From 5cce778d8ca12d28cdc2d542c78afdb4fe78ab87 Mon Sep 17 00:00:00 2001
From: Christopher Tso <christso@gmail.com>
Date: Tue, 30 Jun 2026 06:51:51 +0200
Subject: [PATCH 1/2] fix(artifacts): rename generated task bundle to test
 bundle

---
 CONCEPTS.md                                   |  2 +-
 apps/cli/src/commands/eval/artifact-writer.ts | 14 ++--
 apps/cli/src/commands/eval/task-bundle.ts     | 22 ++---
 apps/cli/src/commands/results/combine-run.ts  |  1 +
 apps/cli/src/commands/results/manifest.ts     |  1 +
 .../src/commands/results/projection-bundle.ts |  2 +
 apps/cli/src/commands/results/serve.ts        |  2 +
 apps/cli/src/commands/runs/rerun.ts           | 37 +++++----
 .../commands/eval/artifact-writer.test.ts     | 50 ++++++-----
 .../test/commands/eval/task-bundle.test.ts    | 19 +++--
 apps/cli/test/commands/results/serve.test.ts  | 26 +++++-
 .../test/commands/results/validate.test.ts    | 63 ++++++++++++++
 apps/cli/test/commands/runs/rerun.test.ts     | 83 ++++++++++++++-----
 .../docs/docs/evaluation/running-evals.mdx    | 10 +--
 .../docs/docs/guides/benchmark-provenance.mdx |  4 +-
 .../docs/docs/reference/result-artifacts.mdx  |  6 +-
 .../src/content/docs/docs/tools/dashboard.mdx |  2 +-
 .../src/content/docs/docs/tools/results.mdx   |  4 +-
 .../docs/docs/tools/wip-checkpoints.mdx       |  2 +-
 ...-result-identity-and-default-experiment.md |  2 +-
 .../0011-result-output-artifact-contract.md   |  6 +-
 .../core/src/evaluation/result-row-schema.ts  |  1 +
 packages/core/src/evaluation/run-artifacts.ts |  4 +-
 23 files changed, 255 insertions(+), 108 deletions(-)

diff --git a/CONCEPTS.md b/CONCEPTS.md
index 92c2136b4..41ec016e2 100644
--- a/CONCEPTS.md
+++ b/CONCEPTS.md
@@ -24,7 +24,7 @@ Shared domain vocabulary for this project — entities, named processes, and sta
 
 **Workspace** — The task environment an eval prepares for the agent: repositories, templates, fixture files, and lifecycle hooks. It is not prompt input; use `input` for instructions and `workspace.repos[]` for multi-repo workspaces the agent can inspect or modify through tools.
 
-**Run manifest** — The root `index.jsonl` file in a run bundle. It is the dashboard and tooling loading contract for per-case result rows and artifact locations, including fields such as `result_dir`, `task_dir`, `summary_path`, and `grading_path`.
+**Run manifest** — The root `index.jsonl` file in a run bundle. It is the dashboard and tooling loading contract for per-case result rows and artifact locations, including fields such as `result_dir`, `test_dir`, `summary_path`, and `grading_path`.
 
 **Result source identity** — The stable source identity for a result row: repo-relative `eval_path`, `test_id`, and `target`. `suite` and `name` are display metadata, not storage or routing identity.
 
diff --git a/apps/cli/src/commands/eval/artifact-writer.ts b/apps/cli/src/commands/eval/artifact-writer.ts
index 32efbaad5..7f9c03c01 100644
--- a/apps/cli/src/commands/eval/artifact-writer.ts
+++ b/apps/cli/src/commands/eval/artifact-writer.ts
@@ -74,13 +74,13 @@ function buildTaskBundleIndexFields(
   taskBundle: MaterializedTaskBundlePaths | undefined,
 ): Pick<
   IndexArtifactEntry,
-  'task_dir' | 'eval_path' | 'targets_path' | 'files_path' | 'graders_path'
+  'test_dir' | 'eval_path' | 'targets_path' | 'files_path' | 'graders_path'
 > {
   if (!taskBundle) {
     return {};
   }
   return {
-    task_dir: toRelativeArtifactPath(outputDir, taskBundle.taskDir),
+    test_dir: toRelativeArtifactPath(outputDir, taskBundle.testDir),
     eval_path: toRelativeArtifactPath(outputDir, taskBundle.evalPath),
     targets_path: toRelativeArtifactPath(outputDir, taskBundle.targetsPath),
     ...(taskBundle.filesPath
@@ -123,14 +123,14 @@ export function buildResultIndexArtifact(
   const artifactSubdir = (buildCoreResultIndexArtifact(result).result_dir ?? '').trim();
   const extraIndexFields = taskBundle
     ? {
-        task_dir: path.posix.join(artifactSubdir, 'task'),
-        eval_path: path.posix.join(artifactSubdir, 'task', 'EVAL.yaml'),
-        targets_path: path.posix.join(artifactSubdir, 'task', 'targets.yaml'),
+        test_dir: path.posix.join(artifactSubdir, 'test'),
+        eval_path: path.posix.join(artifactSubdir, 'test', 'EVAL.yaml'),
+        targets_path: path.posix.join(artifactSubdir, 'test', 'targets.yaml'),
         ...(taskBundle.filesPath
-          ? { files_path: path.posix.join(artifactSubdir, 'task', 'files') }
+          ? { files_path: path.posix.join(artifactSubdir, 'test', 'files') }
           : {}),
         ...(taskBundle.gradersPath
-          ? { graders_path: path.posix.join(artifactSubdir, 'task', 'graders') }
+          ? { graders_path: path.posix.join(artifactSubdir, 'test', 'graders') }
           : {}),
       }
     : undefined;
diff --git a/apps/cli/src/commands/eval/task-bundle.ts b/apps/cli/src/commands/eval/task-bundle.ts
index 80287621b..f6e263d2a 100644
--- a/apps/cli/src/commands/eval/task-bundle.ts
+++ b/apps/cli/src/commands/eval/task-bundle.ts
@@ -15,7 +15,7 @@ import { stringify as stringifyYaml } from 'yaml';
 
 import { toSnakeCaseDeep } from '../../utils/case-conversion.js';
 
-const TASK_DIRNAME = 'task';
+const TEST_BUNDLE_DIRNAME = 'test';
 const TASK_EVAL_FILENAME = 'EVAL.yaml';
 const TASK_TARGETS_FILENAME = 'targets.yaml';
 const TASK_FILES_DIRNAME = 'files';
@@ -62,7 +62,7 @@ export interface MaterializeTaskBundleOptions {
 }
 
 export interface MaterializedTaskBundlePaths {
-  readonly taskDir: string;
+  readonly testDir: string;
   readonly evalPath: string;
   readonly targetsPath: string;
   readonly filesPath?: string;
@@ -950,14 +950,14 @@ export async function materializeTaskBundle(
     return undefined;
   }
 
-  const taskDir = path.join(options.outputDir, TASK_DIRNAME);
-  await mkdir(taskDir, { recursive: true });
+  const testDir = path.join(options.outputDir, TEST_BUNDLE_DIRNAME);
+  await mkdir(testDir, { recursive: true });
 
-  const copiedReferences = await copyReferences(options.test.source.references, taskDir, options);
+  const copiedReferences = await copyReferences(options.test.source.references, testDir, options);
   const rewrites = buildPathRewrites(copiedReferences);
   const evalCase = buildEvalCase(options.test, rewrites);
-  const evalPath = path.join(taskDir, TASK_EVAL_FILENAME);
-  const targetsPath = path.join(taskDir, TASK_TARGETS_FILENAME);
+  const evalPath = path.join(testDir, TASK_EVAL_FILENAME);
+  const targetsPath = path.join(testDir, TASK_TARGETS_FILENAME);
 
   await writeYamlFile(evalPath, {
     execution: { target: options.targetName },
@@ -966,14 +966,14 @@ export async function materializeTaskBundle(
   await writeYamlFile(targetsPath, { targets: targetDefinitions });
 
   return {
-    taskDir,
+    testDir,
     evalPath,
     targetsPath,
     ...(hasCopiedBucket(copiedReferences, 'files')
-      ? { filesPath: path.join(taskDir, TASK_FILES_DIRNAME) }
+      ? { filesPath: path.join(testDir, TASK_FILES_DIRNAME) }
       : {}),
     ...(hasCopiedBucket(copiedReferences, 'graders')
-      ? { gradersPath: path.join(taskDir, TASK_GRADERS_DIRNAME) }
+      ? { gradersPath: path.join(testDir, TASK_GRADERS_DIRNAME) }
       : {}),
   };
 }
@@ -982,7 +982,7 @@ export async function materializeTaskBundle(
  * Materialize a whole eval suite as a portable directory.
  *
  * This reuses the same source snapshots, dependency copying, path rewriting,
- * target slicing, and secret redaction used by per-result task bundles. The
+ * target slicing, and secret redaction used by per-result test bundles. The
  * output eval is intentionally explicit: inherited suite defaults are written
  * onto each bundled test case so the bundle can run without the source tree.
  */
diff --git a/apps/cli/src/commands/results/combine-run.ts b/apps/cli/src/commands/results/combine-run.ts
index 56b0382b6..2d2440562 100644
--- a/apps/cli/src/commands/results/combine-run.ts
+++ b/apps/cli/src/commands/results/combine-run.ts
@@ -377,6 +377,7 @@ const MANIFEST_PATH_FIELDS = [
   'transcript_raw_path',
   'metrics_path',
   'raw_provider_log_path',
+  'test_dir',
   'task_dir',
   'eval_path',
   'targets_path',
diff --git a/apps/cli/src/commands/results/manifest.ts b/apps/cli/src/commands/results/manifest.ts
index 823be5a1f..ceb4c9754 100644
--- a/apps/cli/src/commands/results/manifest.ts
+++ b/apps/cli/src/commands/results/manifest.ts
@@ -66,6 +66,7 @@ export interface ResultManifestRecord {
   readonly external_trace?: ExternalTraceMetadataWire;
   readonly response_path?: string;
   readonly result_dir?: string;
+  readonly test_dir?: string;
   readonly task_dir?: string;
   readonly eval_path?: string;
   readonly targets_path?: string;
diff --git a/apps/cli/src/commands/results/projection-bundle.ts b/apps/cli/src/commands/results/projection-bundle.ts
index 4a2c58fd1..1d4ee85c7 100644
--- a/apps/cli/src/commands/results/projection-bundle.ts
+++ b/apps/cli/src/commands/results/projection-bundle.ts
@@ -95,6 +95,7 @@ export type ProjectionBundleArtifactRefs = Partial<
     | 'transcript_path'
     | 'transcript_raw_path'
     | 'metrics_path'
+    | 'test_dir'
     | 'task_dir'
     | 'eval_path'
     | 'targets_path'
@@ -173,6 +174,7 @@ function artifactRefs(
     transcript_path: indexEntry.transcript_path,
     transcript_raw_path: indexEntry.transcript_raw_path,
     metrics_path: indexEntry.metrics_path,
+    test_dir: indexEntry.test_dir,
     task_dir: indexEntry.task_dir,
     eval_path: indexEntry.eval_path,
     targets_path: indexEntry.targets_path,
diff --git a/apps/cli/src/commands/results/serve.ts b/apps/cli/src/commands/results/serve.ts
index 23e6370aa..3e618cfaa 100644
--- a/apps/cli/src/commands/results/serve.ts
+++ b/apps/cli/src/commands/results/serve.ts
@@ -883,6 +883,7 @@ function resultArtifactTreeRootPaths(
 ): string[] {
   return [
     ...catalog.filter((entry) => entry.storage === 'local').map((entry) => entry.displayPath),
+    record.test_dir,
     record.task_dir,
     record.files_path,
     record.graders_path,
@@ -1212,6 +1213,7 @@ function attachRunDetailReadModelFields<T extends Record<string, unknown>>(
       ...(record.aggregation && { aggregation: record.aggregation }),
       ...(record.eval_path && { eval_path: record.eval_path }),
       ...(record.result_dir && { result_dir: record.result_dir }),
+      ...(record.test_dir && { test_dir: record.test_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 }),
diff --git a/apps/cli/src/commands/runs/rerun.ts b/apps/cli/src/commands/runs/rerun.ts
index 465efa3ec..abab40290 100644
--- a/apps/cli/src/commands/runs/rerun.ts
+++ b/apps/cli/src/commands/runs/rerun.ts
@@ -33,7 +33,7 @@ interface SelectedTaskBundle {
   readonly testId: string;
   readonly sourceTarget: string;
   readonly resultDir: string;
-  readonly taskDir: string;
+  readonly testDir: string;
   readonly evalPath: string;
   readonly targetsPath: string;
   readonly taskTarget: string;
@@ -254,7 +254,7 @@ function forbiddenOutputRoots(
 ): readonly string[] {
   return [
     path.resolve(sourceRunDir),
-    ...selected.flatMap((bundle) => [path.resolve(bundle.resultDir), path.resolve(bundle.taskDir)]),
+    ...selected.flatMap((bundle) => [path.resolve(bundle.resultDir), path.resolve(bundle.testDir)]),
   ];
 }
 
@@ -322,40 +322,41 @@ async function loadSelectedTaskBundles(options: {
     }
 
     const recordLabel = displayRecord(record);
+    const bundleDir = record.test_dir ?? record.task_dir;
     const evalPath =
       resolveRelativeRunPath(options.sourceRunDir, record.eval_path) ??
       resolveRelativeRunPath(
         options.sourceRunDir,
-        record.task_dir && `${record.task_dir}/${TASK_EVAL_FILENAME}`,
+        bundleDir && `${bundleDir}/${TASK_EVAL_FILENAME}`,
       );
     const targetsPath =
       resolveRelativeRunPath(options.sourceRunDir, record.targets_path) ??
       resolveRelativeRunPath(
         options.sourceRunDir,
-        record.task_dir && `${record.task_dir}/${TASK_TARGETS_FILENAME}`,
+        bundleDir && `${bundleDir}/${TASK_TARGETS_FILENAME}`,
       );
-    const taskDir =
-      resolveRelativeRunPath(options.sourceRunDir, record.task_dir) ??
+    const testDir =
+      resolveRelativeRunPath(options.sourceRunDir, bundleDir) ??
       (evalPath ? path.dirname(evalPath) : undefined);
     const resultDir =
       resolveRelativeRunPath(options.sourceRunDir, record.result_dir) ??
-      (taskDir ? path.dirname(taskDir) : undefined);
+      (testDir ? path.dirname(testDir) : undefined);
 
-    if (!evalPath || !targetsPath || !taskDir || !resultDir) {
+    if (!evalPath || !targetsPath || !testDir || !resultDir) {
       throw new Error(
-        `Selected result ${recordLabel} is missing task bundle paths. Re-run requires task/EVAL.yaml and task/targets.yaml.`,
+        `Selected result ${recordLabel} is missing test bundle paths. Re-run requires test/EVAL.yaml and test/targets.yaml.`,
       );
     }
 
-    await ensureFile(evalPath, `Task eval for ${recordLabel}`);
-    await ensureFile(targetsPath, `Task targets for ${recordLabel}`);
+    await ensureFile(evalPath, `Test eval for ${recordLabel}`);
+    await ensureFile(targetsPath, `Test targets for ${recordLabel}`);
     const taskTarget = await readTaskTarget(evalPath, sourceTarget);
     selected.push({
       record,
       testId,
       sourceTarget,
       resultDir,
-      taskDir,
+      testDir,
       evalPath,
       targetsPath,
       taskTarget,
@@ -364,7 +365,7 @@ async function loadSelectedTaskBundles(options: {
 
   if (selected.length === 0) {
     throw new Error(
-      'No captured task bundles matched the provided --test-id/--source-target filters.',
+      'No captured test bundles matched the provided --test-id/--source-target filters.',
     );
   }
   return selected;
@@ -384,7 +385,7 @@ function buildSourceMetadataByEvalFile(
           sourceRunDir: path.resolve(sourceRunDir),
           sourceIndexPath: path.resolve(indexPath),
           sourceResultDir: path.resolve(bundle.resultDir),
-          sourceTaskDir: path.resolve(bundle.taskDir),
+          sourceTestDir: path.resolve(bundle.testDir),
           sourceTestId: bundle.testId,
           sourceTarget: bundle.sourceTarget,
           sourceTimestamp: bundle.record.timestamp,
@@ -396,12 +397,12 @@ function buildSourceMetadataByEvalFile(
 
 export const runsRerunCommand = command({
   name: 'rerun',
-  description: 'Rerun captured task bundles with local target environment',
+  description: 'Rerun captured test bundles with local target environment',
   args: {
     runDir: positional({
       type: string,
       displayName: 'run-dir',
-      description: 'Run workspace directory or run manifest containing task bundles',
+      description: 'Run workspace directory or run manifest containing test bundles',
     }),
     testId: multioption({
       type: array(string),
@@ -490,11 +491,11 @@ export const runsRerunCommand = command({
         targetNamesByFile.set(bundle.targetsPath, names);
       }
       for (const [targetsPath, names] of targetNamesByFile.entries()) {
-        await validateTargetFile(targetsPath, [...names], 'Task bundle targets');
+        await validateTargetFile(targetsPath, [...names], 'Test bundle targets');
       }
     }
 
-    console.log(`Rerunning ${selected.length} captured task bundle(s) from: ${sourceRunDir}`);
+    console.log(`Rerunning ${selected.length} captured test bundle(s) from: ${sourceRunDir}`);
     console.log(`Rerun output directory: ${outputDir}`);
 
     const result = await runEvalCommand({
diff --git a/apps/cli/test/commands/eval/artifact-writer.test.ts b/apps/cli/test/commands/eval/artifact-writer.test.ts
index 8e736131f..2bc53aa26 100644
--- a/apps/cli/test/commands/eval/artifact-writer.test.ts
+++ b/apps/cli/test/commands/eval/artifact-writer.test.ts
@@ -1061,6 +1061,7 @@ describe('writeArtifactsFromResults', () => {
     });
     expect(indexEntry?.result_dir).toBe(repeatRowDir);
     expect(indexEntry?.summary_path).toBe(`${repeatRowDir}/summary.json`);
+    expect(indexEntry?.test_dir).toBeUndefined();
     expect(indexEntry?.task_dir).toBeUndefined();
     expect(indexEntry?.input_path).toBeUndefined();
     expect(indexEntry?.grading_path).toBeUndefined();
@@ -1939,7 +1940,7 @@ describe('writeArtifactsFromResults', () => {
     expect(alpha.result_dir).not.toBe(beta.result_dir);
   });
 
-  it('writes task bundle artifacts with local source paths when source metadata is provided', async () => {
+  it('writes test bundle artifacts with local source paths when source metadata is provided', async () => {
     const sourceRoot = path.join(testDir, 'src');
     await mkdir(sourceRoot, { recursive: true });
     const evalFile = path.join(sourceRoot, 'trace.eval.yaml');
@@ -2058,28 +2059,31 @@ describe('writeArtifactsFromResults', () => {
 
     const [indexLine] = await readIndexLines(paths.indexPath);
     const rowDir = expectRowDir(indexLine, 'trace-case');
-    const taskDir = path.join(outputDir, rowDir, 'task');
-    const evalPath = path.join(taskDir, 'EVAL.yaml');
-    const targetsPath = path.join(taskDir, 'targets.yaml');
+    const testBundleDir = path.join(outputDir, rowDir, 'test');
+    const evalPath = path.join(testBundleDir, 'EVAL.yaml');
+    const targetsPath = path.join(testBundleDir, 'targets.yaml');
     const taskEval = await readFile(evalPath, 'utf8');
     const taskTargets = await readFile(targetsPath, 'utf8');
 
     expect(indexLine).toMatchObject({
       result_dir: rowDir,
-      task_dir: `${rowDir}/task`,
-      eval_path: `${rowDir}/task/EVAL.yaml`,
-      targets_path: `${rowDir}/task/targets.yaml`,
-      files_path: `${rowDir}/task/files`,
-      graders_path: `${rowDir}/task/graders`,
+      test_dir: `${rowDir}/test`,
+      eval_path: `${rowDir}/test/EVAL.yaml`,
+      targets_path: `${rowDir}/test/targets.yaml`,
+      files_path: `${rowDir}/test/files`,
+      graders_path: `${rowDir}/test/graders`,
     });
-    expect(await readFile(path.join(taskDir, 'files', 'src', 'input.txt'), 'utf8')).toBe(
+    expect(indexLine.task_dir).toBeUndefined();
+    expect(await readFile(path.join(testBundleDir, 'files', 'src', 'input.txt'), 'utf8')).toBe(
       'input fixture\n',
     );
-    expect(await readFile(path.join(taskDir, 'files', 'src', '.env'), 'utf8')).toBe('[redacted]\n');
-    expect(await readFile(path.join(taskDir, 'graders', 'src', 'grader.md'), 'utf8')).toBe(
+    expect(await readFile(path.join(testBundleDir, 'files', 'src', '.env'), 'utf8')).toBe(
+      '[redacted]\n',
+    );
+    expect(await readFile(path.join(testBundleDir, 'graders', 'src', 'grader.md'), 'utf8')).toBe(
       'grade this response\n',
     );
-    expect(await readFile(path.join(taskDir, 'graders', 'src', 'prompt.ts'), 'utf8')).toBe(
+    expect(await readFile(path.join(testBundleDir, 'graders', 'src', 'prompt.ts'), 'utf8')).toBe(
       'console.log("prompt");\n',
     );
 
@@ -2102,10 +2106,10 @@ describe('writeArtifactsFromResults', () => {
     expect(taskEval).not.toContain('literal-secret');
     expect(taskTargets).not.toContain('literal-secret');
     await expect(readdir(path.join(outputDir, rowDir, '.agentv', 'results'))).rejects.toThrow();
-    await expect(readdir(path.join(taskDir, '.agentv', 'results'))).rejects.toThrow();
+    await expect(readdir(path.join(testBundleDir, '.agentv', 'results'))).rejects.toThrow();
   });
 
-  it('writes task bundle index links for multi-test runs', async () => {
+  it('writes test bundle index links for multi-test runs', async () => {
     const evalFile = path.join(testDir, 'multi.eval.yaml');
     await mkdir(path.dirname(evalFile), { recursive: true });
     await writeFile(
@@ -2152,18 +2156,19 @@ describe('writeArtifactsFromResults', () => {
 
     const indexLines = await readIndexLines(paths.indexPath);
     const rowDirs = indexLines.map((line) => expectRowDir(line, line.test_id));
-    expect(indexLines.map((line, index) => line.task_dir)).toEqual(
-      rowDirs.map((rowDir) => `${rowDir}/task`),
+    expect(indexLines.map((line) => line.test_dir)).toEqual(
+      rowDirs.map((rowDir) => `${rowDir}/test`),
     );
-    expect(await readdir(path.join(testDir, 'multi-out', rowDirs[0] ?? '', 'task'))).toContain(
+    expect(indexLines.map((line) => line.task_dir)).toEqual([undefined, undefined]);
+    expect(await readdir(path.join(testDir, 'multi-out', rowDirs[0] ?? '', 'test'))).toContain(
       'EVAL.yaml',
     );
-    expect(await readdir(path.join(testDir, 'multi-out', rowDirs[1] ?? '', 'task'))).toContain(
+    expect(await readdir(path.join(testDir, 'multi-out', rowDirs[1] ?? '', 'test'))).toContain(
       'EVAL.yaml',
     );
   });
 
-  it('matches task bundle targets by resolved result target while preserving selected target name', async () => {
+  it('matches test bundle targets by resolved result target while preserving selected target name', async () => {
     const evalFile = path.join(testDir, 'resolved-target.eval.yaml');
     await mkdir(path.dirname(evalFile), { recursive: true });
     await writeFile(evalFile, 'tests:\n  - id: alias-case\n    input: hello\n');
@@ -2204,10 +2209,11 @@ describe('writeArtifactsFromResults', () => {
 
     const [indexLine] = await readIndexLines(paths.indexPath);
     const rowDir = expectRowDir(indexLine, 'alias-case');
-    expect(indexLine.task_dir).toBe(`${rowDir}/task`);
+    expect(indexLine.test_dir).toBe(`${rowDir}/test`);
+    expect(indexLine.task_dir).toBeUndefined();
 
     const taskEval = await readFile(
-      path.join(testDir, 'resolved-target-out', rowDir, 'task', 'EVAL.yaml'),
+      path.join(testDir, 'resolved-target-out', rowDir, 'test', 'EVAL.yaml'),
       'utf8',
     );
     const parsedEval = parseYamlValue(taskEval) as Record<string, unknown>;
diff --git a/apps/cli/test/commands/eval/task-bundle.test.ts b/apps/cli/test/commands/eval/task-bundle.test.ts
index ef80d6e0f..98613183a 100644
--- a/apps/cli/test/commands/eval/task-bundle.test.ts
+++ b/apps/cli/test/commands/eval/task-bundle.test.ts
@@ -18,7 +18,7 @@ describe('materializeTaskBundle', () => {
     await rm(tempDir, { recursive: true, force: true });
   });
 
-  it('writes a self-contained task bundle without executing a provider', async () => {
+  it('writes a self-contained test bundle without executing a provider', async () => {
     const evalFile = path.join(tempDir, 'evals', 'demo.eval.yaml');
     const fixturePath = path.join(tempDir, 'fixtures', 'input.txt');
     const promptPath = path.join(tempDir, 'graders', 'prompt.md');
@@ -104,15 +104,16 @@ describe('materializeTaskBundle', () => {
     });
 
     expect(paths).toBeDefined();
-    const taskDir = paths?.taskDir ?? '';
-    expect(await readdir(taskDir)).toContain('EVAL.yaml');
-    expect(await readFile(path.join(taskDir, 'files', 'fixtures', 'input.txt'), 'utf8')).toBe(
+    const testBundleDir = paths?.testDir ?? '';
+    expect(path.basename(testBundleDir)).toBe('test');
+    expect(await readdir(testBundleDir)).toContain('EVAL.yaml');
+    expect(await readFile(path.join(testBundleDir, 'files', 'fixtures', 'input.txt'), 'utf8')).toBe(
       'fixture text\n',
     );
-    expect(await readFile(path.join(taskDir, 'graders', 'graders', 'prompt.md'), 'utf8')).toBe(
-      'grade carefully\n',
-    );
-    expect(await readFile(path.join(taskDir, 'graders', 'graders', 'check.ts'), 'utf8')).toBe(
+    expect(
+      await readFile(path.join(testBundleDir, 'graders', 'graders', 'prompt.md'), 'utf8'),
+    ).toBe('grade carefully\n');
+    expect(await readFile(path.join(testBundleDir, 'graders', 'graders', 'check.ts'), 'utf8')).toBe(
       'console.log("ok");\n',
     );
 
@@ -133,6 +134,6 @@ describe('materializeTaskBundle', () => {
     expect(taskEval).not.toContain('literal-secret');
     expect(taskTargets).not.toContain('literal-secret');
     await expect(readdir(path.join(tempDir, 'out', '.agentv', 'results'))).rejects.toThrow();
-    await expect(readdir(path.join(taskDir, '.agentv', 'results'))).rejects.toThrow();
+    await expect(readdir(path.join(testBundleDir, '.agentv', 'results'))).rejects.toThrow();
   });
 });
diff --git a/apps/cli/test/commands/results/serve.test.ts b/apps/cli/test/commands/results/serve.test.ts
index 3e867b91e..a41fe79a9 100644
--- a/apps/cli/test/commands/results/serve.test.ts
+++ b/apps/cli/test/commands/results/serve.test.ts
@@ -2751,7 +2751,7 @@ describe('serve app', () => {
       expect(data.source_label).toBe(filename);
     });
 
-    it('loads historical runs without task bundle metadata', async () => {
+    it('loads historical runs without test bundle metadata', async () => {
       const runId = writeLocalRunArtifact(
         tempDir,
         'historical',
@@ -2766,8 +2766,32 @@ describe('serve app', () => {
         results: Array<Record<string, unknown>>;
       };
       expect(data.results[0]).not.toHaveProperty('task_dir');
+      expect(data.results[0]).not.toHaveProperty('test_dir');
       expect(data.results[0]).not.toHaveProperty('source_traceability');
     });
+
+    it('loads generated test bundle metadata from manifest rows', async () => {
+      const runId = writeLocalRunArtifact(tempDir, 'generated-bundle', '2026-03-25T12-00-01-000Z', {
+        ...RESULT_A,
+        result_dir: 'demo/test-greeting',
+        test_dir: 'demo/test-greeting/test',
+        eval_path: 'demo/test-greeting/test/EVAL.yaml',
+        targets_path: 'demo/test-greeting/test/targets.yaml',
+      });
+
+      const app = createApp([], tempDir, tempDir, undefined, { studioDir });
+      const res = await app.request(`/api/runs/${encodeURIComponent(runId)}`);
+      expect(res.status).toBe(200);
+      const data = (await res.json()) as {
+        results: Array<Record<string, unknown>>;
+      };
+      expect(data.results[0]).toMatchObject({
+        result_dir: 'demo/test-greeting',
+        test_dir: 'demo/test-greeting/test',
+        eval_path: 'demo/test-greeting/test/EVAL.yaml',
+      });
+      expect(data.results[0]).not.toHaveProperty('task_dir');
+    });
   });
 
   describe('run combine API', () => {
diff --git a/apps/cli/test/commands/results/validate.test.ts b/apps/cli/test/commands/results/validate.test.ts
index 7b82d7da9..8b483a7bc 100644
--- a/apps/cli/test/commands/results/validate.test.ts
+++ b/apps/cli/test/commands/results/validate.test.ts
@@ -118,4 +118,67 @@ describe('results validate', () => {
       rmSync(tempDir, { recursive: true, force: true });
     }
   });
+
+  it('accepts new test_dir and legacy task_dir bundle metadata', () => {
+    const tempDir = mkdtempSync(path.join(tmpdir(), 'agentv-validate-test-'));
+
+    try {
+      const runDir = path.join(
+        tempDir,
+        '.agentv',
+        'results',
+        'with-bundles',
+        '2026-03-27T12-42-24-429Z',
+      );
+      mkdirSync(runDir, { recursive: true });
+      writeFileSync(
+        path.join(runDir, 'index.jsonl'),
+        `${[
+          JSON.stringify({
+            timestamp: '2026-03-27T12:42:24.429Z',
+            test_id: 'test-new',
+            score: 1,
+            target: 'gpt-4o',
+            scores: [{ name: 'quality', type: 'llm', score: 1, verdict: 'pass' }],
+            execution_status: 'ok',
+            summary_path: 'test-new/summary.json',
+            test_dir: 'test-new/test',
+            eval_path: 'test-new/test/EVAL.yaml',
+            targets_path: 'test-new/test/targets.yaml',
+          }),
+          JSON.stringify({
+            timestamp: '2026-03-27T12:42:24.429Z',
+            test_id: 'test-legacy',
+            score: 1,
+            target: 'gpt-4o',
+            scores: [{ name: 'quality', type: 'llm', score: 1, verdict: 'pass' }],
+            execution_status: 'ok',
+            summary_path: 'test-legacy/summary.json',
+            task_dir: 'test-legacy/task',
+            eval_path: 'test-legacy/task/EVAL.yaml',
+            targets_path: 'test-legacy/task/targets.yaml',
+          }),
+        ].join('\n')}\n`,
+      );
+      for (const testId of ['test-new', 'test-legacy']) {
+        mkdirSync(path.join(runDir, testId), { recursive: true });
+        writeFileSync(
+          path.join(runDir, testId, 'summary.json'),
+          `${JSON.stringify({
+            test_id: testId,
+            score: 1,
+            target: 'gpt-4o',
+            execution_status: 'ok',
+          })}\n`,
+        );
+      }
+      writeFileSync(path.join(runDir, 'summary.json'), '{}\n');
+
+      const { diagnostics } = validateRunDirectory(runDir);
+
+      expect(diagnostics.filter((d) => d.severity === 'error')).toEqual([]);
+    } finally {
+      rmSync(tempDir, { recursive: true, force: true });
+    }
+  });
 });
diff --git a/apps/cli/test/commands/runs/rerun.test.ts b/apps/cli/test/commands/runs/rerun.test.ts
index 2a6be746a..ee9d9701e 100644
--- a/apps/cli/test/commands/runs/rerun.test.ts
+++ b/apps/cli/test/commands/runs/rerun.test.ts
@@ -38,15 +38,17 @@ async function writeTaskBundle(options: {
   readonly sourceRunDir: string;
   readonly testId: string;
   readonly targetsYaml: string;
+  readonly legacyTaskDir?: boolean;
 }): Promise<Record<string, unknown>> {
   const artifactDir = path.join(options.sourceRunDir, options.testId);
-  const taskDir = path.join(artifactDir, 'task');
+  const bundleDirname = options.legacyTaskDir ? 'task' : 'test';
+  const bundleDir = path.join(artifactDir, bundleDirname);
   const outputsDir = path.join(artifactDir, 'outputs');
-  await mkdir(taskDir, { recursive: true });
+  await mkdir(bundleDir, { recursive: true });
   await mkdir(outputsDir, { recursive: true });
 
   await writeFile(
-    path.join(taskDir, 'EVAL.yaml'),
+    path.join(bundleDir, 'EVAL.yaml'),
     `execution:
   target: captured
 tests:
@@ -58,11 +60,17 @@ tests:
 `,
     'utf8',
   );
-  await writeFile(path.join(taskDir, 'targets.yaml'), options.targetsYaml, 'utf8');
+  await writeFile(path.join(bundleDir, 'targets.yaml'), options.targetsYaml, 'utf8');
   await writeFile(path.join(artifactDir, 'grading.json'), '{"assertions":[]}\n', 'utf8');
   await writeFile(path.join(artifactDir, 'timing.json'), '{"duration_ms":1}\n', 'utf8');
   await writeFile(path.join(outputsDir, 'answer.md'), '@[assistant]:\nCaptured answer\n', 'utf8');
 
+  const bundlePaths = {
+    [`${options.legacyTaskDir ? 'task' : 'test'}_dir`]: `${options.testId}/${bundleDirname}`,
+    eval_path: `${options.testId}/${bundleDirname}/EVAL.yaml`,
+    targets_path: `${options.testId}/${bundleDirname}/targets.yaml`,
+  };
+
   return {
     timestamp: '2024-01-01T00:00:00.000Z',
     test_id: options.testId,
@@ -73,13 +81,14 @@ tests:
     timing_path: `${options.testId}/timing.json`,
     output_path: `${options.testId}/outputs/answer.md`,
     answer_path: `${options.testId}/outputs/answer.md`,
-    task_dir: `${options.testId}/task`,
-    eval_path: `${options.testId}/task/EVAL.yaml`,
-    targets_path: `${options.testId}/task/targets.yaml`,
+    ...bundlePaths,
   };
 }
 
-async function createBundleFixture(targetsYaml = DEFAULT_TARGETS): Promise<BundleFixture> {
+async function createBundleFixture(
+  targetsYaml = DEFAULT_TARGETS,
+  options?: { readonly legacyTaskDir?: boolean },
+): Promise<BundleFixture> {
   const baseDir = await mkdtemp(path.join(tmpdir(), 'agentv-rerun-'));
   const cwd = path.join(baseDir, 'workspace');
   const sourceRunDir = path.join(baseDir, 'source-run');
@@ -88,8 +97,18 @@ async function createBundleFixture(targetsYaml = DEFAULT_TARGETS): Promise<Bundl
   await mkdir(sourceRunDir, { recursive: true });
 
   const records = [
-    await writeTaskBundle({ sourceRunDir, testId: 'case-alpha', targetsYaml }),
-    await writeTaskBundle({ sourceRunDir, testId: 'case-beta', targetsYaml }),
+    await writeTaskBundle({
+      sourceRunDir,
+      testId: 'case-alpha',
+      targetsYaml,
+      legacyTaskDir: options?.legacyTaskDir,
+    }),
+    await writeTaskBundle({
+      sourceRunDir,
+      testId: 'case-beta',
+      targetsYaml,
+      legacyTaskDir: options?.legacyTaskDir,
+    }),
   ];
   await writeFile(
     path.join(sourceRunDir, 'index.jsonl'),
@@ -197,7 +216,7 @@ describe('agentv runs rerun', () => {
     return created;
   }
 
-  it('reruns captured task bundles into an explicit output directory with source metadata', async () => {
+  it('reruns captured test bundles into an explicit output directory with source metadata', async () => {
     const created = await fixture();
 
     const result = await runCli(created, [
@@ -210,13 +229,14 @@ describe('agentv runs rerun', () => {
     ]);
 
     expect(result.exitCode).toBe(0);
-    expect(result.stdout).toContain('Rerunning 2 captured task bundle(s)');
+    expect(result.stdout).toContain('Rerunning 2 captured test bundle(s)');
     const { indexPath, rows } = await readOutputBundle(created.outputDir);
     expect(rows.map((row) => row.test_id)).toEqual(['case-alpha', 'case-beta']);
     expect(rows.every((row) => row.target === 'captured')).toBe(true);
     expect(rows[0].metadata).toMatchObject({
       rerun_source: {
         mode: 'rerun',
+        source_test_dir: path.join(created.sourceRunDir, 'case-alpha', 'test'),
         source_test_id: 'case-alpha',
         source_target: 'captured',
       },
@@ -228,6 +248,29 @@ describe('agentv runs rerun', () => {
     expect(answer).not.toContain('Captured answer');
   }, 30_000);
 
+  it('reruns legacy task_dir bundles for backward compatibility', async () => {
+    const created = await createBundleFixture(DEFAULT_TARGETS, { legacyTaskDir: true });
+
+    const result = await runCli(created, [
+      'runs',
+      'rerun',
+      created.sourceRunDir,
+      '--test-id',
+      'case-alpha',
+      '--output',
+      created.outputDir,
+    ]);
+
+    expect(result.exitCode).toBe(0);
+    const { rows } = await readOutputBundle(created.outputDir);
+    expect(rows.map((row) => row.test_id)).toEqual(['case-alpha']);
+    expect(rows[0].metadata).toMatchObject({
+      rerun_source: {
+        source_test_dir: path.join(created.sourceRunDir, 'case-alpha', 'task'),
+      },
+    });
+  }, 30_000);
+
   it('fails clearly for missing env and accepts an explicit env file', async () => {
     const created = await fixture(`targets:
   - name: captured
@@ -269,7 +312,7 @@ describe('agentv runs rerun', () => {
 
   it('fails loudly when selected bundle artifacts are missing', async () => {
     const created = await fixture();
-    await rm(path.join(created.sourceRunDir, 'case-beta', 'task', 'targets.yaml'));
+    await rm(path.join(created.sourceRunDir, 'case-beta', 'test', 'targets.yaml'));
 
     const result = await runCli(created, [
       'runs',
@@ -282,7 +325,7 @@ describe('agentv runs rerun', () => {
     ]);
 
     expect(result.exitCode).toBe(1);
-    expect(result.stderr).toContain('Task targets for case-beta@captured not found');
+    expect(result.stderr).toContain('Test targets for case-beta@captured not found');
   }, 30_000);
 
   it('reruns a selected test subset from index.jsonl', async () => {
@@ -303,29 +346,29 @@ describe('agentv runs rerun', () => {
     expect(rows.map((row) => row.test_id)).toEqual(['case-alpha']);
   }, 30_000);
 
-  it('chooses a default output directory outside the source task folder', async () => {
+  it('chooses a default output directory outside the source test bundle folder', async () => {
     const created = await fixture();
-    const taskDir = path.join(created.sourceRunDir, 'case-alpha', 'task');
+    const testBundleDir = path.join(created.sourceRunDir, 'case-alpha', 'test');
 
     const result = await runCli(
       created,
       ['runs', 'rerun', created.sourceRunDir, '--test-id', 'case-alpha'],
-      { cwd: taskDir },
+      { cwd: testBundleDir },
     );
 
     expect(result.exitCode).toBe(0);
     const outputDir = extractRerunOutputDir(result.stdout);
-    expect(path.relative(taskDir, outputDir).startsWith('..')).toBe(true);
+    expect(path.relative(testBundleDir, outputDir).startsWith('..')).toBe(true);
     const { rows } = await readOutputBundle(outputDir);
     expect(rows.map((row) => row.test_id)).toEqual(['case-alpha']);
   }, 30_000);
 
-  it('rejects explicit output nested under a source task folder', async () => {
+  it('rejects explicit output nested under a source test bundle folder', async () => {
     const created = await fixture();
     const nestedOutput = path.join(
       created.sourceRunDir,
       'case-alpha',
-      'task',
+      'test',
       '.agentv',
       'results',
     );
diff --git a/apps/web/src/content/docs/docs/evaluation/running-evals.mdx b/apps/web/src/content/docs/docs/evaluation/running-evals.mdx
index bb76a9e89..6095d640d 100644
--- a/apps/web/src/content/docs/docs/evaluation/running-evals.mdx
+++ b/apps/web/src/content/docs/docs/evaluation/running-evals.mdx
@@ -118,9 +118,9 @@ agentv eval evals/my-eval.yaml --output ./my-results
 cat ./my-results/index.jsonl
 ```
 
-### Generated Task Bundles
+### Generated Test Bundles
 
-Each result can also include a generated task bundle inside its per-test result
+Each result can also include a generated test bundle inside its per-test result
 directory. The bundle captures the eval slice and target settings that produced
 that row, so reviewers and rerun tooling can inspect the exact run-local source
 instead of relying on a mutable checkout.
@@ -141,7 +141,7 @@ my-results/
       transcript.jsonl
       transcript-raw.jsonl
       outputs/answer.md
-    task/
+    test/
       EVAL.yaml
       targets.yaml
       files/      # copied input files when the case references them
@@ -149,7 +149,7 @@ my-results/
 ```
 
 The `index.jsonl` row links to these generated paths with snake_case fields such
-as `result_dir`, `task_dir`, `eval_path`, `targets_path`, `files_path`, and
+as `result_dir`, `test_dir`, `eval_path`, `targets_path`, `files_path`, and
 `graders_path`. Treat those paths as relative to the run directory. When you need
 a portable artifact for audit, review, Dashboard inspection, or rerun workflows,
 share the generated run directory and its `index.jsonl` manifest. Source-side
@@ -165,7 +165,7 @@ for a result row. `suite` and `name` are display metadata only; do not use them
 to infer storage paths or pick a Dashboard detail row.
 
 If the source eval uses the `PROMPT.md` fallback instead of inline `input`,
-AgentV records the generated task bundle metadata when source artifacts are
+AgentV records the generated test bundle metadata when source artifacts are
 available. It no longer emits a generated prompt sidecar for result rows.
 
 ### Manual or External-Agent Attempts
diff --git a/apps/web/src/content/docs/docs/guides/benchmark-provenance.mdx b/apps/web/src/content/docs/docs/guides/benchmark-provenance.mdx
index e6a195983..7cd9893a5 100644
--- a/apps/web/src/content/docs/docs/guides/benchmark-provenance.mdx
+++ b/apps/web/src/content/docs/docs/guides/benchmark-provenance.mdx
@@ -80,11 +80,11 @@ Benchmark task packs map cleanly onto AgentV fields at authoring time:
 Use this separation only when it makes the source eval easier to maintain. It is
 not a first-class artifact schema. After an eval runs, AgentV writes the portable
 audit surface into the generated run folder: each result can link from
-`index.jsonl` to a run-local `task/` bundle containing `EVAL.yaml`,
+`index.jsonl` to a run-local `test/` bundle containing `EVAL.yaml`,
 `targets.yaml`, and copied `files/` or `graders/` snapshots where applicable.
 Review, Dashboard files views, and rerun workflows should inspect those generated
 run artifacts instead of requiring authors to maintain a parallel source-side
-bundle layout. See [Generated Task Bundles](/docs/evaluation/running-evals/#generated-task-bundles).
+bundle layout. See [Generated Test Bundles](/docs/evaluation/running-evals/#generated-test-bundles).
 
 ## SWE-Style Case
 
diff --git a/apps/web/src/content/docs/docs/reference/result-artifacts.mdx b/apps/web/src/content/docs/docs/reference/result-artifacts.mdx
index 37b6e39b4..dcb0c8150 100644
--- a/apps/web/src/content/docs/docs/reference/result-artifacts.mdx
+++ b/apps/web/src/content/docs/docs/reference/result-artifacts.mdx
@@ -31,7 +31,7 @@ The default local layout is:
       tags.json                 # optional mutable Dashboard tags
       <case-or-allocation>/
         summary.json            # optional per-case aggregate, especially repeats
-        task/                   # optional generated task bundle
+        test/                   # optional generated test bundle
           EVAL.yaml
           targets.yaml
           files/
@@ -81,7 +81,7 @@ query.
 | `timing.json` | Duration, token usage, cost usage, and source labels such as `provider_reported`, `token_estimated`, `aggregate`, or `unavailable`. | Cost/latency reporting and provider-accounting audits. |
 | `transcript.jsonl` | AgentV-normalized transcript/timeline rows. | Portable human review, replay, transcript-aware graders, and tool-trajectory analysis. |
 | `transcript-raw.jsonl` | Native provider or harness evidence when available. | Parser debugging, forensic review, and preserving source bytes without making provider schemas public AgentV fields. |
-| `task/` | Generated task bundle for the exact eval slice and target settings that produced a row. | Audit, external review, and rerun workflows that should not depend on a mutable source checkout. |
+| `test/` | Generated test bundle for the exact eval slice and target settings that produced a row. | Audit, external review, and rerun workflows that should not depend on a mutable source checkout. |
 | `artifact_pointers` | Offload indirection for large detached payload bytes. | Finding payloads published outside the primary metadata/control-plane branch, such as transcript bytes on `agentv/artifacts/v1`. |
 
 `summary.json` and `index.jsonl` are complementary, not redundant. A run list
@@ -127,7 +127,7 @@ Example row:
   "transcript_raw_path": "refund-eligibility/run-1/transcript-raw.jsonl",
   "output_path": "refund-eligibility/run-1/outputs/answer.md",
   "answer_path": "refund-eligibility/run-1/outputs/answer.md",
-  "task_dir": "refund-eligibility/task"
+  "test_dir": "refund-eligibility/test"
 }
 ```
 
diff --git a/apps/web/src/content/docs/docs/tools/dashboard.mdx b/apps/web/src/content/docs/docs/tools/dashboard.mdx
index 64b59a597..66e646d8f 100644
--- a/apps/web/src/content/docs/docs/tools/dashboard.mdx
+++ b/apps/web/src/content/docs/docs/tools/dashboard.mdx
@@ -101,7 +101,7 @@ You can also set the same field globally in `$AGENTV_HOME/config.yaml` or `~/.ag
 
 ## Run Detail
 
-Click any run to see a breakdown by suite, per-test scores, target, duration, and cost. The source label (`local` or `remote`) tells you where the run came from. Files and source views resolve against the generated run artifacts referenced by `index.jsonl`—including per-result task bundles when present—so Dashboard does not require authors to create a separate source-side bundle structure.
+Click any run to see a breakdown by suite, per-test scores, target, duration, and cost. The source label (`local` or `remote`) tells you where the run came from. Files and source views resolve against the generated run artifacts referenced by `index.jsonl`—including per-result test bundles when present—so Dashboard does not require authors to create a separate source-side bundle structure.
 
 In the per-test results table, click a test ID to open its checks, transcript, source, files, and feedback in a row detail panel while the table, filters, and scroll position stay in place. Use **Full page** from the panel when you want the standalone eval detail route.
 
diff --git a/apps/web/src/content/docs/docs/tools/results.mdx b/apps/web/src/content/docs/docs/tools/results.mdx
index c0d0e3d5a..6795a45b1 100644
--- a/apps/web/src/content/docs/docs/tools/results.mdx
+++ b/apps/web/src/content/docs/docs/tools/results.mdx
@@ -99,7 +99,7 @@ Use `results export` when you need the artifact workspace layout itself rather t
 agentv results export <run-workspace-or-index.jsonl> [--out <dir>] [--duplicate-policy update]
 ```
 
-This is useful when a manifest needs to be materialized into a predictable artifact tree for other tooling, review, or archiving. The run workspace is also where generated task bundles live: `index.jsonl` rows may point to per-result `task_dir`, `eval_path`, `targets_path`, `files_path`, and `graders_path` entries. Keep those generated artifacts with the run when sharing or auditing results.
+This is useful when a manifest needs to be materialized into a predictable artifact tree for other tooling, review, or archiving. The run workspace is also where generated test bundles live: `index.jsonl` rows may point to per-result `test_dir`, `eval_path`, `targets_path`, `files_path`, and `graders_path` entries. Keep those generated artifacts with the run when sharing or auditing results.
 
 The export source is still the canonical run bundle described in the
 [Result Artifact Contract](/docs/reference/result-artifacts/): `summary.json`
@@ -175,7 +175,7 @@ Agent Skills eval artifacts map into AgentV like this:
 
 | Agent Skills pattern | AgentV field | Artifact location |
 |----------------------|--------------|-------------------|
-| Authored `evals/evals.json` cases | AgentV eval cases and task bundle paths | Eval source plus optional `task_dir`, `eval_path`, `targets_path`, `files_path`, and `graders_path` in `index.jsonl` |
+| Authored `evals/evals.json` cases | AgentV eval cases and test bundle paths | Eval source plus optional `test_dir`, `eval_path`, `targets_path`, `files_path`, and `graders_path` in `index.jsonl` |
 | Per-case answer | Generated target output artifact | `run-N/outputs/answer.md` |
 | Per-attempt sidecars | Normalized transcript, metrics, and raw provider evidence | `run-N/transcript.jsonl`, `run-N/transcript-raw.jsonl`, `run-N/metrics.json` |
 | Per-attempt `timing.json` | Duration, token totals, cost, and usage source labels | `run-N/timing.json` |
diff --git a/apps/web/src/content/docs/docs/tools/wip-checkpoints.mdx b/apps/web/src/content/docs/docs/tools/wip-checkpoints.mdx
index 2eb389fa9..9baf16d93 100644
--- a/apps/web/src/content/docs/docs/tools/wip-checkpoints.mdx
+++ b/apps/web/src/content/docs/docs/tools/wip-checkpoints.mdx
@@ -85,7 +85,7 @@ git push origin --delete agentv/wip/<hostname>/<run-dir-basename>
 
 - The first remote checkpoint happens on the periodic interval, so a process that dies immediately after startup may only have the local `summary.json` stub.
 - The WIP branch is force-pushed and keeps one snapshot commit. Do not treat it as an audit log.
-- Checkpoint contents can include prompts, outputs, grader evidence, traces, and generated task bundles. Protect the results repo like any other eval artifact store.
+- Checkpoint contents can include prompts, outputs, grader evidence, traces, and generated test bundles. Protect the results repo like any other eval artifact store.
 - Authentication and branch permissions are the same as normal results auto-push. If git or GitHub authentication is missing, AgentV warns and keeps evaluating locally.
 - WIP worktrees are based on the configured storage branch. Missing storage branches are initialized automatically; missing remotes or authentication still prevent WIP pushes until Git credentials are available.
 - Failed or interrupted runs intentionally leave WIP branches behind. Periodically delete old `agentv/wip/...` branches once recovered or obsolete.
diff --git a/docs/adr/0009-eval-path-result-identity-and-default-experiment.md b/docs/adr/0009-eval-path-result-identity-and-default-experiment.md
index 9ce000bbb..6b18791c7 100644
--- a/docs/adr/0009-eval-path-result-identity-and-default-experiment.md
+++ b/docs/adr/0009-eval-path-result-identity-and-default-experiment.md
@@ -102,7 +102,7 @@ they must not drive visible storage hierarchy, semantic routing, Dashboard
 detail selection, rerun lookup, import identity, or artifact discovery.
 
 `index.jsonl` is authoritative for all bundle-relative artifact paths. Per-row
-directories are exposed with `result_dir`. Sidecar paths such as `task_dir`,
+directories are exposed with `result_dir`. Sidecar paths such as `test_dir`,
 `summary_path`, `grading_path`, `metrics_path`, `transcript_path`,
 `targets_path`, `files_path`, and `graders_path` are explicit manifest fields.
 Consumers must use these fields instead of reconstructing paths from
diff --git a/docs/adr/0011-result-output-artifact-contract.md b/docs/adr/0011-result-output-artifact-contract.md
index a5f2a76c9..bf3e5224a 100644
--- a/docs/adr/0011-result-output-artifact-contract.md
+++ b/docs/adr/0011-result-output-artifact-contract.md
@@ -52,7 +52,7 @@ An AgentV result output is a run-centric bundle with this root contract:
   tags.json                 # optional mutable overlay
   <case-or-allocation>/
     summary.json            # optional case aggregate, especially repeats
-    task/                   # optional generated task bundle
+    test/                   # optional generated test bundle
     run-1/
       result.json
       grading.json
@@ -82,7 +82,7 @@ aggregate summaries.
 ordinary per-case sidecars through explicit fields such as `result_dir`,
 `summary_path`, `grading_path`, `metrics_path`, `timing_path`,
 `transcript_path`, `transcript_raw_path`, `answer_path`, `output_path`,
-`task_dir`, `eval_path`, `targets_path`, `files_path`, and `graders_path` when
+`test_dir`, `eval_path`, `targets_path`, `files_path`, and `graders_path` when
 those artifacts exist.
 
 `artifact_pointers` remain an offload indirection for large detached payload
@@ -154,7 +154,7 @@ This aligns with Margin Evals' manifest-first and run-centric lessons:
 
 AgentV does not copy Margin Evals' exact layout. AgentV keeps its own
 `summary.json` plus `index.jsonl` split, AgentV transcript sidecars, repeat-run
-attempt folders, generated task bundles, Git-backed result branch model, and
+attempt folders, generated test bundles, Git-backed result branch model, and
 optional detached `artifact_pointers`.
 
 ## Consequences
diff --git a/packages/core/src/evaluation/result-row-schema.ts b/packages/core/src/evaluation/result-row-schema.ts
index d5b771031..f222c09e2 100644
--- a/packages/core/src/evaluation/result-row-schema.ts
+++ b/packages/core/src/evaluation/result-row-schema.ts
@@ -41,6 +41,7 @@ const RESULT_ROW_ALIASES = {
   summaryPath: 'summary_path',
   targetsPath: 'targets_path',
   taskDir: 'task_dir',
+  testDir: 'test_dir',
   testId: 'test_id',
   timingPath: 'timing_path',
   tokenUsage: 'token_usage',
diff --git a/packages/core/src/evaluation/run-artifacts.ts b/packages/core/src/evaluation/run-artifacts.ts
index 53d59989e..5bd0609bc 100644
--- a/packages/core/src/evaluation/run-artifacts.ts
+++ b/packages/core/src/evaluation/run-artifacts.ts
@@ -465,6 +465,7 @@ export interface IndexArtifactEntry {
   readonly runtime_source?: RunRuntimeSourceMetadata;
   readonly raw_provider_log_path?: string;
   readonly input_path?: string;
+  readonly test_dir?: string;
   readonly task_dir?: string;
   readonly eval_path?: string;
   readonly targets_path?: string;
@@ -485,6 +486,7 @@ export type ResultIndexArtifact = IndexArtifactEntry;
 export type AdditionalResultIndexFields = Partial<
   Pick<
     IndexArtifactEntry,
+    | 'test_dir'
     | 'task_dir'
     | 'eval_path'
     | 'targets_path'
@@ -762,7 +764,7 @@ function toIndexRerunSource(value: unknown): Record<string, unknown> | undefined
     source_run_dir: value.sourceRunDir,
     source_index_path: value.sourceIndexPath,
     source_result_dir: value.sourceResultDir,
-    source_task_dir: value.sourceTaskDir,
+    source_test_dir: value.sourceTestDir ?? value.sourceTaskDir,
     source_test_id: value.sourceTestId,
     source_target: value.sourceTarget,
     source_timestamp: value.sourceTimestamp,

From bea4b3c02c258b9ad7e4f6c3e708638324146cd7 Mon Sep 17 00:00:00 2001
From: Christopher Tso <christso@gmail.com>
Date: Tue, 30 Jun 2026 09:19:16 +0200
Subject: [PATCH 2/2] fix(results): preserve generated test bundles on export

---
 apps/cli/src/commands/eval/artifact-writer.ts |   6 +-
 apps/cli/src/commands/results/export.ts       | 121 +++++++++++++++++-
 apps/cli/test/commands/results/export.test.ts | 105 +++++++++++++++
 3 files changed, 223 insertions(+), 9 deletions(-)

diff --git a/apps/cli/src/commands/eval/artifact-writer.ts b/apps/cli/src/commands/eval/artifact-writer.ts
index 7f9c03c01..edf3704d5 100644
--- a/apps/cli/src/commands/eval/artifact-writer.ts
+++ b/apps/cli/src/commands/eval/artifact-writer.ts
@@ -220,6 +220,7 @@ export async function writePerTestArtifacts(
     repoRoot?: string;
     sourceTests?: readonly EvalTest[];
     taskBundleTargets?: readonly TaskBundleTargetSelection[];
+    additionalArtifacts?: AdditionalResultArtifactsWriter;
     runtimeSource?: RunRuntimeSourceMetadata;
   },
 ): Promise<void> {
@@ -229,7 +230,7 @@ export async function writePerTestArtifacts(
     runId: options?.runId,
     duplicatePolicy: options?.duplicatePolicy,
     sourceTests: options?.sourceTests,
-    additionalArtifacts: createTaskBundleArtifactsWriter(options),
+    additionalArtifacts: options?.additionalArtifacts ?? createTaskBundleArtifactsWriter(options),
     runtimeSource: options?.runtimeSource,
   });
 }
@@ -249,6 +250,7 @@ export async function writeArtifactsFromResults(
     repoRoot?: string;
     sourceTests?: readonly EvalTest[];
     taskBundleTargets?: readonly TaskBundleTargetSelection[];
+    additionalArtifacts?: AdditionalResultArtifactsWriter;
     runtimeSource?: RunRuntimeSourceMetadata;
   },
 ): Promise<{
@@ -265,7 +267,7 @@ export async function writeArtifactsFromResults(
     duplicatePolicy: options?.duplicatePolicy,
     resultGroup: options?.resultGroup,
     sourceTests: options?.sourceTests,
-    additionalArtifacts: createTaskBundleArtifactsWriter(options),
+    additionalArtifacts: options?.additionalArtifacts ?? createTaskBundleArtifactsWriter(options),
     runtimeSource: options?.runtimeSource,
   });
 }
diff --git a/apps/cli/src/commands/results/export.ts b/apps/cli/src/commands/results/export.ts
index ba21b0d8e..aa0994624 100644
--- a/apps/cli/src/commands/results/export.ts
+++ b/apps/cli/src/commands/results/export.ts
@@ -20,12 +20,18 @@
  *   - To add new per-test workspace files, add them under each test directory.
  */
 
-import { readFileSync } from 'node:fs';
+import { cpSync, existsSync, readFileSync } from 'node:fs';
 import path from 'node:path';
 
 import { command, flag, oneOf, option, optional, positional, string } from 'cmd-ts';
 
-import type { EvaluationResult, ExportDuplicatePolicy, IndexArtifactEntry } from '@agentv/core';
+import type {
+  AdditionalResultArtifactsWriter,
+  AdditionalResultIndexFields,
+  EvaluationResult,
+  ExportDuplicatePolicy,
+  IndexArtifactEntry,
+} from '@agentv/core';
 
 import { parseJsonlResults, writeArtifactsFromResults } from '../eval/artifact-writer.js';
 import {
@@ -51,6 +57,7 @@ export async function exportResults(
   options?: { duplicatePolicy?: ExportDuplicatePolicy },
 ): Promise<void> {
   const results = parseJsonlResults(content);
+  const sourceIndexRecords = parseIndexArtifactEntries(content);
 
   if (results.length === 0) {
     throw new Error(`No results found in ${sourceFile}`);
@@ -60,6 +67,11 @@ export async function exportResults(
     evalFile: sourceFile,
     runId: deriveExportRunId(sourceFile),
     duplicatePolicy: options?.duplicatePolicy ?? 'update',
+    additionalArtifacts: createExportBundleArtifactsWriter({
+      outputDir,
+      sourceBaseDir: path.dirname(sourceFile),
+      sourceRecordsByResult: buildSourceRecordMap(results, sourceIndexRecords),
+    }),
   });
 }
 
@@ -100,20 +112,109 @@ export function deriveExportRunId(sourceFile: string): string {
 export async function loadExportSource(
   source: string | undefined,
   cwd: string,
-): Promise<{ sourceFile: string; results: readonly EvaluationResult[] }> {
+): Promise<{
+  sourceFile: string;
+  results: readonly EvaluationResult[];
+  indexRecords?: readonly IndexArtifactEntry[];
+}> {
   const { sourceFile } = await resolveSourceFile(source, cwd);
   const { results } = await loadSharedResults(source, cwd);
-  return { sourceFile, results };
+  const indexRecords = isRunManifestPath(sourceFile)
+    ? readIndexArtifactEntries(sourceFile)
+    : undefined;
+  return { sourceFile, results, indexRecords };
 }
 
-function readIndexArtifactEntries(indexPath: string): IndexArtifactEntry[] {
-  return readFileSync(indexPath, 'utf8')
+function parseIndexArtifactEntries(content: string): IndexArtifactEntry[] {
+  return content
     .split(/\r?\n/)
     .map((line) => line.trim())
     .filter(Boolean)
     .map((line) => JSON.parse(line) as IndexArtifactEntry);
 }
 
+function readIndexArtifactEntries(indexPath: string): IndexArtifactEntry[] {
+  return parseIndexArtifactEntries(readFileSync(indexPath, 'utf8'));
+}
+
+function buildSourceRecordMap(
+  results: readonly EvaluationResult[],
+  sourceRecords: readonly IndexArtifactEntry[],
+): ReadonlyMap<EvaluationResult, IndexArtifactEntry> {
+  return new Map(
+    results.flatMap((result, index) => {
+      const sourceRecord = sourceRecords[index];
+      return sourceRecord ? [[result, sourceRecord] as const] : [];
+    }),
+  );
+}
+
+function isSafeRelativePath(relativePath: string | undefined): relativePath is string {
+  return (
+    typeof relativePath === 'string' &&
+    relativePath.trim().length > 0 &&
+    !path.isAbsolute(relativePath) &&
+    !relativePath.split(/[\\/]+/).includes('..')
+  );
+}
+
+function toRelativeArtifactPath(outputDir: string, filePath: string): string {
+  return path.relative(outputDir, filePath).split(path.sep).join('/');
+}
+
+function hasCopiedSubdir(testBundleDir: string, dirname: string): boolean {
+  return existsSync(path.join(testBundleDir, dirname));
+}
+
+function createExportBundleArtifactsWriter(options: {
+  readonly outputDir: string;
+  readonly sourceBaseDir: string;
+  readonly sourceRecordsByResult: ReadonlyMap<EvaluationResult, IndexArtifactEntry>;
+}): AdditionalResultArtifactsWriter | undefined {
+  if (options.sourceRecordsByResult.size === 0) {
+    return undefined;
+  }
+
+  return async ({ result, testDir }): Promise<AdditionalResultIndexFields | undefined> => {
+    const sourceRecord = options.sourceRecordsByResult.get(result);
+    const sourceBundleDir = sourceRecord?.test_dir ?? sourceRecord?.task_dir;
+    if (!isSafeRelativePath(sourceBundleDir)) {
+      return undefined;
+    }
+
+    const sourceBundlePath = path.join(options.sourceBaseDir, sourceBundleDir);
+    const testBundlePath = path.join(testDir, 'test');
+    if (existsSync(sourceBundlePath)) {
+      cpSync(sourceBundlePath, testBundlePath, { recursive: true, force: true });
+    }
+
+    return {
+      test_dir: toRelativeArtifactPath(options.outputDir, testBundlePath),
+      eval_path: toRelativeArtifactPath(options.outputDir, path.join(testBundlePath, 'EVAL.yaml')),
+      targets_path: toRelativeArtifactPath(
+        options.outputDir,
+        path.join(testBundlePath, 'targets.yaml'),
+      ),
+      ...(sourceRecord?.files_path || hasCopiedSubdir(testBundlePath, 'files')
+        ? {
+            files_path: toRelativeArtifactPath(
+              options.outputDir,
+              path.join(testBundlePath, 'files'),
+            ),
+          }
+        : {}),
+      ...(sourceRecord?.graders_path || hasCopiedSubdir(testBundlePath, 'graders')
+        ? {
+            graders_path: toRelativeArtifactPath(
+              options.outputDir,
+              path.join(testBundlePath, 'graders'),
+            ),
+          }
+        : {}),
+    };
+  };
+}
+
 export function buildProjectionBundleFromExportedIndex(options: {
   readonly sourceFile: string;
   readonly outputDir: string;
@@ -197,7 +298,7 @@ export const resultsExportCommand = command({
     const shouldIncludeRawContent = includeRawContent;
 
     try {
-      const { sourceFile, results } = await loadExportSource(source, cwd);
+      const { sourceFile, results, indexRecords } = await loadExportSource(source, cwd);
 
       const outputDir = out
         ? path.isAbsolute(out)
@@ -212,6 +313,7 @@ export const resultsExportCommand = command({
           cwd,
           duplicatePolicy: policy,
           includeRawContent: shouldIncludeRawContent,
+          indexRecords,
         });
 
       if (shouldDryRun) {
@@ -223,6 +325,11 @@ export const resultsExportCommand = command({
         evalFile: sourceFile,
         runId: deriveExportRunId(sourceFile),
         duplicatePolicy: policy,
+        additionalArtifacts: createExportBundleArtifactsWriter({
+          outputDir,
+          sourceBaseDir: path.dirname(sourceFile),
+          sourceRecordsByResult: buildSourceRecordMap(results, indexRecords ?? []),
+        }),
       });
 
       const bundlePath = shouldWriteProjectionBundle
diff --git a/apps/cli/test/commands/results/export.test.ts b/apps/cli/test/commands/results/export.test.ts
index 88cfa4564..da4df1a0c 100644
--- a/apps/cli/test/commands/results/export.test.ts
+++ b/apps/cli/test/commands/results/export.test.ts
@@ -572,6 +572,111 @@ describe('results export', () => {
     );
   });
 
+  it('exports generated test bundle refs and files from source manifests', async () => {
+    const sourceDir = path.join(tempDir, 'source-run');
+    mkdirSync(path.join(sourceDir, 'case', 'test'), { recursive: true });
+    writeFileSync(
+      path.join(sourceDir, 'case', 'test', 'EVAL.yaml'),
+      'tests:\n  - id: test-greeting\n',
+    );
+    writeFileSync(path.join(sourceDir, 'case', 'test', 'targets.yaml'), 'targets: []\n');
+    const sourceFile = path.join(sourceDir, RESULT_INDEX_FILENAME);
+    const outputDir = path.join(tempDir, 'output');
+    const content = toJsonl({
+      ...RESULT_FULL,
+      result_dir: 'case',
+      test_dir: 'case/test',
+      eval_path: 'case/test/EVAL.yaml',
+      targets_path: 'case/test/targets.yaml',
+    });
+
+    await exportResults(sourceFile, content, outputDir);
+
+    const [entry] = readIndex(outputDir);
+    expect(entry.test_dir).toBe(`${entry.result_dir}/test`);
+    expect(entry.task_dir).toBeUndefined();
+    expect(entry.eval_path).toBe(`${entry.result_dir}/test/EVAL.yaml`);
+    expect(entry.targets_path).toBe(`${entry.result_dir}/test/targets.yaml`);
+    expect(readFileSync(path.join(outputDir, entry.eval_path ?? ''), 'utf8')).toContain(
+      'test-greeting',
+    );
+
+    const bundle = buildProjectionBundleFromExportedIndex({
+      sourceFile,
+      outputDir,
+      cwd: tempDir,
+      includeRawContent: true,
+    });
+    expect(bundle.entries[0].artifact_refs).toMatchObject({
+      status: 'emitted',
+      test_dir: entry.test_dir,
+      eval_path: entry.eval_path,
+      targets_path: entry.targets_path,
+    });
+  });
+
+  it('exports legacy task_dir bundles as new test_dir artifacts', async () => {
+    const sourceDir = path.join(tempDir, 'legacy-run');
+    mkdirSync(path.join(sourceDir, 'case', 'task'), { recursive: true });
+    writeFileSync(
+      path.join(sourceDir, 'case', 'task', 'EVAL.yaml'),
+      'tests:\n  - id: test-greeting\n',
+    );
+    writeFileSync(path.join(sourceDir, 'case', 'task', 'targets.yaml'), 'targets: []\n');
+    const sourceFile = path.join(sourceDir, RESULT_INDEX_FILENAME);
+    const outputDir = path.join(tempDir, 'output');
+    const content = toJsonl({
+      ...RESULT_FULL,
+      result_dir: 'case',
+      task_dir: 'case/task',
+      eval_path: 'case/task/EVAL.yaml',
+      targets_path: 'case/task/targets.yaml',
+    });
+
+    await exportResults(sourceFile, content, outputDir);
+
+    const [entry] = readIndex(outputDir);
+    expect(entry.test_dir).toBe(`${entry.result_dir}/test`);
+    expect(entry.task_dir).toBeUndefined();
+    expect(entry.eval_path).toBe(`${entry.result_dir}/test/EVAL.yaml`);
+    expect(entry.targets_path).toBe(`${entry.result_dir}/test/targets.yaml`);
+    expect(readFileSync(path.join(outputDir, entry.eval_path ?? ''), 'utf8')).toContain(
+      'test-greeting',
+    );
+  });
+
+  it('preserves source bundle refs in dry-run projection inputs', async () => {
+    const sourceDir = path.join(tempDir, 'source-run');
+    mkdirSync(sourceDir, { recursive: true });
+    const sourceFile = path.join(sourceDir, RESULT_INDEX_FILENAME);
+    writeFileSync(
+      sourceFile,
+      toJsonl({
+        ...RESULT_FULL,
+        result_dir: 'case',
+        task_dir: 'case/task',
+        eval_path: 'case/task/EVAL.yaml',
+        targets_path: 'case/task/targets.yaml',
+      }),
+    );
+
+    const source = await loadExportSource(sourceFile, tempDir);
+    const bundle = buildProjectionBundle(source.results, {
+      sourceFile: source.sourceFile,
+      runId: deriveExportRunId(source.sourceFile),
+      cwd: tempDir,
+      includeRawContent: true,
+      indexRecords: source.indexRecords,
+    });
+
+    expect(bundle.entries[0].artifact_refs).toMatchObject({
+      status: 'planned_export',
+      task_dir: 'case/task',
+      eval_path: 'case/task/EVAL.yaml',
+      targets_path: 'case/task/targets.yaml',
+    });
+  });
+
   it('fails duplicate projection artifacts when duplicate policy is error', async () => {
     const sourceFile = path.join(tempDir, 'runs', 'retry-run', 'index.jsonl');
     const outputDir = path.join(tempDir, 'output');