agentv 4.43.0-next.1 → 5.0.0-next.1

package/dist/{chunk-IFQXBQSY.js → chunk-6Q56CEXV.js}

@@ -9,6 +9,7 @@ import {
   buildDefaultRunDirFromName,
   c,
   clearRemoteRunTags,
+  confirmRemoteResultsMerge,
   createRunDirName,
   detectFileType,
   discoverEvalFiles,
@@ -64,12 +65,12 @@ import {
   validateTargetsFile,
   validateWorkspacePaths,
   writeRunTags
-} from "./chunk-ROLNQBYU.js";
+} from "./chunk-IBI7A5RP.js";
 import {
   materializeEvalBundle,
   toSnakeCaseDeep as toSnakeCaseDeep2,
   writeArtifactsFromResults as writeArtifactsFromResults2
-} from "./chunk-E2NL57MZ.js";
+} from "./chunk-6JVM4GJI.js";
 import {
   DEFAULT_CATEGORY,
   deriveCategory,
@@ -86,17 +87,16 @@ import {
   syncProjects,
   transpileEvalYamlFile,
   trimBaselineResult
-} from "./chunk-5NDSGGMJ.js";
+} from "./chunk-UG4JHZYM.js";
 import {
   AGENTV_RESULTS_ARTIFACTS_REF,
   DEFAULT_THRESHOLD,
   EXECUTION_TRACE_SCHEMA_VERSION,
   addProject,
   aggregateRunDir,
-  buildBenchmarkArtifact,
   buildResultIndexArtifact,
+  buildRunSummaryArtifact,
   buildTestTargetKey,
-  buildTimingArtifact,
   buildTraceEnvelopeFromEvaluationResult,
   buildTraceFromMessages,
   createBuiltinRegistry,
@@ -105,6 +105,7 @@ import {
   externalTraceMetadataFromRecord,
   external_exports,
   getAgentvConfigDir,
+  getLocalConfigPath,
   getProject,
   getWorkspacePoolRoot,
   gradePreparedEvalCase,
@@ -113,6 +114,7 @@ import {
   loadConfig,
   loadProjectRegistry,
   loadTestSuite,
+  mergeConfigObjects,
   normalizeLineEndings,
   normalizeResultRow,
   omitExternalTraceMetadataKeys,
@@ -145,7 +147,7 @@ import {
   traceEnvelopeToTranscriptMessages,
   traceFromTranscriptJsonLines,
   writeArtifactsFromResults
-} from "./chunk-RFJM2UBD.js";
+} from "./chunk-GZJMODLW.js";
 import {
   __commonJS,
   __require,
@@ -4258,7 +4260,7 @@ var doctorCommand = command({
 import path5 from "node:path";
 var evalAggregateCommand = command({
   name: "aggregate",
-  description: "Recompute benchmark.json and timing.json from a run directory. Deduplicates by (test_id, target), keeping the last entry.",
+  description: "Recompute summary.json from a run directory. Deduplicates by (test_id, target), keeping the last entry.",
   args: {
     runDir: positional({
       type: string,
@@ -4268,10 +4270,9 @@ var evalAggregateCommand = command({
   },
   handler: async (args) => {
     const runDir = path5.resolve(args.runDir);
-    const { benchmarkPath, timingPath, testCount, targetCount } = await aggregateRunDir(runDir);
+    const { summaryPath, testCount, targetCount } = await aggregateRunDir(runDir);
     console.log(`Aggregated ${testCount} test result(s) across ${targetCount} target(s)`);
-    console.log(`  Benchmark: ${benchmarkPath}`);
-    console.log(`  Timing:    ${timingPath}`);
+    console.log(`  Summary: ${summaryPath}`);
   }
 });
 
@@ -4629,7 +4630,7 @@ var evalRunCommand = command({
       type: optional(string),
       long: "output",
       short: "o",
-      description: "Run artifact directory (writes index.jsonl, benchmark.json, timing, and per-test artifacts)"
+      description: "Run artifact directory (writes index.jsonl, summary.json, and per-case artifacts)"
     }),
     outputFormat: option({
       type: optional(string),
@@ -4827,12 +4828,6 @@ var evalRunCommand = command({
     })
   },
   handler: async (args) => {
-    if (args.evalPaths.length === 0 && process.stdin.isTTY) {
-      const { launchInteractiveWizard } = await import("./interactive-RE4DXH4A.js");
-      await launchInteractiveWizard();
-      return;
-    }
-    const resolvedPaths = await resolveEvalPaths(args.evalPaths, process.cwd());
     if (args.budgetUsd !== void 0 && args.budgetUsd <= 0) {
       console.error("Error: --budget-usd must be a positive number.");
       process.exit(2);
@@ -4891,7 +4886,7 @@ var evalRunCommand = command({
       recordReplay: args.recordReplay,
       recordReplayVariant: args.recordReplayVariant
     };
-    const result = await runEvalCommand({ testFiles: resolvedPaths, rawOptions });
+    const result = await runEvalCommand({ testFiles: args.evalPaths, rawOptions });
     if (result?.allExecutionErrors) {
       process.exit(2);
     }
@@ -9089,7 +9084,6 @@ var evalBenchCommand = command({
       const grading = {
         assertions: allAssertions,
         summary: { passed, failed, total: allAssertions.length, pass_rate: passRate },
-        execution_metrics: { tool_calls: {}, total_tool_calls: 0, errors_encountered: 0 },
         graders: evaluators.map((e) => ({
           name: e.name,
           type: e.type,
@@ -9151,7 +9145,7 @@ var evalBenchCommand = command({
       "utf8"
     );
     const passRateStats = computeStats(allPassRates);
-    const benchmark = {
+    const summary = {
       metadata: {
         eval_file: manifest.eval_file,
         timestamp: manifest.timestamp,
@@ -9166,11 +9160,24 @@ var evalBenchCommand = command({
           tokens: { mean: 0, stddev: 0 }
         }
       },
+      timing: {
+        total_tokens: 0,
+        duration_ms: 0,
+        total_duration_seconds: 0,
+        cost_usd: null,
+        token_usage: { input: 0, output: 0, reasoning: 0 },
+        usage_sources: {
+          token_usage: "unavailable",
+          total_tokens: "unavailable",
+          duration: "unavailable",
+          cost: "unavailable"
+        }
+      },
       notes: []
     };
     await writeFile7(
-      join(exportDir, "benchmark.json"),
-      `${JSON.stringify(benchmark, null, 2)}
+      join(exportDir, "summary.json"),
+      `${JSON.stringify(summary, null, 2)}
 `,
       "utf8"
     );
@@ -10359,10 +10366,10 @@ function parseJsonlLine(line) {
 function readManifestRecords(manifestPath) {
   return readFileSync8(manifestPath, "utf8").split(/\r?\n/).map((line) => line.trim()).filter(Boolean).map(parseJsonlLine);
 }
-function readBenchmarkMetadata(manifestPath) {
+function readSummaryMetadata(manifestPath) {
   try {
-    const benchmarkPath = path19.join(path19.dirname(manifestPath), "benchmark.json");
-    const parsed = JSON.parse(readFileSync8(benchmarkPath, "utf8"));
+    const summaryPath = path19.join(path19.dirname(manifestPath), "summary.json");
+    const parsed = JSON.parse(readFileSync8(summaryPath, "utf8"));
     return {
       timestamp: parsed.metadata?.timestamp,
       displayName: parsed.metadata?.display_name
@@ -10388,7 +10395,7 @@ function loadSources(sources) {
     if (records.length !== results.length) {
       throw new Error(`Manifest could not be hydrated completely: ${manifestPath}`);
     }
-    const metadata = readBenchmarkMetadata(manifestPath);
+    const metadata = readSummaryMetadata(manifestPath);
     return {
       ...source,
       index,
@@ -10529,6 +10536,7 @@ function resolveCombinedExperiment(sources, requestedExperiment) {
 }
 var MANIFEST_PATH_FIELDS = [
   "artifact_dir",
+  "summary_path",
   "grading_path",
   "timing_path",
   "input_path",
@@ -10705,15 +10713,12 @@ function combineRunSources(options) {
   const records = rows.map((row) => rewriteAndCopyRecord(row, runDir, experiment));
   const manifestPath = path19.join(runDir, "index.jsonl");
   writeJsonl(manifestPath, records);
-  const timing = buildTimingArtifact(results);
-  const timingPath = path19.join(runDir, "timing.json");
-  writeJson3(timingPath, timing);
-  const benchmark = buildBenchmarkArtifact(results, "", "combined", results.length);
-  const benchmarkWithMetadata = {
-    ...benchmark,
+  const summary = buildRunSummaryArtifact(results, "", "combined", results.length);
+  const summaryWithMetadata = {
+    ...summary,
     metadata: {
-      ...benchmark.metadata,
-      timestamp: startedAt ?? benchmark.metadata.timestamp,
+      ...summary.metadata,
+      timestamp: startedAt ?? summary.metadata.timestamp,
       display_name: displayName,
       experiment,
       combined_from_run_ids: loadedSources.map((source) => source.id),
@@ -10721,15 +10726,14 @@ function combineRunSources(options) {
       duplicate_policy: options.duplicatePolicy
     }
   };
-  const benchmarkPath = path19.join(runDir, "benchmark.json");
-  writeJson3(benchmarkPath, benchmarkWithMetadata);
+  const summaryPath = path19.join(runDir, "summary.json");
+  writeJson3(summaryPath, summaryWithMetadata);
   const tags = [...new Set(loadedSources.flatMap((source) => source.tags ?? []))].sort();
   return {
     runDir,
     runId: toRunId(options.cwd, runDir),
     manifestPath,
-    benchmarkPath,
-    timingPath,
+    summaryPath,
     displayName,
     experiment,
     combinedFromRunIds: loadedSources.map((source) => source.id),
@@ -10819,7 +10823,7 @@ var resultsCombineCommand = command({
     displayName: option({
       type: optional(string),
       long: "display-name",
-      description: "Display name stored in benchmark.json metadata"
+      description: "Display name stored in summary.json metadata"
     }),
     duplicatePolicy: option({
       type: optional(oneOf(["prompt", "error", "latest"])),
@@ -10871,8 +10875,7 @@ var resultsCombineCommand = command({
       });
       console.log(`Combined ${result.testCount} result row(s) into ${result.runDir}`);
       console.log(`  Run ID: ${result.runId}`);
-      console.log(`  Benchmark: ${result.benchmarkPath}`);
-      console.log(`  Timing:    ${result.timingPath}`);
+      console.log(`  Summary: ${result.summaryPath}`);
       if (result.duplicateConflicts.length > 0) {
         console.log(`  Duplicates handled: ${result.duplicateConflicts.length}`);
       }
@@ -11028,11 +11031,11 @@ function artifactRefs(indexEntry, options) {
   return dropUndefined({
     ...metadataRefs,
     artifact_dir: indexEntry.artifact_dir,
+    summary_path: indexEntry.summary_path,
     grading_path: indexEntry.grading_path,
     input_path: indexEntry.input_path,
     output_path: indexEntry.output_path,
     answer_path: indexEntry.answer_path,
-    response_path: indexEntry.response_path,
     transcript_path: indexEntry.transcript_path,
     metrics_path: indexEntry.metrics_path,
     trace_path: tracePathFor(indexEntry),
@@ -11109,8 +11112,7 @@ function buildEntry(result, options, indexRecord) {
     source: { kind: "agentv_run", path: sourcePath, format: "agentv_result" },
     artifacts: {
       trace_path: tracePathFor(indexRecord ?? plannedIndexEntry),
-      answer_path: result.output.length > 0 ? "outputs/answer.md" : void 0,
-      response_path: result.output.length > 0 ? "outputs/response.md" : void 0
+      answer_path: result.output.length > 0 ? "outputs/answer.md" : void 0
     },
     duplicatePolicy: options.duplicatePolicy,
     capture: captureOptions(includeRawContent),
@@ -13092,14 +13094,14 @@ function normalizeEvalFileLabel(value) {
   }
   return path23.basename(trimmed).replace(/\.results\.jsonl$/i, "").replace(/\.eval\.ya?ml$/i, "").replace(/\.ya?ml$/i, "").replace(/\.jsonl$/i, "");
 }
-function readBenchmarkEvalFile(sourceFile) {
-  const benchmarkPath = path23.join(path23.dirname(sourceFile), "benchmark.json");
-  if (!existsSync11(benchmarkPath)) {
+function readSummaryEvalFile(sourceFile) {
+  const summaryPath = path23.join(path23.dirname(sourceFile), "summary.json");
+  if (!existsSync11(summaryPath)) {
     return void 0;
   }
   try {
-    const benchmark = JSON.parse(readFileSync10(benchmarkPath, "utf8"));
-    return normalizeEvalFileLabel(benchmark.metadata?.eval_file);
+    const summary = JSON.parse(readFileSync10(summaryPath, "utf8"));
+    return normalizeEvalFileLabel(summary.metadata?.eval_file);
   } catch {
     return void 0;
   }
@@ -13107,8 +13109,8 @@ function readBenchmarkEvalFile(sourceFile) {
 function deriveReportPath(sourceFile) {
   return path23.join(path23.dirname(sourceFile), "report.html");
 }
-function serializeReportResult(result, sourceFile, manifestRecord, benchmarkEvalFile) {
-  const fallbackEvalFile = normalizeEvalFileLabel(manifestRecord?.eval_file) ?? benchmarkEvalFile ?? normalizeEvalFileLabel(result.suite) ?? path23.basename(path23.dirname(sourceFile));
+function serializeReportResult(result, sourceFile, manifestRecord, summaryEvalFile) {
+  const fallbackEvalFile = normalizeEvalFileLabel(manifestRecord?.eval_file) ?? summaryEvalFile ?? normalizeEvalFileLabel(result.suite) ?? path23.basename(path23.dirname(sourceFile));
   return {
     timestamp: result.timestamp,
     test_id: result.testId,
@@ -13141,23 +13143,23 @@ async function loadReportSource(source, cwd) {
     sourceFile: resolvedSourceFile,
     results,
     records,
-    benchmarkEvalFile: readBenchmarkEvalFile(resolvedSourceFile)
+    summaryEvalFile: readSummaryEvalFile(resolvedSourceFile)
   };
 }
-function renderResultsReport(results, sourceFile, records, benchmarkEvalFile) {
+function renderResultsReport(results, sourceFile, records, summaryEvalFile) {
   if (!RESULTS_REPORT_TEMPLATE.includes("__DATA_PLACEHOLDER__")) {
     throw new Error("Report template is missing __DATA_PLACEHOLDER__");
   }
   const rows = results.map(
-    (result, index) => serializeReportResult(result, sourceFile, records[index], benchmarkEvalFile)
+    (result, index) => serializeReportResult(result, sourceFile, records[index], summaryEvalFile)
   );
   const dataJson = JSON.stringify(rows).replace(/<\//g, "<\\/");
   return RESULTS_REPORT_TEMPLATE.replace("__DATA_PLACEHOLDER__", () => dataJson);
 }
 async function writeResultsReport(source, outputPath, cwd) {
-  const { sourceFile, results, records, benchmarkEvalFile } = await loadReportSource(source, cwd);
+  const { sourceFile, results, records, summaryEvalFile } = await loadReportSource(source, cwd);
   const resolvedOutputPath = outputPath ? path23.isAbsolute(outputPath) ? outputPath : path23.resolve(cwd, outputPath) : deriveReportPath(sourceFile);
-  const html = renderResultsReport(results, sourceFile, records, benchmarkEvalFile);
+  const html = renderResultsReport(results, sourceFile, records, summaryEvalFile);
   mkdirSync3(path23.dirname(resolvedOutputPath), { recursive: true });
   writeFileSync4(resolvedOutputPath, html, "utf8");
   const written = readFileSync10(resolvedOutputPath, "utf8");
@@ -13423,10 +13425,10 @@ function checkIndexJsonl(runDir) {
           message: `index.jsonl line ${i + 1} (${entry.test_id ?? "?"}): missing 'target'`
         });
       }
-      if (!entry.grading_path) {
+      if (!entry.summary_path) {
         diagnostics.push({
-          severity: "warning",
-          message: `index.jsonl line ${i + 1} (${entry.test_id ?? "?"}): missing 'grading_path'`
+          severity: "error",
+          message: `index.jsonl line ${i + 1} (${entry.test_id ?? "?"}): missing 'summary_path'`
         });
       }
       if (!entry.scores || !Array.isArray(entry.scores) || entry.scores.length === 0) {
@@ -13481,6 +13483,35 @@ function checkArtifactFiles(runDir, entries2) {
   const diagnostics = [];
   for (const entry of entries2) {
     const testId = entry.test_id ?? "?";
+    if (entry.summary_path) {
+      const summaryPath2 = path24.join(runDir, entry.summary_path);
+      if (!existsSync13(summaryPath2)) {
+        diagnostics.push({
+          severity: "error",
+          message: `${testId}: summary.json not found at '${entry.summary_path}'`
+        });
+      }
+    }
+    for (const trial of entry.trials ?? []) {
+      if (!entry.artifact_dir || !trial.run_path) {
+        continue;
+      }
+      const runDirPath = path24.join(runDir, entry.artifact_dir, trial.run_path);
+      const resultPath = path24.join(runDirPath, "result.json");
+      const gradingPath = path24.join(runDirPath, "grading.json");
+      if (!existsSync13(resultPath)) {
+        diagnostics.push({
+          severity: "error",
+          message: `${testId}: result.json not found at '${path24.posix.join(entry.artifact_dir, trial.run_path, "result.json")}'`
+        });
+      }
+      if (!existsSync13(gradingPath)) {
+        diagnostics.push({
+          severity: "error",
+          message: `${testId}: grading.json not found at '${path24.posix.join(entry.artifact_dir, trial.run_path, "grading.json")}'`
+        });
+      }
+    }
     if (entry.grading_path) {
       const gradingPath = path24.join(runDir, entry.grading_path);
       if (!existsSync13(gradingPath)) {
@@ -13521,9 +13552,9 @@ function checkArtifactFiles(runDir, entries2) {
       }
     }
   }
-  const benchmarkPath = path24.join(runDir, "benchmark.json");
-  if (!existsSync13(benchmarkPath)) {
-    diagnostics.push({ severity: "warning", message: "benchmark.json is missing" });
+  const summaryPath = path24.join(runDir, "summary.json");
+  if (!existsSync13(summaryPath)) {
+    diagnostics.push({ severity: "error", message: "summary.json is missing" });
   }
   return diagnostics;
 }
@@ -14167,8 +14198,8 @@ var DEFAULTS = {
 function loadStudioConfig(agentvDir) {
   const localConfigPath = path26.join(agentvDir, "config.yaml");
   const globalConfigPath = path26.join(getAgentvConfigDir(), "config.yaml");
-  const localConfig = loadParsedConfig(localConfigPath);
-  const globalConfig = path26.resolve(globalConfigPath) === path26.resolve(localConfigPath) ? void 0 : loadParsedConfig(globalConfigPath);
+  const localConfig = loadParsedConfigPair(localConfigPath);
+  const globalConfig = path26.resolve(path26.dirname(globalConfigPath)) === path26.resolve(path26.dirname(localConfigPath)) ? void 0 : loadParsedConfigPair(globalConfigPath);
   const threshold = [
     readThreshold(localConfig?.dashboard),
     readThreshold(localConfig?.studio),
@@ -14190,6 +14221,12 @@ function loadParsedConfig(configPath) {
   if (!parsed || typeof parsed !== "object" || Array.isArray(parsed)) return void 0;
   return parsed;
 }
+function loadParsedConfigPair(configPath) {
+  const base = loadParsedConfig(configPath);
+  const local = loadParsedConfig(getLocalConfigPath(configPath));
+  if (base && local) return mergeConfigObjects(base, local);
+  return local ?? base;
+}
 function readThreshold(section) {
   if (!section || typeof section !== "object" || Array.isArray(section)) return void 0;
   const values = section;
@@ -14350,21 +14387,25 @@ function flattenFileTree(nodes) {
   }
   return files;
 }
-function ensureCatalogFileNode(root, entry) {
-  const segments = entry.displayPath.split("/").filter(Boolean);
+function ensureCatalogFileNode(root, entry, rootPrefix) {
+  const fullPath = entry.displayPath;
+  const nestPath = rootPrefix && fullPath.startsWith(`${rootPrefix}/`) ? fullPath.slice(rootPrefix.length + 1) : fullPath;
+  const usePrefix = nestPath !== fullPath ? rootPrefix : void 0;
+  const segments = nestPath.split("/").filter(Boolean);
   if (segments.length === 0) return;
   let siblings = root;
-  let currentPath = "";
+  let nestedPath = "";
   for (let i = 0; i < segments.length; i += 1) {
     const segment = segments[i];
     if (!segment) continue;
-    currentPath = currentPath ? `${currentPath}/${segment}` : segment;
+    nestedPath = nestedPath ? `${nestedPath}/${segment}` : segment;
+    const fullNodePath = usePrefix ? `${usePrefix}/${nestedPath}` : nestedPath;
     const isFile = i === segments.length - 1;
     let node = siblings.find((candidate) => candidate.name === segment);
     if (!node) {
       node = {
         name: segment,
-        path: currentPath,
+        path: fullNodePath,
         type: isFile ? "file" : "dir",
         ...isFile ? {
           kind: entry.kind,
@@ -14394,6 +14435,15 @@ function ensureCatalogFileNode(root, entry) {
     siblings = node.children;
   }
 }
+function overlayCatalogFileNodes(files, catalog, rootPrefix) {
+  const localFilePaths = new Set(flattenFileTree(files).map((file) => file.path));
+  for (const entry of catalog) {
+    if (entry.storage === "git" || !localFilePaths.has(entry.displayPath)) {
+      ensureCatalogFileNode(files, entry, rootPrefix);
+    }
+  }
+  return files;
+}
 function inferLanguage(filePath) {
   const ext = path27.extname(filePath).toLowerCase();
   const langMap = {
@@ -14564,6 +14614,28 @@ function normalizeArtifactRelativePath(relativePath) {
   }
   return segments.join("/");
 }
+function requestedArtifactDir(c4) {
+  const raw = c4.req.query("artifact_dir")?.trim();
+  if (!raw) {
+    return {};
+  }
+  const normalized = normalizeArtifactRelativePath(raw);
+  if (!normalized) {
+    return { error: "Invalid artifact_dir" };
+  }
+  return { value: normalized };
+}
+function manifestRecordSelection(records, evalId, artifactDir) {
+  return records.map((record, index) => ({ record, index })).find(({ record }) => {
+    if (record.test_id !== evalId) {
+      return false;
+    }
+    if (!artifactDir) {
+      return true;
+    }
+    return normalizeArtifactRelativePath(record.artifact_dir ?? "") === artifactDir;
+  });
+}
 function relativeRunPathFromNormalizedManifestPath(manifestPath) {
   const parts = manifestPath.split("/").filter(Boolean);
   const runsIndex = parts.lastIndexOf("runs");
@@ -14656,6 +14728,39 @@ function displayPathFromArtifactKey(key, runPath) {
   if (!normalizedKey.startsWith(runPrefix)) return normalizedKey;
   return normalizeArtifactRelativePath(normalizedKey.slice(runPrefix.length)) ?? normalizedKey;
 }
+function addTrialRunCatalogEntries(entries2, seen, record) {
+  const artifactDir = record.artifact_dir ? normalizeArtifactRelativePath(record.artifact_dir) : void 0;
+  if (!artifactDir) return;
+  for (const trial of record.trials ?? []) {
+    const runPath = trial.run_path ? normalizeArtifactRelativePath(trial.run_path) : void 0;
+    if (!runPath) continue;
+    const runDir = path27.posix.join(artifactDir, runPath);
+    addDirectArtifactCatalogEntry(
+      entries2,
+      seen,
+      path27.posix.join(runDir, "result.json"),
+      "artifact"
+    );
+    addDirectArtifactCatalogEntry(
+      entries2,
+      seen,
+      path27.posix.join(runDir, "grading.json"),
+      "artifact"
+    );
+    addDirectArtifactCatalogEntry(
+      entries2,
+      seen,
+      path27.posix.join(runDir, "metrics.json"),
+      "artifact"
+    );
+    addDirectArtifactCatalogEntry(
+      entries2,
+      seen,
+      path27.posix.join(runDir, "timing.json"),
+      "artifact"
+    );
+  }
+}
 function buildResultArtifactCatalog(record, options) {
   const entries2 = [];
   const seen = /* @__PURE__ */ new Set();
@@ -14666,6 +14771,7 @@ function buildResultArtifactCatalog(record, options) {
   addPointerArtifactCatalogEntry(entries2, seen, transcript, "transcript", options?.runPath);
   addPointerArtifactCatalogEntry(entries2, seen, trace, "trace", options?.runPath);
   addPointerArtifactCatalogEntry(entries2, seen, answer, "answer", options?.runPath);
+  addDirectArtifactCatalogEntry(entries2, seen, record.summary_path, "artifact");
   addDirectArtifactCatalogEntry(entries2, seen, record.grading_path, "artifact");
   addDirectArtifactCatalogEntry(entries2, seen, record.timing_path, "artifact");
   addDirectArtifactCatalogEntry(entries2, seen, record.input_path, "artifact");
@@ -14676,6 +14782,7 @@ function buildResultArtifactCatalog(record, options) {
   addDirectArtifactCatalogEntry(entries2, seen, recordWithTrace.trace_path, "trace");
   addDirectArtifactCatalogEntry(entries2, seen, record.eval_path, "artifact");
   addDirectArtifactCatalogEntry(entries2, seen, record.targets_path, "artifact");
+  addTrialRunCatalogEntries(entries2, seen, record);
   return entries2;
 }
 function resultArtifactTreeRootPaths(record, catalog) {
@@ -14686,16 +14793,25 @@ function resultArtifactTreeRootPaths(record, catalog) {
     record.graders_path
   ].filter((p, index, all) => !!p && all.indexOf(p) === index);
 }
-function buildLocalResultArtifactTree(baseDir, record, catalog) {
+function artifactTreeCommonDir(record, catalog) {
   const knownPaths = resultArtifactTreeRootPaths(record, catalog);
-  if (knownPaths.length === 0) return [];
+  if (knownPaths.length === 0) return void 0;
   const artifactDirs = knownPaths.map((p) => path27.dirname(p));
   let commonDir = artifactDirs[0];
   for (const dir of artifactDirs) {
     while (!dir.startsWith(commonDir)) {
-      commonDir = path27.dirname(commonDir);
+      const parent = path27.dirname(commonDir);
+      if (parent === commonDir) break;
+      commonDir = parent;
     }
   }
+  return commonDir === "." || commonDir === "" ? void 0 : commonDir;
+}
+function buildLocalResultArtifactTree(baseDir, record, catalog) {
+  const commonDir = artifactTreeCommonDir(record, catalog);
+  if (commonDir === void 0) {
+    return resultArtifactTreeRootPaths(record, catalog).length === 0 ? [] : buildFileTree(baseDir, baseDir);
+  }
   return buildFileTree(path27.join(baseDir, commonDir), baseDir);
 }
 function catalogEntryForDiscoveredLocalFile(nodes, displayPath) {
@@ -14804,6 +14920,87 @@ function stripHeavyFields(results) {
     };
   });
 }
+function readArtifactJsonObject(baseDir, relativePath) {
+  if (!relativePath) return void 0;
+  const resolved = resolveReadableRunArtifactFile(baseDir, relativePath);
+  if (!resolved.absolutePath) return void 0;
+  try {
+    const parsed = JSON.parse(readFileSync14(resolved.absolutePath, "utf8"));
+    return parsed && typeof parsed === "object" && !Array.isArray(parsed) ? parsed : void 0;
+  } catch {
+    return void 0;
+  }
+}
+function numberField(record, key) {
+  const value = record?.[key];
+  return typeof value === "number" && Number.isFinite(value) ? value : void 0;
+}
+function objectField(record, key) {
+  const value = record?.[key];
+  return value && typeof value === "object" && !Array.isArray(value) ? value : void 0;
+}
+function caseTrialArtifactPath(artifactDir, runPath, filePath) {
+  if (!artifactDir || !runPath) return void 0;
+  return path27.posix.join(artifactDir, runPath, filePath);
+}
+function buildRepeatTrialReadModels(baseDir, record) {
+  if (!record.trials || record.trials.length === 0) return void 0;
+  const artifactDir = record.artifact_dir ? normalizeArtifactRelativePath(record.artifact_dir) : void 0;
+  return record.trials.map((trial) => {
+    const runPath = trial.run_path ? normalizeArtifactRelativePath(trial.run_path) : void 0;
+    const metricsPath = caseTrialArtifactPath(artifactDir, runPath, "metrics.json");
+    const timingPath = caseTrialArtifactPath(artifactDir, runPath, "timing.json");
+    const gradingPath = caseTrialArtifactPath(artifactDir, runPath, "grading.json");
+    const transcriptPath = caseTrialArtifactPath(artifactDir, runPath, "transcript-raw.jsonl");
+    const answerPath = caseTrialArtifactPath(artifactDir, runPath, "outputs/answer.md");
+    const metrics = readArtifactJsonObject(baseDir, metricsPath);
+    const timing = readArtifactJsonObject(baseDir, timingPath);
+    const toolCalls = objectField(metrics, "tool_calls");
+    const tokenUsage = objectField(timing, "token_usage");
+    return {
+      ...trial,
+      ...numberField(timing, "duration_ms") !== void 0 && {
+        duration_ms: numberField(timing, "duration_ms")
+      },
+      ...numberField(timing, "total_tokens") !== void 0 && {
+        total_tokens: numberField(timing, "total_tokens")
+      },
+      ...numberField(timing, "cost_usd") !== void 0 && {
+        cost_usd: numberField(timing, "cost_usd")
+      },
+      ...tokenUsage && { token_usage: tokenUsage },
+      ...numberField(metrics, "total_tool_calls") !== void 0 && {
+        total_tool_calls: numberField(metrics, "total_tool_calls")
+      },
+      ...toolCalls && { tool_calls: toolCalls },
+      ...metricsPath && { metrics_path: metricsPath },
+      ...timingPath && { timing_path: timingPath },
+      ...gradingPath && { grading_path: gradingPath },
+      ...transcriptPath && { transcript_path: transcriptPath },
+      ...answerPath && { answer_path: answerPath }
+    };
+  });
+}
+function attachRunDetailReadModelFields(results, records, baseDir) {
+  return results.map((result, index) => {
+    const record = records[index];
+    if (!record) return result;
+    const trials = buildRepeatTrialReadModels(baseDir, record);
+    return {
+      ...result,
+      ...record.aggregation && { aggregation: record.aggregation },
+      ...record.artifact_dir && { artifact_dir: record.artifact_dir },
+      ...record.summary_path && { summary_path: record.summary_path },
+      ...record.grading_path && { grading_path: record.grading_path },
+      ...record.timing_path && { timing_path: record.timing_path },
+      ...record.metrics_path && { metrics_path: record.metrics_path },
+      ...record.transcript_path && { transcript_path: record.transcript_path },
+      ...record.output_path && { output_path: record.output_path },
+      ...record.answer_path && { answer_path: record.answer_path },
+      ...trials && { trials }
+    };
+  });
+}
 function inferExperimentFromRunId(runId) {
   const separatorIndex = runId.lastIndexOf("::");
   if (separatorIndex === -1) {
@@ -15104,8 +15301,12 @@ async function handleRunDetail(c4, { searchDir, projectId }) {
     const resumeMeta = meta.source === "local" ? deriveResumeMeta(searchDir, meta.path) : {};
     const liveStatus = meta.source === "local" ? getActiveRunStatus(meta.path) : void 0;
     const tagFields = await readRunTagFields(searchDir, meta, projectId);
+    const baseDir = path27.dirname(meta.path);
     return c4.json({
-      results: attachExternalTraceFields(stripHeavyFields(loaded), records),
+      results: attachExternalTraceFields(
+        attachRunDetailReadModelFields(stripHeavyFields(loaded), records, baseDir),
+        records
+      ),
       source: meta.source,
       source_label: meta.displayName,
       ...tagFields,
@@ -15146,9 +15347,9 @@ function deriveResumeMeta(cwd, manifestPath) {
   const relative3 = path27.relative(cwd, runDir);
   out.run_dir = relative3 !== "" && !relative3.startsWith("..") ? relative3 : runDir;
   try {
-    const benchmarkPath = path27.join(runDir, "benchmark.json");
-    if (existsSync16(benchmarkPath)) {
-      const parsed = JSON.parse(readFileSync14(benchmarkPath, "utf8"));
+    const summaryPath = path27.join(runDir, "summary.json");
+    if (existsSync16(summaryPath)) {
+      const parsed = JSON.parse(readFileSync14(summaryPath, "utf8"));
       const evalFile = parsed.metadata?.eval_file;
       if (typeof evalFile === "string" && evalFile.trim()) {
         out.suite_filter = evalFile.trim();
@@ -15261,14 +15462,24 @@ async function handleCategorySuites(c4, { searchDir, agentvDir, projectId }) {
 }
 async function handleEvalDetail(c4, { searchDir, projectId }) {
   const filename = c4.req.param("filename") ?? "";
-  const evalId = c4.req.param("evalId");
+  const evalId = c4.req.param("evalId") ?? "";
+  if (!evalId) return c4.json({ error: "Eval id is required" }, 400);
+  const artifactDir = requestedArtifactDir(c4);
+  if (artifactDir.error) return c4.json({ error: artifactDir.error }, 400);
   const meta = await findRunById(searchDir, filename, projectId);
   if (!meta) return c4.json({ error: "Run not found" }, 404);
   try {
     const loaded = await loadManifestResultsForMeta(searchDir, meta, projectId);
-    const result = loaded.find((r) => r.testId === evalId);
-    if (!result) return c4.json({ error: "Eval not found" }, 404);
-    const [stripped] = stripHeavyFields([result]);
+    const records = await parseManifestForMeta(searchDir, meta, projectId);
+    const selection = manifestRecordSelection(records, evalId, artifactDir.value);
+    const result = selection ? loaded[selection.index] : void 0;
+    if (!selection || !result) return c4.json({ error: "Eval not found" }, 404);
+    const baseDir = path27.dirname(meta.path);
+    const [stripped] = attachRunDetailReadModelFields(
+      stripHeavyFields([result]),
+      [selection.record],
+      baseDir
+    );
     return c4.json({ eval: stripped });
   } catch {
     return c4.json({ error: "Failed to load eval" }, 500);
@@ -15276,24 +15487,24 @@ async function handleEvalDetail(c4, { searchDir, projectId }) {
 }
 async function handleEvalFiles(c4, { searchDir, projectId }) {
   const filename = c4.req.param("filename") ?? "";
-  const evalId = c4.req.param("evalId");
+  const evalId = c4.req.param("evalId") ?? "";
+  if (!evalId) return c4.json({ error: "Eval id is required" }, 400);
+  const artifactDir = requestedArtifactDir(c4);
+  if (artifactDir.error) return c4.json({ error: artifactDir.error }, 400);
   const meta = await findRunById(searchDir, filename, projectId);
   if (!meta) return c4.json({ error: "Run not found" }, 404);
   try {
     const records = await parseManifestForMeta(searchDir, meta, projectId);
-    const record = records.find((r) => r.test_id === evalId);
-    if (!record) return c4.json({ error: "Eval not found" }, 404);
+    const selection = manifestRecordSelection(records, evalId, artifactDir.value);
+    if (!selection) return c4.json({ error: "Eval not found" }, 404);
+    const { record } = selection;
     const baseDir = path27.dirname(meta.path);
     const catalog = buildResultArtifactCatalog(record, {
       runPath: relativeRunPathFromManifestPath(meta.path)
     });
     const files = buildLocalResultArtifactTree(baseDir, record, catalog);
-    const localFilePaths = new Set(flattenFileTree(files).map((file) => file.path));
-    for (const entry of catalog) {
-      if (entry.storage === "git" || !localFilePaths.has(entry.displayPath)) {
-        ensureCatalogFileNode(files, entry);
-      }
-    }
+    const rootPrefix = artifactTreeCommonDir(record, catalog);
+    overlayCatalogFileNodes(files, catalog, rootPrefix);
     return c4.json({ files });
   } catch {
     return c4.json({ error: "Failed to load file tree" }, 500);
@@ -15301,7 +15512,10 @@ async function handleEvalFiles(c4, { searchDir, projectId }) {
 }
 async function handleEvalFileContent(c4, { searchDir, projectId }) {
   const filename = c4.req.param("filename") ?? "";
-  const evalId = c4.req.param("evalId");
+  const evalId = c4.req.param("evalId") ?? "";
+  if (!evalId) return c4.json({ error: "Eval id is required" }, 400);
+  const artifactDir = requestedArtifactDir(c4);
+  if (artifactDir.error) return c4.json({ error: artifactDir.error }, 400);
   const meta = await findRunById(searchDir, filename, projectId);
   if (!meta) return c4.json({ error: "Run not found" }, 404);
   const marker = "/files/";
@@ -15316,8 +15530,9 @@ async function handleEvalFileContent(c4, { searchDir, projectId }) {
   if (!filePath) return c4.json({ error: "No file path specified" }, 400);
   await ensureRunReadable(searchDir, meta, projectId);
   const records = parseResultManifest(readFileSync14(meta.path, "utf8"));
-  const record = records.find((r) => r.test_id === evalId);
-  if (!record) return c4.json({ error: "Eval not found" }, 404);
+  const selection = manifestRecordSelection(records, evalId, artifactDir.value);
+  if (!selection) return c4.json({ error: "Eval not found" }, 404);
+  const { record } = selection;
   const catalog = buildResultArtifactCatalog(record, {
     runPath: relativeRunPathFromManifestPath(meta.path)
   });
@@ -15337,13 +15552,17 @@ async function handleEvalFileContent(c4, { searchDir, projectId }) {
 }
 async function handleEvalTraceSession(c4, { searchDir, projectId }) {
   const filename = c4.req.param("filename") ?? "";
-  const evalId = c4.req.param("evalId");
+  const evalId = c4.req.param("evalId") ?? "";
+  if (!evalId) return c4.json({ error: "Eval id is required" }, 400);
+  const artifactDir = requestedArtifactDir(c4);
+  if (artifactDir.error) return c4.json({ error: artifactDir.error }, 400);
   const meta = await findRunById(searchDir, filename, projectId);
   if (!meta) return c4.json({ error: "Run not found" }, 404);
   try {
     const records = await parseManifestForMeta(searchDir, meta, projectId);
-    const record = records.find((r) => r.test_id === evalId);
-    if (!record) return c4.json({ error: "Eval not found" }, 404);
+    const selection = manifestRecordSelection(records, evalId, artifactDir.value);
+    if (!selection) return c4.json({ error: "Eval not found" }, 404);
+    const { record } = selection;
     const trace = resolveRecordArtifactPointer(record, "trace");
     const runPath = relativeRunPathFromManifestPath(meta.path);
     const traceEntry = findPointerArtifactCatalogEntry(
@@ -15453,13 +15672,17 @@ async function handleEvalTraceSession(c4, { searchDir, projectId }) {
 }
 async function handleEvalTranscript(c4, { searchDir, projectId }) {
   const filename = c4.req.param("filename") ?? "";
-  const evalId = c4.req.param("evalId");
+  const evalId = c4.req.param("evalId") ?? "";
+  if (!evalId) return c4.json({ error: "Eval id is required" }, 400);
+  const artifactDir = requestedArtifactDir(c4);
+  if (artifactDir.error) return c4.json({ error: artifactDir.error }, 400);
   const meta = await findRunById(searchDir, filename, projectId);
   if (!meta) return c4.json({ error: "Run not found" }, 404);
   try {
     const records = await parseManifestForMeta(searchDir, meta, projectId);
-    const record = records.find((r) => r.test_id === evalId);
-    if (!record) return c4.json({ error: "Eval not found" }, 404);
+    const selection = manifestRecordSelection(records, evalId, artifactDir.value);
+    if (!selection) return c4.json({ error: "Eval not found" }, 404);
+    const { record } = selection;
     const transcript = resolveRecordArtifactPointer(record, "transcript");
     const answer = resolveRecordArtifactPointer(record, "answer");
     const runPath = relativeRunPathFromManifestPath(meta.path);
@@ -16328,6 +16551,10 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
     "/api/remote/sync",
     async (c4) => c4.json(await syncRemoteResults(searchDir, defaultCtx.projectId))
   );
+  app2.post(
+    "/api/remote/sync/confirm-merge",
+    async (c4) => c4.json(await confirmRemoteResultsMerge(searchDir, defaultCtx.projectId))
+  );
   app2.get("/api/runs", (c4) => handleRuns(c4, defaultCtx));
   app2.post("/api/runs/combine", (c4) => {
     if (readOnly) {
@@ -16447,6 +16674,13 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
       async (ctx, dataCtx) => ctx.json(await syncRemoteResults(dataCtx.searchDir, dataCtx.projectId))
     )
   );
+  app2.post(
+    "/api/projects/:projectId/remote/sync/confirm-merge",
+    (c4) => withProject(
+      c4,
+      async (ctx, dataCtx) => ctx.json(await confirmRemoteResultsMerge(dataCtx.searchDir, dataCtx.projectId))
+    )
+  );
   app2.get("/api/projects/:projectId/runs", (c4) => withProject(c4, handleRuns));
   app2.post("/api/projects/:projectId/runs/combine", (c4) => {
     if (readOnly) {
@@ -18765,4 +18999,4 @@ export {
   preprocessArgv,
   runCli
 };
-//# sourceMappingURL=chunk-IFQXBQSY.js.map
+//# sourceMappingURL=chunk-6Q56CEXV.js.map