@iowarp/clio-coder 0.4.2 → 0.4.3

package/dist/{eval-C45FYRJ6.js → eval-7G7SGAYO.js}

@@ -1,11 +1,14 @@
 import { createRequire as __clioCreateRequire } from "node:module"; const require = __clioCreateRequire(import.meta.url);
 import {
   EVAL_SUITE_V2_VERSION
-} from "./chunk-RKSR6VSF.js";
+} from "./chunk-4IUZQIJ3.js";
 import {
   loadFragments,
   renderCodewikiDigest
-} from "./chunk-AZ4WMN4W.js";
+} from "./chunk-Y3CBHOR6.js";
+import {
+  discoverAgentRecipes
+} from "./chunk-RRNP2ANY.js";
 import {
   EvalTaskFileError,
   TRUST_STATUS_AXES,
@@ -20,12 +23,11 @@ import {
   loadEvalTaskFile,
   summarizeTrustStatus,
   verifyReceiptIntegrity
-} from "./chunk-B74PXLU7.js";
+} from "./chunk-IWT4SF4R.js";
 import {
   listSessionLedgerRefs,
   parseSessionEntries
-} from "./chunk-X7IARSHT.js";
-import "./chunk-3DPEIQKN.js";
+} from "./chunk-JKKCYP3C.js";
 import {
   EVAL_BEHAVIOR_METRIC_DEFINITIONS_V1,
   EVAL_EXECUTION_ENVELOPE_SCHEMA_V1,
@@ -46,38 +48,60 @@ import {
   renderEvalServingConfiguration,
   sameEvalServingConfiguration,
   writeEvalArtifactV4
-} from "./chunk-5PFYMY2V.js";
+} from "./chunk-FTMGRKEF.js";
 import "./chunk-W6GROXXM.js";
+import "./chunk-RKKLTLYB.js";
 import {
   shellQuote
 } from "./chunk-TXOTCRLG.js";
+import "./chunk-LDJG7DW3.js";
+import "./chunk-NDDGJUXD.js";
+import "./chunk-CE5AX47J.js";
+import "./chunk-3DPEIQKN.js";
+import "./chunk-7DRAWPTZ.js";
+import "./chunk-5MEZN6CB.js";
+import "./chunk-PUVDKJ2Y.js";
 import "./chunk-HVDIIIQW.js";
 import {
-  discoverAgentRecipes
-} from "./chunk-CRFOIAX3.js";
-import "./chunk-O3YUNJZ2.js";
-import "./chunk-NDDGJUXD.js";
-import "./chunk-XGDPUNND.js";
-import "./chunk-RKKLTLYB.js";
-import "./chunk-HLAFFSEK.js";
-import "./chunk-77QIVUZB.js";
-import "./chunk-HIICAHCJ.js";
-import "./chunk-Q4XWMHX6.js";
+  printError
+} from "./chunk-VPKWYKEY.js";
+import "./chunk-5TSRNF4G.js";
+import "./chunk-EKCHAPYA.js";
+import "./chunk-L47TF46W.js";
 import {
   createSafetyPolicyEngine
-} from "./chunk-FVDGR2ZL.js";
+} from "./chunk-XIVNBFZS.js";
 import "./chunk-H7IXIC72.js";
-import "./chunk-VN3SHNBN.js";
-import "./chunk-FSP7CMNU.js";
+import "./chunk-7DICMOS6.js";
 import "./chunk-HHV2GANA.js";
 import {
   agentSpecFingerprint,
   normalizeAgentSpec
-} from "./chunk-PGF63K6I.js";
-import "./chunk-IKOZFYBN.js";
+} from "./chunk-B4OAX3SI.js";
+import "./chunk-I7ZPNEJM.js";
+import "./chunk-GYV6VZOC.js";
 import "./chunk-UL3WSD3F.js";
 import "./chunk-ECH6PKUQ.js";
+import "./chunk-JGRC33J2.js";
 import "./chunk-K6BSR66V.js";
+import "./chunk-CFGTUFWB.js";
+import {
+  extractReasoningTokens
+} from "./chunk-NZMNUPZZ.js";
+import "./chunk-HAXOFFRH.js";
+import {
+  InvalidIdError
+} from "./chunk-R346GLFC.js";
+import "./chunk-KXDSS5WJ.js";
+import "./chunk-GHS5EBTQ.js";
+import {
+  namingCompatibilityEnvironment
+} from "./chunk-D7E7WCDF.js";
+import {
+  normalizeClioCoderEventRecord,
+  normalizeClioCoderEventType
+} from "./chunk-BTK3SSSK.js";
+import "./chunk-F4EKGO4N.js";
 import {
   readCodewiki,
   structuralCodewikiHash
@@ -90,50 +114,27 @@ import {
 } from "./chunk-33YXPOE3.js";
 import "./chunk-7CR24IG7.js";
 import "./chunk-XPLRXC72.js";
-import "./chunk-ZW55JB7N.js";
-import {
-  printError
-} from "./chunk-VPKWYKEY.js";
-import "./chunk-5TSRNF4G.js";
-import "./chunk-CFGTUFWB.js";
-import {
-  extractReasoningTokens
-} from "./chunk-CWVRRIEI.js";
-import "./chunk-MQXIVJ35.js";
-import {
-  InvalidIdError
-} from "./chunk-R346GLFC.js";
-import {
-  normalizeClioCoderEventRecord,
-  normalizeClioCoderEventType
-} from "./chunk-BTK3SSSK.js";
-import "./chunk-IHXBNWMM.js";
-import "./chunk-34BHNEE3.js";
-import {
-  namingCompatibilityEnvironment
-} from "./chunk-D7E7WCDF.js";
-import "./chunk-HJWWJ6IL.js";
-import "./chunk-F4EKGO4N.js";
-import "./chunk-JIEGK6UF.js";
-import "./chunk-EBY63B7G.js";
-import "./chunk-6EJMN2Y3.js";
 import "./chunk-IWHMRKLL.js";
 import "./chunk-UXCU4E3T.js";
 import "./chunk-LL4KHSZI.js";
+import "./chunk-JSC3U7TI.js";
 import "./chunk-S66XZJOF.js";
-import "./chunk-WNP7O5WZ.js";
+import "./chunk-JIEGK6UF.js";
+import "./chunk-EBY63B7G.js";
+import "./chunk-6EJMN2Y3.js";
+import "./chunk-ID64D7PE.js";
 import "./chunk-336NSSKI.js";
 import "./chunk-IKCO5N3L.js";
 import "./chunk-3I7MS7N2.js";
 import {
   require_dist
 } from "./chunk-APJ265NV.js";
+import "./chunk-WEPFGWHJ.js";
+import "./chunk-YXLYO42X.js";
 import {
   clioDataDir,
   clioStateDir
 } from "./chunk-BYP5D4HI.js";
-import "./chunk-WEPFGWHJ.js";
-import "./chunk-YXLYO42X.js";
 import {
   __toESM,
   init_esm_shims
@@ -1466,7 +1467,15 @@ function artifactMatrixIdentity(targets) {
 }
 function resolveTargets(targets, options) {
   const filtered = options.target === void 0 ? [...targets] : targets.filter((target) => target.id === options.target);
-  const selected = filtered.length > 0 ? filtered : options.target === void 0 ? [...targets] : [{ id: options.target }];
+  const selected = filtered.length > 0 ? filtered : options.target === void 0 ? [...targets] : [
+    {
+      id: options.target,
+      // A one-row suite still owns its thinking dimension when the CLI
+      // substitutes the target/model pair. Dropping it silently changes
+      // an off baseline into the target's configured default.
+      ...targets.length === 1 && targets[0]?.thinking !== void 0 ? { thinking: targets[0].thinking } : {}
+    }
+  ];
   return selected.map((target) => ({
     ...target,
     ...options.model === void 0 ? {} : { model: options.model }
@@ -2285,6 +2294,16 @@ function zeroToolCallMetrics() {
 init_esm_shims();
 import { readFile as readFile2 } from "node:fs/promises";
 import { dirname as dirname2, join as join2 } from "node:path";
+function selectEvalLedgerEntries(durable, stream) {
+  const sessionCalls = assistantCalls(durable).length;
+  const streamCalls = assistantCalls(stream).length;
+  return {
+    entries: sessionCalls > 0 ? durable : [...durable, ...stream],
+    source: sessionCalls > 0 ? "session" : "stream",
+    sessionCalls,
+    streamCalls
+  };
+}
 async function readEvalLedgerSnapshot(stateDir) {
   const refs = await listSessionLedgerRefs(stateDir);
   const entries = [];
@@ -2384,7 +2403,7 @@ function emptyEvalTrackedMetrics(source = "estimated") {
     reasoningTokens: { value: null, source },
     toolCalls: zero(),
     toolErrors: zero(),
-    ttftMsFirstCall: zero(),
+    ttftMsFirstCall: { value: null, source: "estimated" },
     wallClockMs: zero(),
     contextTokensAtEnd: zero(),
     compactions: zero(),
@@ -2400,6 +2419,7 @@ function assistantCalls(entries) {
     if (promptCache === null && timing === null && usage === null) return [];
     return [
       {
+        timestamp: entry.payload.timestampEstimated === true ? null : entry.timestamp,
         payload: entry.payload,
         promptCache,
         backend: promptCache === null ? null : recordField(promptCache, "backend"),
@@ -2459,9 +2479,19 @@ function reasoningMetric(receipt, calls, compactionUsage) {
   return measured ? { value: total, source: "ledger" } : { value: null, source: "estimated" };
 }
 function firstCallTtft(calls) {
-  const first = calls[0];
+  let first;
+  let firstAt = Number.POSITIVE_INFINITY;
+  for (const call of calls) {
+    if (call.timestamp === null) return { value: null, source: "estimated" };
+    const at = Date.parse(call.timestamp);
+    if (!Number.isFinite(at)) return { value: null, source: "estimated" };
+    if (at < firstAt) {
+      first = call;
+      firstAt = at;
+    }
+  }
   const value = first?.timing === null || first?.timing === void 0 ? null : nonNegativeNumber(first.timing.ttftMs);
-  return value === null ? { value: 0, source: "estimated" } : { value, source: "ledger" };
+  return value === null ? { value: null, source: "estimated" } : { value, source: "ledger" };
 }
 function wallClockMetric(receipt, fallback) {
   if (receipt !== null) {
@@ -2665,13 +2695,15 @@ init_esm_shims();
 var UNMEASURED_TOKEN_USAGE = {
   measured: false,
   tokens: { input: 0, output: 0, total: 0, cacheRead: 0, cacheWrite: 0 },
-  costUsd: 0
+  costUsd: 0,
+  provider: emptyProviderObservation()
 };
 function createTokenUsageFold() {
   const tokens = { input: 0, output: 0, total: 0, cacheRead: 0, cacheWrite: 0 };
   let costUsd = 0;
   let measured = false;
   let pending = "";
+  const provider = emptyProviderObservation();
   const consume = (line) => {
     if (line.trim().length === 0) return;
     let event;
@@ -2680,23 +2712,16 @@ function createTokenUsageFold() {
     } catch {
       return;
     }
-    if (!isRecord6(event) || event.type !== "message_end") return;
-    const message = isRecord6(event.message) ? event.message : void 0;
-    if (message === void 0 || message.role !== "assistant") return;
-    const usage = isRecord6(message.usage) ? message.usage : void 0;
-    if (usage === void 0) return;
-    measured = true;
-    const input = numberField2(usage, "input");
-    const output = numberField2(usage, "output");
-    const cacheRead = numberField2(usage, "cacheRead");
-    const cacheWrite = numberField2(usage, "cacheWrite");
-    const totalTokens = numberField2(usage, "totalTokens");
-    tokens.input += input;
-    tokens.output += output;
-    tokens.cacheRead += cacheRead;
-    tokens.cacheWrite += cacheWrite;
-    tokens.total += totalTokens > 0 ? totalTokens : input + output + cacheRead + cacheWrite;
-    if (isRecord6(usage.cost)) costUsd += numberField2(usage.cost, "total");
+    if (!isRecord6(event)) return;
+    observeProviderEvent(provider, event);
+    if (event.type !== "message_end" || !isRecord6(event.message) || event.message.role !== "assistant") return;
+    const usage = readMessageUsage(event.message);
+    if (usage === null) return;
+    if (Object.keys(usage.tokens).length > 0) {
+      measured = true;
+      for (const field of TOKEN_FIELDS) tokens[field] += usage.tokens[field] ?? 0;
+    }
+    costUsd += usage.costUsd ?? 0;
   };
   return {
     push(chunk) {
@@ -2713,7 +2738,7 @@ function createTokenUsageFold() {
         consume(pending.replace(/\r$/u, ""));
         pending = "";
       }
-      return { measured, tokens: { ...tokens }, costUsd };
+      return { measured, tokens: { ...tokens }, costUsd, provider: structuredClone(provider) };
     }
   };
 }
@@ -2727,24 +2752,145 @@ function addTokenStreamUsage(left, right) {
       cacheRead: left.tokens.cacheRead + right.tokens.cacheRead,
       cacheWrite: left.tokens.cacheWrite + right.tokens.cacheWrite
     },
-    costUsd: left.costUsd + right.costUsd
+    costUsd: left.costUsd + right.costUsd,
+    provider: addProviderObservations(left.provider, right.provider)
   };
 }
 function tokenMetricEntries(usage) {
-  if (!usage.measured) return { "tokens.measured": false };
   return {
-    "tokens.measured": true,
-    "tokens.input": usage.tokens.input,
-    "tokens.output": usage.tokens.output,
-    "tokens.total": usage.tokens.total,
-    "tokens.cacheRead": usage.tokens.cacheRead,
-    "tokens.cacheWrite": usage.tokens.cacheWrite,
+    ...providerMetricEntries(usage.provider),
+    "tokens.measured": usage.measured,
+    ...usage.measured ? {
+      "tokens.input": usage.tokens.input,
+      "tokens.output": usage.tokens.output,
+      "tokens.total": usage.tokens.total,
+      "tokens.cacheRead": usage.tokens.cacheRead,
+      "tokens.cacheWrite": usage.tokens.cacheWrite
+    } : {},
     ...usage.costUsd > 0 ? { "cost.usd": usage.costUsd } : {}
   };
 }
-function numberField2(record, field) {
-  const value = record[field];
-  return typeof value === "number" && Number.isFinite(value) ? value : 0;
+var TOKEN_FIELDS = ["input", "output", "total", "cacheRead", "cacheWrite"];
+var STOP_REASONS = ["stop", "toolUse", "length", "error", "aborted", "other"];
+var RETRY_PHASES = {
+  scheduled: "retryScheduled",
+  retrying: "retryStarted",
+  cancelled: "retryCancelled",
+  exhausted: "retryExhausted",
+  recovered: "retryRecovered"
+};
+function emptyProviderObservation() {
+  return {
+    measured: false,
+    stops: { stop: 0, toolUse: 0, length: 0, error: 0, aborted: 0, other: 0 },
+    retries: { scheduled: 0, retrying: 0, cancelled: 0, exhausted: 0, recovered: 0 },
+    errorUsageObservedCalls: 0,
+    errorUsageUnobservedCalls: 0,
+    errorUsageIncompleteCalls: 0,
+    errorCostUnobservedCalls: 0,
+    errorReasoningUnobservedCalls: 0,
+    errorReasoningTokens: null,
+    errorTokens: {},
+    errorCostUsd: null
+  };
+}
+function readMessageUsage(message) {
+  if (!isRecord6(message.usage)) return null;
+  const usage = message.usage;
+  const failed = message.stopReason === "error" || message.stopReason === "aborted";
+  const raw = {};
+  for (const field of TOKEN_FIELDS) {
+    const value = nonNegativeNumber2(usage[field === "total" ? "totalTokens" : field]);
+    if (value !== null && (!failed || value > 0)) raw[field] = value;
+  }
+  const cost = isRecord6(usage.cost) ? nonNegativeNumber2(usage.cost.total) : null;
+  const costUsd = cost !== null && cost > 0 ? cost : null;
+  const reportedReasoning = extractReasoningTokens({ ...usage, reasoningTokens: void 0 });
+  const reasoningTokens = reportedReasoning !== null && reportedReasoning > 0 ? reportedReasoning : null;
+  const tokens = raw;
+  if (Object.keys(tokens).length === 0 && costUsd === null && reasoningTokens === null) return null;
+  const completeTokens = Object.keys(tokens).length === TOKEN_FIELDS.length;
+  if (Object.keys(tokens).length > 0 && !(tokens.total !== void 0 && tokens.total > 0)) {
+    tokens.total = (tokens.input ?? 0) + (tokens.output ?? 0) + (tokens.cacheRead ?? 0) + (tokens.cacheWrite ?? 0);
+  }
+  return { tokens, costUsd, completeTokens, reasoningTokens };
+}
+function observeProviderEvent(provider, event) {
+  if (event.type === "retry_status" && isRecord6(event.status)) {
+    const phase = event.status.phase;
+    if (typeof phase === "string" && Object.hasOwn(RETRY_PHASES, phase)) {
+      provider.measured = true;
+      provider.retries[phase] += 1;
+    }
+    return;
+  }
+  if (event.type !== "message_end" || !isRecord6(event.message) || event.message.role !== "assistant") return;
+  const reason = event.message.stopReason;
+  if (typeof reason !== "string" || reason.length === 0 || reason === "pending") return;
+  provider.measured = true;
+  const stop = STOP_REASONS.includes(reason) ? reason : "other";
+  provider.stops[stop] += 1;
+  if (reason !== "error") return;
+  const usage = readMessageUsage(event.message);
+  if (usage === null) provider.errorUsageUnobservedCalls += 1;
+  else {
+    provider.errorUsageObservedCalls += 1;
+    for (const field of TOKEN_FIELDS) {
+      const value = usage.tokens[field];
+      if (value !== void 0) provider.errorTokens[field] = (provider.errorTokens[field] ?? 0) + value;
+    }
+  }
+  if (usage?.completeTokens !== true) provider.errorUsageIncompleteCalls += 1;
+  if (usage?.costUsd === void 0 || usage.costUsd === null) provider.errorCostUnobservedCalls += 1;
+  else provider.errorCostUsd = (provider.errorCostUsd ?? 0) + usage.costUsd;
+  if (usage?.reasoningTokens === void 0 || usage.reasoningTokens === null)
+    provider.errorReasoningUnobservedCalls += 1;
+  else provider.errorReasoningTokens = (provider.errorReasoningTokens ?? 0) + usage.reasoningTokens;
+}
+function addProviderObservations(left, right) {
+  const result = emptyProviderObservation();
+  result.measured = left.measured || right.measured;
+  for (const reason of STOP_REASONS) result.stops[reason] = left.stops[reason] + right.stops[reason];
+  for (const phase of Object.keys(RETRY_PHASES))
+    result.retries[phase] = left.retries[phase] + right.retries[phase];
+  result.errorUsageObservedCalls = left.errorUsageObservedCalls + right.errorUsageObservedCalls;
+  result.errorUsageUnobservedCalls = left.errorUsageUnobservedCalls + right.errorUsageUnobservedCalls;
+  result.errorUsageIncompleteCalls = left.errorUsageIncompleteCalls + right.errorUsageIncompleteCalls;
+  result.errorCostUnobservedCalls = left.errorCostUnobservedCalls + right.errorCostUnobservedCalls;
+  result.errorReasoningUnobservedCalls = left.errorReasoningUnobservedCalls + right.errorReasoningUnobservedCalls;
+  if (left.errorReasoningTokens !== null || right.errorReasoningTokens !== null)
+    result.errorReasoningTokens = (left.errorReasoningTokens ?? 0) + (right.errorReasoningTokens ?? 0);
+  for (const field of TOKEN_FIELDS) {
+    if (left.errorTokens[field] !== void 0 || right.errorTokens[field] !== void 0) {
+      result.errorTokens[field] = (left.errorTokens[field] ?? 0) + (right.errorTokens[field] ?? 0);
+    }
+  }
+  if (left.errorCostUsd !== null || right.errorCostUsd !== null)
+    result.errorCostUsd = (left.errorCostUsd ?? 0) + (right.errorCostUsd ?? 0);
+  return result;
+}
+function providerMetricEntries(provider) {
+  if (!provider.measured) return { "provider.measured": false };
+  const metrics = { "provider.measured": true };
+  for (const reason of STOP_REASONS) metrics[`provider.stopReason.${reason}`] = provider.stops[reason];
+  for (const phase of Object.keys(RETRY_PHASES))
+    metrics[`provider.${RETRY_PHASES[phase]}`] = provider.retries[phase];
+  if (provider.stops.error === 0) return metrics;
+  metrics["provider.errorUsageObservedCalls"] = provider.errorUsageObservedCalls;
+  metrics["provider.errorUsageUnobservedCalls"] = provider.errorUsageUnobservedCalls;
+  metrics["provider.errorUsageIncompleteCalls"] = provider.errorUsageIncompleteCalls;
+  metrics["provider.errorCostUnobservedCalls"] = provider.errorCostUnobservedCalls;
+  metrics["provider.errorReasoningUnobservedCalls"] = provider.errorReasoningUnobservedCalls;
+  if (provider.errorReasoningTokens !== null) metrics["provider.errorReasoningTokens"] = provider.errorReasoningTokens;
+  for (const field of TOKEN_FIELDS) {
+    const value = provider.errorTokens[field];
+    if (value !== void 0) metrics[`provider.errorTokens.${field}`] = value;
+  }
+  if (provider.errorCostUsd !== null) metrics["provider.errorCostUsd"] = provider.errorCostUsd;
+  return metrics;
+}
+function nonNegativeNumber2(value) {
+  return typeof value === "number" && Number.isFinite(value) && value >= 0 ? value : null;
 }
 function isRecord6(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
@@ -2771,7 +2917,7 @@ function createEvalCallLedgerFold(now = () => performance.now()) {
       activeFirstOutputAt = null;
       return;
     }
-    if (event.type === "message_update" && activeStartedAt !== null && activeFirstOutputAt === null) {
+    if (isAssistantOutput(event) && activeStartedAt !== null && activeFirstOutputAt === null) {
       activeFirstOutputAt = now();
       return;
     }
@@ -2785,24 +2931,28 @@ function createEvalCallLedgerFold(now = () => performance.now()) {
     }
     const endedAt = now();
     const promptCache = {
-      input: nonNegativeNumber2(usage.input) ?? 0,
-      cacheRead: nonNegativeNumber2(usage.cacheRead) ?? 0,
-      cacheWrite: nonNegativeNumber2(usage.cacheWrite) ?? 0,
+      input: nonNegativeNumber3(usage.input) ?? 0,
+      cacheRead: nonNegativeNumber3(usage.cacheRead) ?? 0,
+      cacheWrite: nonNegativeNumber3(usage.cacheWrite) ?? 0,
       backendVerdict: "unknown"
     };
     if (isRecord7(message.backendTimings)) promptCache.backend = structuredClone(message.backendTimings);
     const previous = entries.at(-1);
+    const timestamp = messageTimestamp(message.timestamp);
     entries.push({
       kind: "message",
       role: "assistant",
       turnId: `eval-call-${entries.length + 1}`,
       parentTurnId: previous?.turnId ?? null,
-      timestamp: messageTimestamp(message.timestamp),
+      timestamp: timestamp ?? (/* @__PURE__ */ new Date(0)).toISOString(),
       payload: {
+        // Retain the legacy ISO placeholder without treating it as observed
+        // chronology. Per-call monotonic timing remains usable on its own.
+        ...timestamp === null ? { timestampEstimated: true } : {},
         promptCache,
-        timing: {
-          ttftMs: activeStartedAt === null ? null : Math.round(Math.max(0, (activeFirstOutputAt ?? endedAt) - activeStartedAt)),
-          apiMs: activeStartedAt === null ? 0 : Math.round(Math.max(0, endedAt - activeStartedAt))
+        timing: activeStartedAt === null ? null : {
+          ttftMs: activeFirstOutputAt === null ? null : Math.round(Math.max(0, activeFirstOutputAt - activeStartedAt)),
+          apiMs: Math.round(Math.max(0, endedAt - activeStartedAt))
         },
         usage: structuredClone(usage)
       }
@@ -2829,16 +2979,22 @@ function createEvalCallLedgerFold(now = () => performance.now()) {
     }
   };
 }
+function isAssistantOutput(event) {
+  const output = event.type === "message_update" ? event.assistantMessageEvent : event;
+  if (!isRecord7(output)) return false;
+  if (output.type === "toolcall_start") return event.type === "message_update";
+  return (output.type === "text_delta" || output.type === "thinking_delta" || output.type === "toolcall_delta") && typeof output.delta === "string" && output.delta.length > 0;
+}
 function isAssistantMessage(value) {
   return isRecord7(value) && value.role === "assistant";
 }
 function messageTimestamp(value) {
-  const milliseconds = nonNegativeNumber2(value);
-  if (milliseconds === null) return (/* @__PURE__ */ new Date(0)).toISOString();
+  const milliseconds = nonNegativeNumber3(value);
+  if (milliseconds === null) return null;
   const date = new Date(milliseconds);
-  return Number.isFinite(date.getTime()) ? date.toISOString() : (/* @__PURE__ */ new Date(0)).toISOString();
+  return Number.isFinite(date.getTime()) ? date.toISOString() : null;
 }
-function nonNegativeNumber2(value) {
+function nonNegativeNumber3(value) {
   return typeof value === "number" && Number.isFinite(value) && value >= 0 ? value : null;
 }
 function parseRecord(line) {
@@ -3227,16 +3383,19 @@ function toolBehaviorMetricEntriesFromJsonl(stdout, cwd, readObservation) {
   }
   const terminals = canonicalFinishes.length > 0 ? canonicalFinishes : executionEnds;
   const calls = /* @__PURE__ */ new Map();
+  const succeeded = /* @__PURE__ */ new Map();
   const blocked = /* @__PURE__ */ new Map();
   for (const terminal of terminals) {
     const tool = metricToolName(terminal.tool);
     calls.set(tool, (calls.get(tool) ?? 0) + 1);
+    if (terminal.outcome === "ok") succeeded.set(tool, (succeeded.get(tool) ?? 0) + 1);
     if (terminal.outcome === "blocked") blocked.set(tool, (blocked.get(tool) ?? 0) + 1);
   }
-  const namedTools = /* @__PURE__ */ new Set(["bash", "dispatch", "read", ...calls.keys(), ...blocked.keys()]);
+  const namedTools = /* @__PURE__ */ new Set(["bash", "dispatch", "read", ...calls.keys(), ...succeeded.keys(), ...blocked.keys()]);
   const entries = { "tools.read.distinctPaths": readPaths.size };
   for (const tool of [...namedTools].sort()) {
     entries[`tools.calls.${tool}`] = calls.get(tool) ?? 0;
+    entries[`tools.succeeded.${tool}`] = succeeded.get(tool) ?? 0;
     entries[`tools.blocked.${tool}`] = blocked.get(tool) ?? 0;
   }
   if (readObservation !== void 0) {
@@ -3378,21 +3537,21 @@ async function runContextInitRunner(runner, cwd, clioEntry, timeoutMs, target, e
     wallTimeMs: result.wallTimeMs,
     metrics: {
       "latency.wallMs": result.wallTimeMs,
-      "latency.modelMs": numberField3(run, "durationMs"),
-      "tokens.input": numberField3(tokens, "input"),
-      "tokens.output": numberField3(tokens, "output"),
-      "tokens.total": numberField3(tokens, "total"),
-      "tokens.cacheRead": numberField3(tokens, "cacheRead"),
-      "tokens.cacheWrite": numberField3(tokens, "cacheWrite"),
-      "tools.totalCalls": numberField3(run, "toolCalls"),
-      "tools.failed": numberField3(run, "toolFailures"),
-      "tools.blocked": numberField3(run, "toolBlocked"),
+      "latency.modelMs": numberField2(run, "durationMs"),
+      "tokens.input": numberField2(tokens, "input"),
+      "tokens.output": numberField2(tokens, "output"),
+      "tokens.total": numberField2(tokens, "total"),
+      "tokens.cacheRead": numberField2(tokens, "cacheRead"),
+      "tokens.cacheWrite": numberField2(tokens, "cacheWrite"),
+      "tools.totalCalls": numberField2(run, "toolCalls"),
+      "tools.failed": numberField2(run, "toolFailures"),
+      "tools.blocked": numberField2(run, "toolBlocked"),
       "context.clioMdBytes": clioMdBytes,
       "context.initMode": stringField3(generation, "mode") ?? "unknown",
       "context.initParserOutcome": stringField3(generation, "parserOutcome") ?? "unknown",
       "context.initFallback": stringField3(generation, "fallbackReason") !== null,
-      "context.initPromptBytes": numberField3(run, "promptBytes"),
-      "context.initOutputBytes": numberField3(run, "outputBytes"),
+      "context.initPromptBytes": numberField2(run, "promptBytes"),
+      "context.initOutputBytes": numberField2(run, "outputBytes"),
       "context.initTargetId": effectiveTarget,
       "context.initModelId": effectiveModel,
       "context.initRuntimeId": effectiveRuntime,
@@ -3491,7 +3650,7 @@ function recordField2(value, field) {
   const selected = field && typeof value === "object" && value !== null && !Array.isArray(value) ? value[field] : value;
   return typeof selected === "object" && selected !== null && !Array.isArray(selected) ? selected : null;
 }
-function numberField3(value, field) {
+function numberField2(value, field) {
   const record = recordField2(value);
   const selected = record?.[field];
   return typeof selected === "number" && Number.isFinite(selected) ? selected : null;
@@ -3916,11 +4075,23 @@ async function runMatrixItem(loaded, task, target, repeatIndex, clioEntry, fresh
       }
     };
     const snapshot = await readEvalLedgerSnapshot(stateDir);
-    const ledgerEntries = [...snapshot.entries, ...runner.ledgerEntries ?? []];
+    const selected = selectEvalLedgerEntries(snapshot.entries, runner.ledgerEntries ?? []);
+    result.artifacts.trackedMetricSources = JSON.stringify({
+      assistantCalls: selected.source,
+      sessionCalls: selected.sessionCalls,
+      streamCalls: selected.streamCalls,
+      compactions: "session"
+    });
+    if (selected.sessionCalls > 0 && selected.streamCalls > 0) {
+      result.artifacts.callLedger ??= JSON.stringify(runner.ledgerEntries);
+      if (selected.sessionCalls !== selected.streamCalls || task.runner.kind === "external-command") {
+        result.artifacts.trackedMetricWarning = "Session and stream call coverage may differ; tracked metrics use session calls, not a reconciled union. Inspect callLedger before treating these counts as complete: no shared call identity proves overlap.";
+      }
+    }
     result.verdict = adaptSuiteV2ResultToVerdictV1(
       result,
       buildEvalTrackedMetrics({
-        ledgerEntries,
+        ledgerEntries: selected.entries,
         receipt: receipt ?? null,
         fallbackWallClockMs: runner.wallTimeMs
       })
@@ -4326,7 +4497,7 @@ function parseEvalArgs(args) {
 }
 async function runEvalCommand(args) {
   if (args[0] === "inventory") {
-    const { runEvalInventory } = await import("./eval-inventory-6DEJPLBF.js");
+    const { runEvalInventory } = await import("./eval-inventory-Y6QRFOH5.js");
     return runEvalInventory(args.slice(1));
   }
   let parsed;
@@ -4501,4 +4672,4 @@ function comparisonFormat(value) {
 export {
   runEvalCommand
 };
-//# sourceMappingURL=eval-C45FYRJ6.js.map
+//# sourceMappingURL=eval-7G7SGAYO.js.map

package/dist/{eval-inventory-6DEJPLBF.js → eval-inventory-Y6QRFOH5.js}

@@ -2,19 +2,19 @@ import { createRequire as __clioCreateRequire } from "node:module"; const requir
 import {
   CANONICAL_METRICS,
   CANONICAL_METRIC_PREFIXES
-} from "./chunk-RKSR6VSF.js";
+} from "./chunk-4IUZQIJ3.js";
 import {
   EVAL_EXECUTION_MATRIX_DIMENSIONS_V1,
   evalRoot,
   evalServingConfigurationOf,
   loadEvalArtifactV4,
   sameEvalServingConfiguration
-} from "./chunk-5PFYMY2V.js";
+} from "./chunk-FTMGRKEF.js";
 import "./chunk-R346GLFC.js";
+import "./chunk-WEPFGWHJ.js";
 import {
   clioDataDir
 } from "./chunk-BYP5D4HI.js";
-import "./chunk-WEPFGWHJ.js";
 import {
   init_esm_shims
 } from "./chunk-3R73A4XB.js";
@@ -312,4 +312,4 @@ export {
   EVAL_INVENTORY_MAX_SCENARIOS,
   runEvalInventory
 };
-//# sourceMappingURL=eval-inventory-6DEJPLBF.js.map
+//# sourceMappingURL=eval-inventory-Y6QRFOH5.js.map