fix(results): isolate target bundles under run timestamps

apps/cli/src/commands/eval/artifact-writer.ts

@@ -19,6 +19,8 @@ import {
   buildAggregateGradingArtifact,
   buildIndexArtifactEntry as buildCoreIndexArtifactEntry,
   buildResultIndexArtifact as buildCoreResultIndexArtifact,
+  buildEvalTestTargetKey,
+  buildEvaluationResultTargetKey,
   buildGradingArtifact,
   buildRunSummaryArtifact,
   buildTestTargetKey,
@@ -41,6 +43,8 @@ import {
 export {
   aggregateRunDir,
   buildAggregateGradingArtifact,
+  buildEvalTestTargetKey,
+  buildEvaluationResultTargetKey,
   buildRunSummaryArtifact,
   buildGradingArtifact,
   buildTestTargetKey,

apps/cli/src/commands/eval/progress-display.ts

@@ -65,8 +65,6 @@ export class ProgressDisplay {
   private readonly workers: Map<number, WorkerProgress> = new Map();
   private totalTests = 0;
   private completedTests = 0;
-  private readonly logPaths: string[] = [];
-  private readonly logPathSet = new Set<string>();
   private started = false;
   private finished = false;
   private readonly verbose: boolean;
@@ -133,24 +131,7 @@ export class ProgressDisplay {
   }
 
   addLogPaths(paths: readonly string[]): void {
-    const newPaths: string[] = [];
-    for (const path of paths) {
-      if (this.logPathSet.has(path)) {
-        continue;
-      }
-      this.logPathSet.add(path);
-      newPaths.push(path);
-    }
-
-    if (newPaths.length === 0) {
-      return;
-    }
-
-    this.logPaths.push(...newPaths);
-
-    for (const p of newPaths) {
-      console.log(`Provider log: ${p}`);
-    }
+    void paths;
   }
 
   finish(): void {

apps/cli/src/commands/eval/result-layout.ts

@@ -1,4 +1,4 @@
-import { existsSync, statSync } from 'node:fs';
+import { type Dirent, existsSync, readdirSync, statSync } from 'node:fs';
 import path from 'node:path';
 
 export const RESULT_INDEX_FILENAME = 'index.jsonl';
@@ -76,6 +76,37 @@ export function resolveExistingRunPrimaryPath(runDir: string): string | undefine
   return undefined;
 }
 
+export function discoverRunManifestPaths(runDir: string): readonly string[] {
+  const direct = resolveExistingRunPrimaryPath(runDir);
+  if (direct) {
+    return [direct];
+  }
+
+  const manifests: string[] = [];
+  function walk(currentDir: string): void {
+    const primary = resolveExistingRunPrimaryPath(currentDir);
+    if (primary) {
+      manifests.push(primary);
+      return;
+    }
+
+    let entries: Dirent<string>[];
+    try {
+      entries = readdirSync(currentDir, { withFileTypes: true });
+    } catch {
+      return;
+    }
+    for (const entry of entries) {
+      if (entry.isDirectory()) {
+        walk(path.join(currentDir, entry.name));
+      }
+    }
+  }
+
+  walk(runDir);
+  return manifests.sort();
+}
+
 export function isDirectoryPath(filePath: string): boolean {
   try {
     return statSync(filePath).isDirectory();
@@ -90,11 +121,20 @@ export function resolveWorkspaceOrFilePath(filePath: string): string {
   }
 
   const existing = resolveExistingRunPrimaryPath(filePath);
-  if (!existing) {
-    throw new Error(`Result workspace is missing ${RESULT_INDEX_FILENAME}: ${filePath}`);
+  if (existing) {
+    return existing;
   }
 
-  return existing;
+  const nested = discoverRunManifestPaths(filePath);
+  if (nested.length === 1) {
+    return nested[0];
+  }
+  if (nested.length > 1) {
+    throw new Error(
+      `Result workspace contains multiple ${RESULT_INDEX_FILENAME} manifests; pass one bundle directory or manifest: ${filePath}`,
+    );
+  }
+  throw new Error(`Result workspace is missing ${RESULT_INDEX_FILENAME}: ${filePath}`);
 }
 
 export function resolveRunManifestPath(filePath: string): string {

apps/cli/src/commands/eval/run-cache.ts

@@ -4,6 +4,7 @@ import path from 'node:path';
 
 import {
   RESULT_INDEX_FILENAME,
+  discoverRunManifestPaths,
   resolveExistingRunPrimaryPath,
   resolveRunIndexPath,
 } from './result-layout.js';
@@ -27,7 +28,11 @@ export interface RunCache {
  */
 export function resolveRunCacheFile(cache: RunCache): string {
   if (cache.lastRunDir) {
-    return resolveExistingRunPrimaryPath(cache.lastRunDir) ?? resolveRunIndexPath(cache.lastRunDir);
+    const direct = resolveExistingRunPrimaryPath(cache.lastRunDir);
+    if (direct) {
+      return direct;
+    }
+    return discoverRunManifestPaths(cache.lastRunDir)[0] ?? resolveRunIndexPath(cache.lastRunDir);
   }
   return '';
 }
@@ -61,14 +66,12 @@ export async function resolveCachedRunDir(cwd: string): Promise<string | undefin
 }
 
 export async function saveRunCache(cwd: string, resultPath: string): Promise<void> {
-  if (path.basename(resultPath) !== RESULT_INDEX_FILENAME) {
-    return;
-  }
-
   const dir = path.join(cwd, '.agentv');
+  const lastRunDir =
+    path.basename(resultPath) === RESULT_INDEX_FILENAME ? path.dirname(resultPath) : resultPath;
   await mkdir(dir, { recursive: true });
   const cache: RunCache = {
-    lastRunDir: path.dirname(resultPath),
+    lastRunDir,
     timestamp: new Date().toISOString(),
   };
   await writeFile(cachePath(cwd), `${JSON.stringify(cache, null, 2)}\n`, 'utf-8');