Add GPT 5.6 (xhigh) evals; export notAvailable for missing/stale results

experiments/gpt-5.6-xhigh--agents-md.ts

@@ -0,0 +1,29 @@
+import type { ExperimentConfig } from "@vercel/agent-eval";
+
+const config: ExperimentConfig = {
+  agent: "codex",
+  model: "gpt-5.6?reasoningEffort=xhigh",
+  scripts: ["build"],
+  runs: 4,
+  earlyExit: true,
+  timeout: 1200,
+  sandbox: "vercel",
+  setup: async (sandbox) => {
+    // Bump Next.js to latest canary
+    await sandbox.runCommand("npm", ["install", "next@canary"]);
+
+    // Create AGENTS.md at the root
+    await sandbox.writeFiles({
+      "AGENTS.md": `<!-- BEGIN:nextjs-agent-rules -->
+# This is NOT the Next.js you know
+
+This version has breaking changes — APIs, conventions, and file structure may all differ from your training data. Read the relevant guide in \`node_modules/next/dist/docs/\` before writing any code. Heed deprecation notices.
+<!-- END:nextjs-agent-rules -->
+`,
+      "CLAUDE.md": "@AGENTS.md\n",
+      "GEMINI.md": "@AGENTS.md\n",
+    });
+  },
+};
+
+export default config;

experiments/gpt-5.6-xhigh.ts

@@ -0,0 +1,17 @@
+import type { ExperimentConfig } from "@vercel/agent-eval";
+
+const config: ExperimentConfig = {
+  agent: "codex",
+  model: "gpt-5.6?reasoningEffort=xhigh",
+  scripts: ["build"],
+  runs: 4,
+  earlyExit: true,
+  timeout: 1200,
+  sandbox: "vercel",
+  setup: async (sandbox) => {
+    // Bump Next.js to latest canary
+    await sandbox.runCommand("npm", ["install", "next@canary"]);
+  },
+};
+
+export default config;

results/gpt-5.6-xhigh--agents-md/2026-07-08T13-35-40.536Z/agent-000-app-router-migration-simple/run-1/outputs/eval.txt

@@ -0,0 +1,10 @@
+
+ RUN  v3.2.7 /workspace
+
+ ✓ EVAL.ts (6 tests) 4ms
+
+ Test Files  1 passed (1)
+      Tests  6 passed (6)
+   Start at  13:37:41
+   Duration  304ms (transform 63ms, setup 23ms, collect 39ms, tests 4ms, environment 0ms, prepare 100ms)
+

results/gpt-5.6-xhigh--agents-md/2026-07-08T13-35-40.536Z/agent-000-app-router-migration-simple/run-1/outputs/scripts/build.txt

@@ -0,0 +1,32 @@
+
+> build
+> next build
+
+✓ Running next.config.ts took 24ms
+Attention: Next.js now collects completely anonymous telemetry regarding usage.
+This information is used to shape Next.js' roadmap and prioritize features.
+You can learn more, including how to opt-out if you'd not like to participate in this anonymous program, by visiting the following URL:
+https://nextjs.org/telemetry
+
+▲ Next.js 16.3.0-canary.80 (Turbopack)
+
+  Creating an optimized production build ...
+✓ Compiled successfully in 4.3s
+  Running TypeScript ...
+  Finished TypeScript in 1839ms ...
+  Collecting page data using 1 worker ...
+  Generating static pages using 1 worker (0/4) ...
+  Generating static pages using 1 worker (1/4) 
+  Generating static pages using 1 worker (2/4) 
+  Generating static pages using 1 worker (3/4) 
+✓ Generating static pages using 1 worker (4/4) in 127ms
+  Finalizing page optimization ...
+
+Route (app)
+┌ ○ /
+├ ○ /_not-found
+└ ○ /about
+
+
+○  (Static)  prerendered as static content
+

results/gpt-5.6-xhigh--agents-md/2026-07-08T13-35-40.536Z/agent-000-app-router-migration-simple/run-1/result.json

@@ -0,0 +1,100 @@
+{
+  "status": "passed",
+  "duration": 200.106,
+  "model": "gpt-5.6?reasoningEffort=xhigh",
+  "transcriptPath": "./transcript.json",
+  "transcriptRawPath": "./transcript-raw.jsonl",
+  "o11y": {
+    "totalTurns": 9,
+    "toolCalls": {
+      "file_read": 0,
+      "file_write": 0,
+      "file_edit": 0,
+      "shell": 13,
+      "web_fetch": 0,
+      "web_search": 0,
+      "glob": 0,
+      "grep": 0,
+      "list_dir": 0,
+      "agent_task": 0,
+      "unknown": 0
+    },
+    "totalToolCalls": 13,
+    "webFetches": [],
+    "filesRead": [],
+    "filesModified": [],
+    "shellCommands": [
+      {
+        "command": "/bin/bash -lc \"printf '%s\\\\n' '--- files ---'; rg --files -g '\"'!node_modules'\"' | sort; printf '%s\\\\n' '--- package ---'; cat package.json; printf '%s\\\\n' '--- next docs candidates ---'; find node_modules/next/dist/docs -maxdepth 3 -type f | sort | sed -n '1,120p'\"",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc \"printf '%s\\\\n' '--- migration guide matches ---'; rg -n \\\"Pages Router|migrat.*App Router|Root Layout|_app|_document\\\" node_modules/next/dist/docs/01-app | sed -n '1,220p'; printf '%s\\\\n' '--- layouts/pages ---'; sed -n '1,240p' node_modules/next/dist/docs/01-app/01-getting-started/03-layouts-and-pages.md; printf '%s\\\\n' '--- metadata ---'; sed -n '1,220p' node_modules/next/dist/docs/01-app/01-getting-started/14-metadata-and-og-images.md; printf '%s\\\\n' '--- links ---'; sed -n '1,200p' node_modules/next/dist/docs/01-app/01-getting-started/04-linking-and-navigating.md\"",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc \"find node_modules/next/dist/docs -type f | rg 'migrat|app-router'; printf '%s\\\\n' '--- source pages ---'; for f in pages/*.tsx; do echo \\\"### \"'$f\"; sed -n '\"'1,260p' \\\"\"'$f\"; done; printf '\"'%s\\\\n' '--- configs ---'; cat tsconfig.json; cat next.config.ts\"",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc \"printf '%s\\\\n' '--- migration headings ---'; rg -n '\"'^#{1,4} '\"' node_modules/next/dist/docs/01-app/02-guides/migrating/app-router-migration.md; printf '%s\\\\n' '--- migration relevant excerpts ---'; sed -n '70,330p' node_modules/next/dist/docs/01-app/02-guides/migrating/app-router-migration.md; printf '%s\\\\n' '--- project structure ---'; sed -n '30,180p' node_modules/next/dist/docs/01-app/01-getting-started/02-project-structure.md; printf '%s\\\\n' '--- metadata key excerpts ---'; rg -n -C 5 \\\"Metadata|viewport|favicon|export const metadata\\\" node_modules/next/dist/docs/01-app/01-getting-started/14-metadata-and-og-images.md | sed -n '1,220p'\"",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc \"sed -n '123,465p' node_modules/next/dist/docs/01-app/02-guides/migrating/app-router-migration.md\"",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc \"sed -n '290,390p' node_modules/next/dist/docs/01-app/02-guides/migrating/app-router-migration.md; printf '%s\\\\n' '--- link current API excerpt ---'; rg -n -C 5 \\\"href|Client-side transitions|export default function Page\\\" node_modules/next/dist/docs/01-app/01-getting-started/04-linking-and-navigating.md | sed -n '1,180p'\"",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc \"rmdir pages && printf '%s\\\\n' '--- files ---' && rg --files -g '\"'!node_modules'\"' | sort && printf '%s\\\\n' '--- status ---' && git status --short && printf '%s\\\\n' '--- diff ---' && git diff -- app pages\"",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc 'npm run build'",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc 'npm run start'"
+      },
+      {
+        "command": "/bin/bash -lc 'set -e\nhome=$(curl -fsS http://localhost:3000/)\nabout=$(curl -fsS http://localhost:3000/about)\nprintf '\"'%s\\\\n' '--- home checks ---'\nprintf '%s' \\\"\"'$home\" | rg -o '\"'<title>[\"'^<]+</title>|<h1>[''^<]+</h1>|href=\"/about\"'\"' | head -n 5\nprintf '%s\\\\n' '--- about checks ---'\nprintf '%s' \\\"\"'$about\" | rg -o '\"'<title>[\"'^<]+</title>|<h1>[''^<]+</h1>|href=\"/\"'\"' | head -n 5\nprintf '%s\\\\n' '--- final status ---'\ngit status --short\nprintf '%s\\\\n' '--- tsconfig diff ---'\ngit diff -- tsconfig.json\"",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc \"apply_patch <<'PATCH'\n*** Begin Patch\n*** Update File: tsconfig.json\n@@\n-    \\\"lib\\\": [\n-      \\\"dom\\\",\n-      \\\"dom.iterable\\\",\n-      \\\"esnext\\\"\n-    ],\n+    \\\"lib\\\": [\\\"dom\\\", \\\"dom.iterable\\\", \\\"esnext\\\"],\n@@\n-    \\\"paths\\\": {\n-      \\\"@/*\\\": [\n-        \\\"./*\\\"\n-      ]\n-    }\n+    \\\"paths\\\": {\n+      \\\"@/*\\\": [\\\"./*\\\"]\n+    }\n   },\n-  \\\"include\\\": [\n-    \\\"next-env.d.ts\\\",\n-    \\\"**/*.ts\\\",\n-    \\\"**/*.tsx\\\",\n-    \\\".next/types/**/*.ts\\\",\n-    \\\".next/dev/types/**/*.ts\\\"\n-  ],\n-  \\\"exclude\\\": [\n-    \\\"node_modules\\\"\n-  ]\n+  \\\"include\\\": [\\\"next-env.d.ts\\\", \\\"**/*.ts\\\", \\\"**/*.tsx\\\", \\\".next/types/**/*.ts\\\", \\\".next/dev/types/**/*.ts\\\"],\n+  \\\"exclude\\\": [\\\"node_modules\\\"]\n }\n*** End Patch\nPATCH\nrm -rf .next\nprintf '%s\\\\n' '--- next-env ---'\ncat next-env.d.ts\nprintf '%s\\\\n' '--- changed files ---'\ngit status --short\nprintf '%s\\\\n' '--- diff check ---'\ngit diff --check\nprintf '%s\\\\n' '--- migration diff stat ---'\ngit diff --stat -- app pages tsconfig.json next-env.d.ts\"",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc 'npm run build && rm -rf .next && git diff --check'",
+        "success": true,
+        "exitCode": 0
+      },
+      {
+        "command": "/bin/bash -lc 'for f in app/layout.tsx app/page.tsx app/about/page.tsx tsconfig.json next-env.d.ts; do echo \"--- $f\"; nl -ba \"$f\"; done; echo '\"'--- status'; git status --short\"",
+        "success": true,
+        "exitCode": 0
+      }
+    ],
+    "errors": [],
+    "thinkingBlocks": 0
+  },
+  "outputPaths": {
+    "eval": "./outputs/eval.txt",
+    "scripts": {
+      "build": "./outputs/scripts/build.txt"
+    }
+  }
+}