval run. */
@@ -4436,9 +4436,9 @@ declare function extractApiCalls(spans: EvalTraceSpan$1[], config: ResolvedApiCa
  * - `refresh`: never read, always write (forces re-execution and overwrites).
  */
 declare const cacheModeSchema: z$1.ZodEnum<{
-  refresh: "refresh";
-  bypass: "bypass";
   use: "use";
+  bypass: "bypass";
+  refresh: "refresh";
 }>;
 /** Mode controlling how cached spans behave during a run. */
 type CacheMode = z$1.infer<typeof cacheModeSchema>;
@@ -4459,10 +4459,10 @@ declare const cacheOperationTypeSchema: z$1.ZodEnum<{
 type CacheOperationType = z$1.infer<typeof cacheOperationTypeSchema>;
 /** Status of a cache lookup recorded on a span or case scope. */
 declare const cacheStatusSchema: z$1.ZodEnum<{
+  bypass: "bypass";
+  refresh: "refresh";
   hit: "hit";
   miss: "miss";
-  refresh: "refresh";
-  bypass: "bypass";
 }>;
 /** Status of a cache lookup recorded on a span or case scope. */
 type CacheStatus = z$1.infer<typeof cacheStatusSchema>;
@@ -4479,10 +4479,10 @@ declare const traceCacheRefSchema: z$1.ZodObject<{
   namespace: z$1.ZodString;
   key: z$1.ZodString;
   status: z$1.ZodEnum<{
+    bypass: "bypass";
+    refresh: "refresh";
     hit: "hit";
     miss: "miss";
-    refresh: "refresh";
-    bypass: "bypass";
   }>;
   read: z$1.ZodOptional<z$1.ZodBoolean>;
   stored: z$1.ZodOptional<z$1.ZodBoolean>;
@@ -5435,8 +5435,8 @@ declare const createRunRequestSchema$1: z$1.ZodObject<{
   target: z$1.ZodObject<{
     mode: z$1.ZodEnum<{
       all: "all";
-      caseIds: "caseIds";
       evalIds: "evalIds";
+      caseIds: "caseIds";
     }>;
     evalKeys: z$1.ZodOptional<z$1.ZodArray<z$1.ZodString>>;
     files: z$1.ZodOptional<z$1.ZodArray<z$1.ZodString>>;
@@ -5448,9 +5448,9 @@ declare const createRunRequestSchema$1: z$1.ZodObject<{
   temporary: z$1.ZodOptional<z$1.ZodBoolean>;
   cache: z$1.ZodOptional<z$1.ZodObject<{
     mode: z$1.ZodDefault<z$1.ZodEnum<{
-      refresh: "refresh";
-      bypass: "bypass";
       use: "use";
+      bypass: "bypass";
+      refresh: "refresh";
     }>>;
   }, z$1.core.$strip>>;
   manualInputs: z$1.ZodOptional<z$1.ZodRecord<z$1.ZodString, z$1.ZodUnknown>>;

package/dist/index.mjs CHANGED Viewed

@@ -1,4 +1,4 @@
-import { $ as startEvalBackgroundJob, A as repoFile, B as getCurrentScope, C as evalTracer, D as deserializeCacheValue, E as deserializeCacheRecording, F as appendToEvalOutput, G as mergeEvalOutput, H as incrementEvalOutput, J as runInEvalScope, K as nextEvalId, L as evalAssert, M as readManualInputFile, N as evalExpect, Nt as getEvalRegistry, O as serializeCacheRecording, P as EvalAssertionError, Q as setScopeCacheContext, R as evalLog, S as evalSpan, T as hashCacheKeySync, U as isInEvalScope, V as getEvalCaseInput, Y as runInExistingEvalScope, Z as setEvalOutput, at as extractLlmCalls, b as buildTraceTree, it as extractApiCalls, j as manualInputFileValueSchema, k as serializeCacheValue, lt as getNestedAttribute, nt as extractCacheEntries, ot as simulateLlmCallCost, q as runInEvalRuntimeScope, rt as extractCacheHits, st as simulateTokenAllocation, w as hashCacheKey, x as captureEvalSpanError, y as z, z as evalTime } from "./runExecution-6lrtj48K.mjs";
-import { a as materializeManualInputFiles, i as isManualInputFileValue, n as createRunner, o as stageManualInputFile, r as cleanupStagedManualInputFiles, s as stageManualInputFileFromPath, t as runCli } from "./cli-DQO2Fpt2.mjs";
-import { n as matchesEvalTags, t as defineEval } from "./src-DCGrFAmO.mjs";
+import { $ as startEvalBackgroundJob, A as repoFile, B as getCurrentScope, C as evalTracer, D as deserializeCacheValue, E as deserializeCacheRecording, F as appendToEvalOutput, G as mergeEvalOutput, H as incrementEvalOutput, J as runInEvalScope, K as nextEvalId, L as evalAssert, M as readManualInputFile, N as evalExpect, Nt as getEvalRegistry, O as serializeCacheRecording, P as EvalAssertionError, Q as setScopeCacheContext, R as evalLog, S as evalSpan, T as hashCacheKeySync, U as isInEvalScope, V as getEvalCaseInput, Y as runInExistingEvalScope, Z as setEvalOutput, at as extractLlmCalls, b as buildTraceTree, it as extractApiCalls, j as manualInputFileValueSchema, k as serializeCacheValue, lt as getNestedAttribute, nt as extractCacheEntries, ot as simulateLlmCallCost, q as runInEvalRuntimeScope, rt as extractCacheHits, st as simulateTokenAllocation, w as hashCacheKey, x as captureEvalSpanError, y as z, z as evalTime } from "./runExecution-BH7DlMXl.mjs";
+import { a as materializeManualInputFiles, i as isManualInputFileValue, n as createRunner, o as stageManualInputFile, r as cleanupStagedManualInputFiles, s as stageManualInputFileFromPath, t as runCli } from "./cli-Ck0mqxd-.mjs";
+import { n as matchesEvalTags, t as defineEval } from "./src-B3iq-tuv.mjs";
 export { EvalAssertionError, appendToEvalOutput, buildTraceTree, captureEvalSpanError, cleanupStagedManualInputFiles, createRunner, defineEval, deserializeCacheRecording, deserializeCacheValue, evalAssert, evalExpect, evalLog, evalSpan, evalTime, evalTracer, extractApiCalls, extractCacheEntries, extractCacheHits, extractLlmCalls, getCurrentScope, getEvalCaseInput, getEvalRegistry, getNestedAttribute, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, isManualInputFileValue, manualInputFileValueSchema, matchesEvalTags, materializeManualInputFiles, mergeEvalOutput, nextEvalId, readManualInputFile, repoFile, runCli, runInEvalRuntimeScope, runInEvalScope, runInExistingEvalScope, serializeCacheRecording, serializeCacheValue, setEvalOutput, setScopeCacheContext, simulateLlmCallCost, simulateTokenAllocation, stageManualInputFile, stageManualInputFileFromPath, startEvalBackgroundJob, z };

package/dist/runChild.mjs CHANGED Viewed

@@ -1,5 +1,5 @@
-import { At as evalChartsConfigSchema, Ct as buildEvalKey, Dt as evalStatAggregateSchema, I as configureEvalRunLogs, Ot as evalStatsConfigSchema, bt as runSummarySchema, et as createRunRequestSchema, jt as columnDefSchema, kt as manualInputDescriptorSchema, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, v as createFsCacheStore, yt as runManifestSchema } from "./runExecution-6lrtj48K.mjs";
-import { S as parseEvalDiscovery, m as persistRunState, r as getTargetEvals$1, t as executeRun } from "./runOrchestration-BYaN2mzS.mjs";
+import { At as evalChartsConfigSchema, Ct as buildEvalKey, Dt as evalStatAggregateSchema, I as configureEvalRunLogs, Ot as evalStatsConfigSchema, bt as runSummarySchema, et as createRunRequestSchema, jt as columnDefSchema, kt as manualInputDescriptorSchema, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, v as createFsCacheStore, yt as runManifestSchema } from "./runExecution-BH7DlMXl.mjs";
+import { S as parseEvalDiscovery, m as persistRunState, r as getTargetEvals$1, t as executeRun } from "./runOrchestration-C1Ex9QI-.mjs";
 import { z } from "zod/v4";
 import { readFile } from "node:fs/promises";
 import { relative } from "node:path";

package/dist/{runExecution-6lrtj48K.mjs → runExecution-BH7DlMXl.mjs} RENAMED Viewed

@@ -6181,6 +6181,10 @@ function sumNullable(values) {
 	}
 	return hasValue ? total : void 0;
 }
+function getMaxLlmTurns(calls) {
+	if (calls.length === 0) return void 0;
+	return Math.max(...calls.map((call) => Math.max(call.stepCount ?? 1, 1)));
+}
 function assignIfMissing(params) {
 	if (!params.activeKeys.has(params.key)) return;
 	if (params.key in params.outputs) return;
@@ -6202,7 +6206,7 @@ function addDefaultOutputs(params) {
 	assignIfMissing({
 		outputs: params.outputs,
 		key: "llmTurns",
-		value: calls.length,
+		value: getMaxLlmTurns(calls),
 		activeKeys
 	});
 	assignIfMissing({

package/dist/{runOrchestration-BYaN2mzS.mjs → runOrchestration-C1Ex9QI-.mjs} RENAMED Viewed

@@ -1,4 +1,4 @@
-import { Et as caseRowSchema, Pt as runWithEvalRegistry, Tt as caseDetailSchema, X as runWithEvalClock, _t as validateEvalTagName, bt as runSummarySchema, d as loadEvalModule, f as resolveEvalDefaultConfig, ft as deriveScopedSummaryFromCases, g as commitPendingCacheWrites, gt as matchesTagsFilter, ht as dedupeEvalTags, i as isCaseChildMessage, m as buildDeclaredColumnDefs, mt as deriveStatusFromChildStatuses, n as resolveRunnableEvalCases, o as stripTerminalControlCodes, pt as deriveStatusFromCaseRows, q as runInEvalRuntimeScope, t as filterEvalCases, u as runWithModuleIsolation, vt as validateTagsFilterExpression, wt as getCaseRowCaseKey, yt as runManifestSchema } from "./runExecution-6lrtj48K.mjs";
+import { Et as caseRowSchema, Pt as runWithEvalRegistry, Tt as caseDetailSchema, X as runWithEvalClock, _t as validateEvalTagName, bt as runSummarySchema, d as loadEvalModule, f as resolveEvalDefaultConfig, ft as deriveScopedSummaryFromCases, g as commitPendingCacheWrites, gt as matchesTagsFilter, ht as dedupeEvalTags, i as isCaseChildMessage, m as buildDeclaredColumnDefs, mt as deriveStatusFromChildStatuses, n as resolveRunnableEvalCases, o as stripTerminalControlCodes, pt as deriveStatusFromCaseRows, q as runInEvalRuntimeScope, t as filterEvalCases, u as runWithModuleIsolation, vt as validateTagsFilterExpression, wt as getCaseRowCaseKey, yt as runManifestSchema } from "./runExecution-BH7DlMXl.mjs";
 import { readFile, readdir, rm, writeFile } from "node:fs/promises";
 import { dirname, join } from "node:path";
 import { existsSync } from "node:fs";

package/dist/{runner-DYlwuAT3.mjs → runner-B3hEOT_I.mjs} RENAMED Viewed

@@ -1,5 +1,5 @@
-import { n as createRunner } from "./cli-DQO2Fpt2.mjs";
-import "./src-DCGrFAmO.mjs";
+import { n as createRunner } from "./cli-Ck0mqxd-.mjs";
+import "./src-B3iq-tuv.mjs";
 //#region ../../apps/server/src/runner.ts
 let runnerInstance = null;
 function getRunnerInstance() {

package/dist/{runner-C3CiS2o7.mjs → runner-DbVYcapC.mjs} RENAMED Viewed

@@ -1,2 +1,2 @@
-import { n as initRunner, t as getRunnerInstance } from "./runner-DYlwuAT3.mjs";
+import { n as initRunner, t as getRunnerInstance } from "./runner-B3hEOT_I.mjs";
 export { getRunnerInstance, initRunner };

package/dist/{src-DCGrFAmO.mjs → src-B3iq-tuv.mjs} RENAMED Viewed

@@ -1,5 +1,5 @@
-import { Mt as defineEval$1, W as matchesEvalTags$1 } from "./runExecution-6lrtj48K.mjs";
-import "./cli-DQO2Fpt2.mjs";
+import { Mt as defineEval$1, W as matchesEvalTags$1 } from "./runExecution-BH7DlMXl.mjs";
+import "./cli-Ck0mqxd-.mjs";
 //#region src/index.ts
 /** Register an eval definition with typed tag support. */
 function defineEval(definition) {

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "@ls-stack/agent-eval",
-  "version": "0.56.0",
+  "version": "0.56.1",
   "type": "module",
   "bin": {
     "agent-evals": "./dist/bin.mjs"
@@ -33,8 +33,8 @@
     "@types/node": "^24.7.2",
     "typescript": "^5.9.2",
     "@agent-evals/runner": "0.0.1",
-    "@agent-evals/sdk": "0.0.1",
-    "@agent-evals/shared": "0.0.1"
+    "@agent-evals/shared": "0.0.1",
+    "@agent-evals/sdk": "0.0.1"
   },
   "scripts": {
     "build": "pnpm --filter @agent-evals/web build && pnpm --filter @agent-evals/shared build && pnpm --filter @agent-evals/sdk build && pnpm --filter @agent-evals/runner build && tsdown --filter cli-js && tsdown --filter cli-types",

package/skills/agent-eval/SKILL.md CHANGED Viewed

@@ -393,7 +393,9 @@ See `EvalScoreDef` / `EvalManualScoreDef` in the types for the full shape
   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.
+  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