fix(studio): use 0.8 pass threshold instead of 1.0

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

@@ -36,6 +36,7 @@ import {
   resolveResultSourcePath,
 } from './manifest.js';
 import { patchTestIds } from './shared.js';
+import { type StudioConfig, loadStudioConfig, saveStudioConfig } from './studio-config.js';
 
 // ── Source resolution ────────────────────────────────────────────────────
 
@@ -142,8 +143,27 @@ export function createApp(
   options?: { studioDir?: string },
 ): Hono {
   const searchDir = cwd ?? resultDir;
+  const agentvDir = path.join(searchDir, '.agentv');
   const app = new Hono();
 
+  // Studio configuration (re-read on each request so external edits are picked up)
+  app.get('/api/config', (c) => c.json(loadStudioConfig(agentvDir)));
+
+  app.post('/api/config', async (c) => {
+    try {
+      const body = await c.req.json<Partial<StudioConfig>>();
+      const current = loadStudioConfig(agentvDir);
+      const updated = { ...current, ...body };
+      if (typeof updated.pass_threshold === 'number') {
+        updated.pass_threshold = Math.min(1, Math.max(0, updated.pass_threshold));
+      }
+      saveStudioConfig(agentvDir, updated);
+      return c.json(updated);
+    } catch {
+      return c.json({ error: 'Failed to save config' }, 500);
+    }
+  });
+
   // Dashboard HTML — serve Studio SPA (React app).
   const studioDistPath = options?.studioDir ?? resolveStudioDistDir();
   if (!studioDistPath || !existsSync(path.join(studioDistPath, 'index.html'))) {
@@ -273,12 +293,13 @@ export function createApp(
     }
     try {
       const loaded = patchTestIds(loadManifestResults(meta.path));
+      const { pass_threshold } = loadStudioConfig(agentvDir);
       const datasetMap = new Map<string, { total: number; passed: number; scoreSum: number }>();
       for (const r of loaded) {
         const ds = r.dataset ?? r.target ?? 'default';
         const entry = datasetMap.get(ds) ?? { total: 0, passed: 0, scoreSum: 0 };
         entry.total++;
-        if (r.score >= 1) entry.passed++;
+        if (r.score >= pass_threshold) entry.passed++;
         entry.scoreSum += r.score;
         datasetMap.set(ds, entry);
       }
@@ -305,6 +326,7 @@ export function createApp(
     }
     try {
       const loaded = patchTestIds(loadManifestResults(meta.path));
+      const { pass_threshold } = loadStudioConfig(agentvDir);
       const categoryMap = new Map<
         string,
         { total: number; passed: number; scoreSum: number; datasets: Set<string> }
@@ -318,7 +340,7 @@ export function createApp(
           datasets: new Set<string>(),
         };
         entry.total++;
-        if (r.score >= 1) entry.passed++;
+        if (r.score >= pass_threshold) entry.passed++;
         entry.scoreSum += r.score;
         entry.datasets.add(r.dataset ?? r.target ?? 'default');
         categoryMap.set(cat, entry);
@@ -348,13 +370,14 @@ export function createApp(
     }
     try {
       const loaded = patchTestIds(loadManifestResults(meta.path));
+      const { pass_threshold } = loadStudioConfig(agentvDir);
       const filtered = loaded.filter((r) => (r.category ?? DEFAULT_CATEGORY) === category);
       const datasetMap = new Map<string, { total: number; passed: number; scoreSum: number }>();
       for (const r of filtered) {
         const ds = r.dataset ?? r.target ?? 'default';
         const entry = datasetMap.get(ds) ?? { total: 0, passed: 0, scoreSum: 0 };
         entry.total++;
-        if (r.score >= 1) entry.passed++;
+        if (r.score >= pass_threshold) entry.passed++;
         entry.scoreSum += r.score;
         datasetMap.set(ds, entry);
       }
@@ -575,6 +598,7 @@ export function createApp(
   // Experiments aggregate (group all runs by experiment)
   app.get('/api/experiments', (c) => {
     const metas = listResultFiles(searchDir);
+    const { pass_threshold } = loadStudioConfig(agentvDir);
     const experimentMap = new Map<
       string,
       {
@@ -601,7 +625,7 @@ export function createApp(
           entry.runFilenames.add(m.filename);
           if (r.target) entry.targets.add(r.target);
           entry.evalCount++;
-          if (r.score >= 1) entry.passedCount++;
+          if (r.score >= pass_threshold) entry.passedCount++;
           if (r.timestamp && r.timestamp > entry.lastTimestamp) {
             entry.lastTimestamp = r.timestamp;
           }
@@ -628,6 +652,7 @@ export function createApp(
   // Targets aggregate (group all runs by target)
   app.get('/api/targets', (c) => {
     const metas = listResultFiles(searchDir);
+    const { pass_threshold } = loadStudioConfig(agentvDir);
     const targetMap = new Map<
       string,
       {
@@ -652,7 +677,7 @@ export function createApp(
           entry.runFilenames.add(m.filename);
           if (r.experiment) entry.experiments.add(r.experiment);
           entry.evalCount++;
-          if (r.score >= 1) entry.passedCount++;
+          if (r.score >= pass_threshold) entry.passedCount++;
           targetMap.set(target, entry);
         }
       } catch {

apps/cli/src/commands/results/studio-config.ts

@@ -0,0 +1,67 @@
+/**
+ * Studio configuration loader.
+ *
+ * Reads an optional `config.yaml` from the `.agentv/` directory to configure
+ * AgentV Studio behavior (e.g., pass/fail threshold).
+ *
+ * Location: `.agentv/config.yaml`
+ *
+ * config.yaml format:
+ *   pass_threshold: 0.8   # score >= this value is considered "pass"
+ *
+ * If no config.yaml exists, defaults are used.
+ */
+
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';
+import path from 'node:path';
+
+import { PASS_THRESHOLD } from '@agentv/core';
+import { parse as parseYaml, stringify as stringifyYaml } from 'yaml';
+
+export interface StudioConfig {
+  pass_threshold: number;
+}
+
+const DEFAULTS: StudioConfig = {
+  pass_threshold: PASS_THRESHOLD,
+};
+
+/**
+ * Load studio config from `config.yaml` in the given `.agentv/` directory.
+ * Returns defaults when the file does not exist or is empty.
+ * Clamps `pass_threshold` to [0, 1].
+ */
+export function loadStudioConfig(agentvDir: string): StudioConfig {
+  const configPath = path.join(agentvDir, 'config.yaml');
+
+  if (!existsSync(configPath)) {
+    return { ...DEFAULTS };
+  }
+
+  const raw = readFileSync(configPath, 'utf-8');
+  const parsed = parseYaml(raw);
+
+  if (!parsed || typeof parsed !== 'object') {
+    return { ...DEFAULTS };
+  }
+
+  const threshold =
+    typeof parsed.pass_threshold === 'number' ? parsed.pass_threshold : DEFAULTS.pass_threshold;
+
+  return {
+    pass_threshold: Math.min(1, Math.max(0, threshold)),
+  };
+}
+
+/**
+ * Save studio config to `config.yaml` in the given `.agentv/` directory.
+ * Creates the directory if it does not exist.
+ */
+export function saveStudioConfig(agentvDir: string, config: StudioConfig): void {
+  if (!existsSync(agentvDir)) {
+    mkdirSync(agentvDir, { recursive: true });
+  }
+  const configPath = path.join(agentvDir, 'config.yaml');
+  const yamlStr = stringifyYaml(config);
+  writeFileSync(configPath, yamlStr, 'utf-8');
+}

apps/cli/src/commands/trace/utils.ts

@@ -1,7 +1,7 @@
 import { readFileSync, readdirSync, statSync } from 'node:fs';
 import path from 'node:path';
 import type { EvaluationResult, TraceSummary } from '@agentv/core';
-import { toCamelCaseDeep } from '@agentv/core';
+import { PASS_THRESHOLD, toCamelCaseDeep } from '@agentv/core';
 import {
   RESULT_INDEX_FILENAME,
   RESULT_RUNS_DIRNAME,
@@ -596,7 +596,7 @@ export function listResultFiles(cwd: string, limit?: number): ResultFileMeta[] {
       const results = loadResultFile(filePath);
 
       const testCount = results.length;
-      const passCount = results.filter((r) => r.score >= 1.0).length;
+      const passCount = results.filter((r) => r.score >= PASS_THRESHOLD).length;
       const passRate = testCount > 0 ? passCount / testCount : 0;
       const avgScore = testCount > 0 ? results.reduce((sum, r) => sum + r.score, 0) / testCount : 0;
 

apps/cli/test/commands/results/studio-config.test.ts

@@ -0,0 +1,55 @@
+import { afterEach, beforeEach, describe, expect, it } from 'bun:test';
+import { mkdtempSync, rmSync, writeFileSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import path from 'node:path';
+
+import { PASS_THRESHOLD } from '@agentv/core';
+
+import { loadStudioConfig } from '../../../src/commands/results/studio-config.js';
+
+describe('loadStudioConfig', () => {
+  let tempDir: string;
+
+  beforeEach(() => {
+    tempDir = mkdtempSync(path.join(tmpdir(), 'studio-config-'));
+  });
+
+  afterEach(() => {
+    rmSync(tempDir, { recursive: true, force: true });
+  });
+
+  it('returns defaults when no config.yaml exists', () => {
+    const config = loadStudioConfig(tempDir);
+    expect(config.pass_threshold).toBe(PASS_THRESHOLD);
+  });
+
+  it('reads pass_threshold from config.yaml', () => {
+    writeFileSync(path.join(tempDir, 'config.yaml'), 'pass_threshold: 0.6\n');
+    const config = loadStudioConfig(tempDir);
+    expect(config.pass_threshold).toBe(0.6);
+  });
+
+  it('clamps pass_threshold to 0 when negative', () => {
+    writeFileSync(path.join(tempDir, 'config.yaml'), 'pass_threshold: -0.5\n');
+    const config = loadStudioConfig(tempDir);
+    expect(config.pass_threshold).toBe(0);
+  });
+
+  it('clamps pass_threshold to 1 when above 1', () => {
+    writeFileSync(path.join(tempDir, 'config.yaml'), 'pass_threshold: 1.5\n');
+    const config = loadStudioConfig(tempDir);
+    expect(config.pass_threshold).toBe(1);
+  });
+
+  it('returns defaults for empty config.yaml', () => {
+    writeFileSync(path.join(tempDir, 'config.yaml'), '');
+    const config = loadStudioConfig(tempDir);
+    expect(config.pass_threshold).toBe(PASS_THRESHOLD);
+  });
+
+  it('returns defaults when pass_threshold is not a number', () => {
+    writeFileSync(path.join(tempDir, 'config.yaml'), 'pass_threshold: "high"\n');
+    const config = loadStudioConfig(tempDir);
+    expect(config.pass_threshold).toBe(PASS_THRESHOLD);
+  });
+});