feat(scripts): deploy example eval results repos (#1267)

* feat(scripts): deploy example eval results repos

* fix(studio): clarify synced remote runs

* fix(studio): dedupe synced runs in detail views

* fix(results): store remote runs in agentv results layout

Author: christso

apps/cli/src/commands/results/remote.ts

@@ -290,7 +290,10 @@ export async function ensureRemoteRunAvailable(
     throw new Error(`Remote manifest path is outside the results repo clone: ${meta.path}`);
   }
 
-  const relativeRunPath = path.posix.relative('runs', path.posix.dirname(relativeManifestPath));
+  const relativeRunPath = path.posix.relative(
+    '.agentv/results/runs',
+    path.posix.dirname(relativeManifestPath),
+  );
   await materializeGitRun(config.path, relativeRunPath);
 }
 

apps/cli/src/commands/results/serve.ts

@@ -376,12 +376,14 @@ async function handleRuns(c: C, { searchDir, agentvDir }: DataContext) {
       let target: string | undefined;
       let experiment = inferExperimentFromRunId(m.raw_filename);
       let passRate = m.passRate;
+      let avgScore = m.avgScore;
       try {
         const records = await loadLightweightResultsForMeta(searchDir, m);
         if (records.length > 0) {
           target = records[0].target;
           experiment = records[0].experiment ?? experiment;
           passRate = records.filter((r) => r.score >= passThreshold).length / records.length;
+          avgScore = records.reduce((sum, r) => sum + r.score, 0) / records.length;
         } else {
           // Run is in-progress with 0 results written yet — fall back to the
           // in-memory target stored when the Studio launched this run.
@@ -402,7 +404,7 @@ async function handleRuns(c: C, { searchDir, agentvDir }: DataContext) {
         timestamp: m.timestamp,
         test_count: m.testCount,
         pass_rate: passRate,
-        avg_score: m.avgScore,
+        avg_score: avgScore,
         size_bytes: m.sizeBytes,
         source: m.source,
         ...(target && { target }),

apps/cli/test/commands/results/serve.test.ts

@@ -107,7 +107,7 @@ function writeRemoteRunArtifact(
     /^(\d{4}-\d{2}-\d{2})T(\d{2})-(\d{2})-(\d{2})-(\d{3})Z$/,
     '$1T$2:$3:$4.$5Z',
   );
-  const runDir = path.join(cloneDir, 'runs', experiment, timestamp);
+  const runDir = path.join(cloneDir, '.agentv', 'results', 'runs', experiment, timestamp);
   mkdirSync(runDir, { recursive: true });
   writeFileSync(path.join(runDir, 'index.jsonl'), toJsonl(resultRecord));
   writeFileSync(
@@ -653,6 +653,8 @@ describe('serve app', () => {
           process.env.AGENTV_HOME,
           'results',
           'EntityProcess-agentv-evals',
+          '.agentv',
+          'results',
           'runs',
           'default',
           '2026-03-26T10-00-00-000Z',
@@ -705,14 +707,21 @@ describe('serve app', () => {
       const listRes = await app.request('/api/runs');
       expect(listRes.status).toBe(200);
       const listData = (await listRes.json()) as {
-        runs: Array<{ filename: string; source: string; experiment?: string; pass_rate?: number }>;
+        runs: Array<{
+          filename: string;
+          source: string;
+          experiment?: string;
+          pass_rate?: number;
+          avg_score?: number;
+        }>;
       };
       expect(listData.runs).toHaveLength(1);
       expect(listData.runs[0]).toMatchObject({
         filename: `remote::${runId}`,
         source: 'remote',
         experiment: 'green-uat',
         pass_rate: 1,
+        avg_score: 1,
       });
 
       const detailRes = await app.request(
@@ -749,6 +758,8 @@ describe('serve app', () => {
 
       const runManifestPath = path.join(
         cloneDir,
+        '.agentv',
+        'results',
         'runs',
         'external-sync',
         '2026-03-26T11-00-00-000Z',

apps/studio/src/components/ExperimentDetail.tsx

@@ -13,6 +13,7 @@ import {
   projectRunListOptions,
   runListOptions,
 } from '~/lib/api';
+import { dedupeSyncedRuns } from '~/lib/run-dedupe';
 
 import { RunList } from './RunList';
 
@@ -45,12 +46,12 @@ export function ExperimentDetail({ experimentName, projectId }: ExperimentDetail
   }
 
   const experiment = experimentsData?.experiments?.find((entry) => entry.name === experimentName);
-  const runs = (runListData?.runs ?? []).filter(
-    (run) => (run.experiment ?? 'default') === experimentName,
+  const runs = dedupeSyncedRuns(
+    (runListData?.runs ?? []).filter((run) => (run.experiment ?? 'default') === experimentName),
   );
 
   const passRate = experiment?.pass_rate ?? 0;
-  const runCount = experiment?.run_count ?? runs.length;
+  const runCount = runs.length;
   const targetCount = experiment?.target_count ?? 0;
 
   return (

apps/studio/src/components/TargetsTab.tsx

@@ -15,6 +15,7 @@ import {
   runListOptions,
   targetsOptions,
 } from '~/lib/api';
+import { dedupeSyncedRuns } from '~/lib/run-dedupe';
 import type { RunMeta, TargetsResponse } from '~/lib/types';
 
 import { PassRatePill } from './PassRatePill';
@@ -68,7 +69,7 @@ export function TargetsTab({ projectId }: TargetsTabProps = {}) {
     }
 
     return [...groups.entries()]
-      .map(([name, experimentRuns]) => buildExperimentGroup(name, experimentRuns))
+      .map(([name, experimentRuns]) => buildExperimentGroup(name, dedupeSyncedRuns(experimentRuns)))
       .sort((a, b) => {
         if (a.latestTimestamp && b.latestTimestamp && a.latestTimestamp !== b.latestTimestamp) {
           return b.latestTimestamp.localeCompare(a.latestTimestamp);

apps/studio/src/lib/run-dedupe.test.ts

@@ -0,0 +1,33 @@
+import { describe, expect, it } from 'bun:test';
+
+import { dedupeSyncedRuns } from './run-dedupe';
+import type { RunMeta } from './types';
+
+function run(filename: string, source: RunMeta['source']): RunMeta {
+  return {
+    filename,
+    display_name: filename,
+    path: `/tmp/${filename}`,
+    timestamp: '2026-05-28T08:21:09.063Z',
+    test_count: 8,
+    pass_rate: 1,
+    avg_score: 1,
+    size_bytes: 1024,
+    source,
+  };
+}
+
+describe('dedupeSyncedRuns', () => {
+  it('collapses local and remote copies of the same run in all-runs views', () => {
+    const runs = [
+      run('remote::2026-05-28T08-21-09-063Z', 'remote'),
+      run('2026-05-28T08-21-09-063Z', 'local'),
+      run('remote::2026-05-27T08-21-09-063Z', 'remote'),
+    ];
+
+    expect(dedupeSyncedRuns(runs).map((r) => r.filename)).toEqual([
+      '2026-05-28T08-21-09-063Z',
+      'remote::2026-05-27T08-21-09-063Z',
+    ]);
+  });
+});

apps/studio/src/lib/run-dedupe.ts

@@ -0,0 +1,23 @@
+import type { RunMeta } from './types';
+
+const REMOTE_RUN_PREFIX = 'remote::';
+
+function canonicalRunId(filename: string): string {
+  return filename.startsWith(REMOTE_RUN_PREFIX)
+    ? filename.slice(REMOTE_RUN_PREFIX.length)
+    : filename;
+}
+
+export function dedupeSyncedRuns(runs: readonly RunMeta[]): RunMeta[] {
+  const byRunId = new Map<string, RunMeta>();
+
+  for (const run of runs) {
+    const key = canonicalRunId(run.filename);
+    const existing = byRunId.get(key);
+    if (!existing || (existing.source === 'remote' && run.source === 'local')) {
+      byRunId.set(key, run);
+    }
+  }
+
+  return [...byRunId.values()];
+}

apps/studio/src/routes/index.tsx

@@ -33,6 +33,7 @@ import {
   resolveIndexRoute,
   resolveInitialProjectRedirect,
 } from '~/lib/navigation';
+import { dedupeSyncedRuns } from '~/lib/run-dedupe';
 import type { RunMeta } from '~/lib/types';
 type TabId = StudioTabId;
 
@@ -234,7 +235,7 @@ function SingleProjectHome() {
   const activeTab: TabId = tabs.some((t) => t.id === tab) ? (tab as TabId) : 'experiments';
   const filteredRuns =
     sourceFilter === 'all'
-      ? (data?.runs ?? [])
+      ? dedupeSyncedRuns(data?.runs ?? [])
       : (data?.runs ?? []).filter((run) => run.source === sourceFilter);
 
   async function handleSyncRemote() {

apps/studio/src/routes/projects/$projectId.tsx

@@ -22,6 +22,7 @@ import {
   useRemoteStatus,
   useStudioConfig,
 } from '~/lib/api';
+import { dedupeSyncedRuns } from '~/lib/run-dedupe';
 
 type TabId = 'runs' | 'experiments' | 'analytics' | 'targets';
 
@@ -121,7 +122,7 @@ function ProjectRunsTab({ projectId }: { projectId: string }) {
 
   const filteredRuns =
     sourceFilter === 'all'
-      ? (data?.runs ?? [])
+      ? dedupeSyncedRuns(data?.runs ?? [])
       : (data?.runs ?? []).filter((run) => run.source === sourceFilter);
 
   async function handleSyncRemote() {

packages/core/src/evaluation/results-repo.ts

@@ -17,6 +17,8 @@ import { getAgentvHome } from '../paths.js';
 import type { ResultsConfig } from './loaders/config-loader.js';
 
 const execFileAsync = promisify(execFile);
+const RESULTS_REPO_RESULTS_DIR = '.agentv/results';
+const RESULTS_REPO_RUNS_DIR = `${RESULTS_REPO_RESULTS_DIR}/runs`;
 
 export interface ResultsRepoLocalPaths {
   readonly rootDir: string;
@@ -345,7 +347,7 @@ export async function stageResultsArtifacts(params: {
 
 export function resolveResultsRepoRunsDir(config: ResultsConfig): string {
   const normalized = normalizeResultsConfig(config);
-  return path.join(normalized.path, 'runs');
+  return path.join(normalized.path, RESULTS_REPO_RESULTS_DIR, 'runs');
 }
 
 export async function directorySizeBytes(targetPath: string): Promise<number> {
@@ -443,7 +445,12 @@ export async function directPushResults(params: {
   const baseBranch = await resolveDefaultBranch(repoDir);
   await fetchResultsRepo(repoDir);
 
-  const destinationDir = path.join(repoDir, 'runs', params.destinationPath);
+  const destinationDir = path.join(
+    repoDir,
+    RESULTS_REPO_RESULTS_DIR,
+    'runs',
+    params.destinationPath,
+  );
   await stageResultsArtifacts({
     repoDir,
     sourceDir: params.sourceDir,
@@ -655,9 +662,12 @@ function parseGitBatchBlobs(output: Buffer): GitBatchBlob[] {
 }
 
 export async function listGitRuns(repoDir: string, ref = 'origin/main'): Promise<GitListedRun[]> {
-  const { stdout: treeOut } = await runGit(['ls-tree', '-r', '--name-only', ref, 'runs'], {
-    cwd: repoDir,
-  });
+  const { stdout: treeOut } = await runGit(
+    ['ls-tree', '-r', '--name-only', ref, RESULTS_REPO_RUNS_DIR],
+    {
+      cwd: repoDir,
+    },
+  );
 
   const benchmarkPaths = treeOut
     .split(/\r?\n/)
@@ -679,7 +689,7 @@ export async function listGitRuns(repoDir: string, ref = 'origin/main'): Promise
     const benchmarkPath = benchmarkPaths[index];
     const benchmark = JSON.parse(blob.content.toString('utf8')) as GitRunBenchmark;
     const runDir = path.posix.dirname(benchmarkPath);
-    const relativeRunPath = path.posix.relative('runs', runDir);
+    const relativeRunPath = path.posix.relative(RESULTS_REPO_RUNS_DIR, runDir);
     const runId = buildGitRunId(relativeRunPath);
     const timestamp = benchmark.metadata?.timestamp?.trim() || path.posix.basename(runDir);
     const targets = benchmark.metadata?.targets ?? [];
@@ -712,7 +722,7 @@ export async function materializeGitRun(
   ref = 'origin/main',
 ): Promise<void> {
   const normalizedRunPath = relativeRunPath.split(path.sep).join('/');
-  const runTreePath = path.posix.join('runs', normalizedRunPath);
+  const runTreePath = path.posix.join(RESULTS_REPO_RUNS_DIR, normalizedRunPath);
   const targetRunDir = path.join(repoDir, ...runTreePath.split('/'));
   const { stdout: treeOut } = await runGit(['ls-tree', '-r', '--name-only', ref, runTreePath], {
     cwd: repoDir,