pi-chrome: snapshot pre-migration state

23 files changed, 1918 insertions, 87 deletions

diff --git a/test-suite/README.md b/test-suite/README.md
index bb8becb..a96e0d0 100644
--- a/test-suite/README.md
+++ b/test-suite/README.md
@@ -1,13 +1,14 @@
 # pi-chrome browser-control benchmark
 
-Static benchmark pages for evaluating tools that let agents control Chrome. The
-suite covers two related questions:
+Static benchmark pages for evaluating tools that let agents control Chrome. The suite has two layers:
 
-1. **Can the agent complete normal browser work?** Forms, scroll containers,
-   contenteditable, files, frames, Shadow DOM, network/console inspection.
-2. **Does the interaction look like a human/browser action when sites care?**
-   `isTrusted`, user activation, pointer paths, key cadence, native controls,
-   drag/drop, touch, paste, and scroll momentum.
+1. **Unit challenges** (`manifest.json`) — MiniWoB-style capability probes for
+   forms, scroll containers, contenteditable, files, frames, Shadow DOM,
+   network/console inspection, `isTrusted`, user activation, pointer paths, key
+   cadence, native controls, drag/drop, touch, paste, and scroll momentum.
+2. **Long-horizon hermetic tasks** (`task-manifest.json`) — WebArena /
+   BrowserGym-inspired multi-step tasks with fresh run IDs and deterministic
+   programmatic graders.
 
 ## Run
 
@@ -24,11 +25,17 @@ Each challenge page exposes:
 - `window.__reason` — array of reasons
 - `window.__events` — raw event log for forensics
 
-`manifest.json` is the source of truth for benchmark metadata: category, goal,
-expected result per mode, prerequisites, flake risk, manual baseline status, and
-canonical tool recipe. `manifest.schema.json` documents the manifest shape.
-Recipes express tool intent; runners may need to adapt descriptive selectors
-(e.g. shadow/iframe notation) and expand path placeholders like `$PWD`.
+`manifest.json` is the source of truth for unit-challenge metadata: category,
+goal, expected result per mode, prerequisites, flake risk, manual baseline
+status, and canonical tool recipe. `manifest.schema.json` documents the manifest
+shape. Recipes express tool intent; runners may need to adapt descriptive
+selectors (e.g. shadow/iframe notation) and expand path placeholders like `$PWD`.
+
+`task-manifest.json` is the source of truth for long-horizon tasks: BrowserGym-style
+`taskId`, seed, viewport, goal object, difficulty tier, max steps, declared
+action subsets, reset/setup URL, validate hook, optional cheat recipe, and
+programmatic grader expression. `task-manifest.schema.json` documents this shape.
+`browsergym-action-space.json` records BrowserGym-compatible action subsets.
 
 ## Modes / expected outcomes
 
@@ -49,7 +56,7 @@ Expected values in `manifest.json`:
 Manual baselines are tracked separately with `manualBaseline`. `unverified`
 means the manual expectation is a target, not a recorded contract.
 
-## Recommended agent flow
+## Recommended unit-challenge agent flow
 
 1. Navigate to dashboard:
    `http://127.0.0.1:8765/`.
@@ -70,6 +77,29 @@ means the manual expectation is a target, not a recorded contract.
 4. Return to dashboard and compare actual verdicts with expected values.
 5. Copy JSON report from dashboard for PRs or regression notes.
 
+## Recommended long-horizon task flow
+
+1. Load `task-manifest.json`.
+2. Replace `$RUN_ID` in `startUrl` with a fresh value.
+3. Navigate to the start URL and read the visible task instruction.
+4. Solve using normal browser tools only; avoid direct state mutation unless the
+   benchmark mode explicitly allows evaluate-based actions.
+5. Click **Grade now** or evaluate the task grader expression:
+   ```js
+   JSON.stringify({ v: window.__taskVerdict, r: window.__taskReason })
+   ```
+6. Record action count, observations used, tools used, verdict, and reason.
+
+## Design principles copied from browser-agent benchmarks
+
+- Prefer hermetic sites and deterministic graders over live sites and LLM judges.
+- Report action API and observation format; these strongly affect scores.
+- Use difficulty tiers: L1 atomic, L2 compositional, L3 cross-page/context-rich.
+- Include tedious cross-page memory and exact-value transfer tasks; short unit
+  probes hide these failures.
+- Keep synthetic-event-gated tests because extension bridges face failures that
+  CDP/Playwright-style benchmarks usually do not measure.
+
 ## Challenge categories
 
 - `trusted-input` — browser-trusted click/key events.
@@ -136,3 +166,7 @@ The dashboard renders this from `manifest.json`. In brief:
   software WebGL in VMs, remote desktops, or policy-constrained environments.
 - `notes/bypass-ideas.md` is historical guidance for older synthetic-only
   versions. Prefer `manifest.json` for current expected outcomes.
+- `notes/browsergym-compat.md` defines the reset/step/validate/observation/BID
+  contract for external BrowserGym-style agents.
+- `notes/runner-spec.md`, `notes/scoring.md`, and `notes/profiles.md` define
+  runner output, scoring, retry policy, and environment metadata.
diff --git a/test-suite/_lib.js b/test-suite/_lib.js
index 95d6393..f3845e5 100644
--- a/test-suite/_lib.js
+++ b/test-suite/_lib.js
@@ -8,6 +8,8 @@
     verdict: "PENDING",
     reason: [],
     events,
+    details: [],
+    thresholds: {},
   };
 
   function render() {
@@ -31,8 +33,9 @@
   }
 
   const Challenge = {
-    init({ id, instructions }) {
+    init({ id, instructions, thresholds = {} }) {
       state.id = id;
+      state.thresholds = parseThresholds(thresholds);
       document.title = `[${state.verdict}] ${id}`;
       const root = document.body;
       const bar = document.createElement("div");
@@ -82,6 +85,20 @@
       document.title = `[WARN] ${state.id}`;
       persist(); render();
     },
+    partial({ name, pass, reason, data }) {
+      state.details.push({ name, pass: !!pass, reason: reason || "", data });
+      log("partial", { name, pass: !!pass, reason, data });
+      persist(); render();
+      return !!pass;
+    },
+    finishPartials() {
+      const failed = state.details.filter(d => !d.pass);
+      if (failed.length) Challenge.fail(...failed.map(d => `${d.name}: ${d.reason || "failed"}`));
+      else Challenge.pass(...state.details.map(d => `${d.name}: ok`));
+    },
+    getThreshold(name, fallback) {
+      return Object.prototype.hasOwnProperty.call(state.thresholds, name) ? state.thresholds[name] : fallback;
+    },
     log,
     state,
   };
@@ -90,10 +107,24 @@
     try {
       localStorage.setItem(
         "pi-chrome-suite:" + state.id,
-        JSON.stringify({ id: state.id, verdict: state.verdict, reason: state.reason, events: state.events.slice(-50), ts: Date.now() })
+        JSON.stringify({ id: state.id, verdict: state.verdict, reason: state.reason, details: state.details, thresholds: state.thresholds, events: state.events.slice(-50), ts: Date.now() })
       );
     } catch {}
   }
 
+  function parseThresholds(defaults) {
+    const out = { ...defaults };
+    try {
+      const qs = new URLSearchParams(location.search);
+      for (const [k, v] of qs) {
+        if (!k.startsWith("threshold.")) continue;
+        const key = k.slice("threshold.".length);
+        const num = Number(v);
+        out[key] = Number.isFinite(num) ? num : v;
+      }
+    } catch {}
+    return out;
+  }
+
   window.Challenge = Challenge;
 })();
diff --git a/test-suite/browsergym-action-space.json b/test-suite/browsergym-action-space.json
new file mode 100644
index 0000000..5d7c873
--- /dev/null
+++ b/test-suite/browsergym-action-space.json
@@ -0,0 +1,44 @@
+{
+  "subsets": {
+    "chat": ["send_msg_to_user"],
+    "infeas": ["report_infeasible"],
+    "bid": ["click", "dblclick", "hover", "fill", "clear", "press", "focus", "select_option", "scroll", "drag_and_drop", "upload_file"],
+    "coord": ["mouse_move", "mouse_up", "mouse_down", "mouse_click", "mouse_dblclick", "mouse_drag_and_drop", "mouse_upload_file", "scroll_at", "keyboard_down", "keyboard_up", "keyboard_press", "keyboard_type", "keyboard_insert_text"],
+    "nav": ["go_back", "go_forward", "goto"],
+    "tab": ["tab_close", "tab_focus", "new_tab"]
+  },
+  "signatures": {
+    "fill": "fill(bid, value, enable_autocomplete_menu=false)",
+    "click": "click(bid, button='left', modifiers=[])",
+    "dblclick": "dblclick(bid, button='left', modifiers=[])",
+    "hover": "hover(bid)",
+    "press": "press(bid, key_comb)",
+    "focus": "focus(bid)",
+    "clear": "clear(bid)",
+    "select_option": "select_option(bid, options)",
+    "scroll": "scroll(delta_x, delta_y)",
+    "drag_and_drop": "drag_and_drop(from_bid, to_bid)",
+    "upload_file": "upload_file(bid, file)",
+    "mouse_move": "mouse_move(x, y)",
+    "mouse_click": "mouse_click(x, y, button='left')",
+    "mouse_dblclick": "mouse_dblclick(x, y, button='left')",
+    "mouse_up": "mouse_up(x, y, button='left')",
+    "mouse_down": "mouse_down(x, y, button='left')",
+    "mouse_drag_and_drop": "mouse_drag_and_drop(from_x, from_y, to_x, to_y)",
+    "scroll_at": "scroll_at(x, y, dx, dy)",
+    "keyboard_press": "keyboard_press(key)",
+    "keyboard_down": "keyboard_down(key)",
+    "keyboard_up": "keyboard_up(key)",
+    "keyboard_type": "keyboard_type(text)",
+    "keyboard_insert_text": "keyboard_insert_text(text)",
+    "goto": "goto(url)",
+    "go_back": "go_back()",
+    "go_forward": "go_forward()",
+    "tab_close": "tab_close()",
+    "tab_focus": "tab_focus(index)",
+    "new_tab": "new_tab()",
+    "send_msg_to_user": "send_msg_to_user(text)",
+    "report_infeasible": "report_infeasible(reason)",
+    "noop": "noop(wait_ms=1000)"
+  }
+}
diff --git a/test-suite/challenges/34-dialog-handling.html b/test-suite/challenges/34-dialog-handling.html
new file mode 100644
index 0000000..25845a7
--- /dev/null
+++ b/test-suite/challenges/34-dialog-handling.html
@@ -0,0 +1,28 @@
+<!doctype html>
+<meta charset="utf-8">
+<title>34 dialog handling</title>
+<link rel="stylesheet" href="../_style.css">
+<script src="../_lib.js"></script>
+<body>
+<main>
+  <p>Goal: handle browser dialogs. Click Prompt, enter <code>pi-chrome</code>, accept it; then click Confirm and accept it. Browser automation must expose/dismiss native dialogs.</p>
+  <button id="promptBtn">Prompt</button>
+  <button id="confirmBtn">Confirm</button>
+  <button id="alertBtn">Alert smoke</button>
+  <pre id="rep" class="code"></pre>
+</main>
+<script>
+Challenge.init({ id: "dialog-handling", instructions: "accept prompt('pi-chrome') and confirm()" });
+const state = { prompt: null, confirm: null, alert: false };
+function paint(){ rep.textContent = JSON.stringify(state, null, 2); }
+function check(){
+  paint();
+  if (state.prompt === "pi-chrome" && state.confirm === true) Challenge.pass("prompt and confirm handled with expected values");
+}
+promptBtn.addEventListener("click", () => { state.prompt = prompt("Enter token", ""); Challenge.log("prompt-result", { value: state.prompt }); check(); });
+confirmBtn.addEventListener("click", () => { state.confirm = confirm("Accept benchmark confirm?"); Challenge.log("confirm-result", { value: state.confirm }); check(); });
+alertBtn.addEventListener("click", () => { alert("pi-chrome alert smoke"); state.alert = true; Challenge.log("alert-dismissed", {}); paint(); });
+window.addEventListener("beforeunload", e => { e.preventDefault(); e.returnValue = "benchmark beforeunload"; });
+paint();
+</script>
+</body>
diff --git a/test-suite/challenges/35-target-blank-popup.html b/test-suite/challenges/35-target-blank-popup.html
new file mode 100644
index 0000000..0813625
--- /dev/null
+++ b/test-suite/challenges/35-target-blank-popup.html
@@ -0,0 +1,35 @@
+<!doctype html>
+<meta charset="utf-8">
+<title>35 target blank popup</title>
+<link rel="stylesheet" href="../_style.css">
+<script src="../_lib.js"></script>
+<body>
+<main>
+  <p>Goal: click <code>target=_blank</code> link, detect new tab, switch to it, and verify child page reports PASS.</p>
+  <a id="open" target="_blank" rel="opener" href="35-target-blank-popup.html?child=1" style="font-size:18px">Open child tab</a>
+  <p id="msg"></p>
+</main>
+<script>
+Challenge.init({ id: "target-blank-popup", instructions: "open child tab and inspect it" });
+const qs = new URLSearchParams(location.search);
+if (qs.get("child") === "1") {
+  document.getElementById("msg").textContent = "Child tab opened.";
+  Challenge.pass("child tab loaded on same origin");
+  try { localStorage.setItem("pi-chrome-suite:target-blank-popup", JSON.stringify({ id:"target-blank-popup", verdict:"PASS", reason:["✓ child tab loaded"], ts: Date.now() })); } catch {}
+} else {
+  document.getElementById("msg").textContent = "Parent page. Link should open new tab.";
+  function maybePassFromStorage(raw) {
+    if (!raw) return;
+    try {
+      const v = JSON.parse(raw);
+      if (v.verdict === "PASS") Challenge.pass("child reported PASS via storage event/localStorage");
+    } catch {}
+  }
+  window.addEventListener("storage", e => {
+    if (e.key === "pi-chrome-suite:target-blank-popup") maybePassFromStorage(e.newValue);
+  });
+  maybePassFromStorage(localStorage.getItem("pi-chrome-suite:target-blank-popup"));
+  document.getElementById("open").addEventListener("click", e => Challenge.log("popup-click", { trusted: e.isTrusted }));
+}
+</script>
+</body>
diff --git a/test-suite/challenges/36-modal-focus-trap.html b/test-suite/challenges/36-modal-focus-trap.html
new file mode 100644
index 0000000..811207c
--- /dev/null
+++ b/test-suite/challenges/36-modal-focus-trap.html
@@ -0,0 +1,49 @@
+<!doctype html>
+<meta charset="utf-8">
+<title>36 modal focus trap</title>
+<link rel="stylesheet" href="../_style.css">
+<script src="../_lib.js"></script>
+<body>
+<main>
+  <p>Goal: open modal, use Tab to cycle focus inside modal only, then Escape closes it and focus returns to opener.</p>
+  <button id="open">Open modal</button>
+  <div id="modal" role="dialog" aria-modal="true" aria-label="Focus trap modal" hidden style="position:fixed;inset:20%;background:#222;border:2px solid #6cf;border-radius:10px;padding:20px;z-index:5">
+    <button id="first">First</button>
+    <input id="middle" aria-label="Modal input">
+    <button id="last">Last</button>
+  </div>
+</main>
+<script>
+Challenge.init({ id: "modal-focus-trap", instructions: "open modal, Tab cycles inside, Escape closes" });
+const order = [];
+const modal = document.getElementById("modal");
+const nodes = [first, middle, last];
+let escaped = false, modalOpened = false, outsideFocusWhileOpen = false;
+for (const el of [open, ...nodes]) el.addEventListener("focus", () => { order.push(el.id); Challenge.log("focus", { id: el.id }); });
+document.addEventListener("focusin", e => {
+  if (modalOpened && !modal.hidden && !modal.contains(e.target)) outsideFocusWhileOpen = true;
+});
+open.addEventListener("click", () => { modal.hidden = false; modalOpened = true; first.focus(); });
+document.addEventListener("keydown", e => {
+  if (modal.hidden) return;
+  if (e.key === "Tab") {
+    const i = nodes.indexOf(document.activeElement);
+    if (i >= 0) { e.preventDefault(); nodes[(i + (e.shiftKey ? -1 : 1) + nodes.length) % nodes.length].focus(); }
+  }
+  if (e.key === "Escape") { modal.hidden = true; escaped = true; open.focus(); queueMicrotask(check); }
+});
+function hasAdjacent(seq, a, b) {
+  return seq.some((x, i) => x === a && seq[i + 1] === b);
+}
+function check(){
+  const afterOpen = order.slice(order.indexOf("first"));
+  const bad = [];
+  if (!escaped) bad.push("Escape did not close modal");
+  if (document.activeElement !== open) bad.push("focus did not return to opener");
+  if (!afterOpen.includes("middle") || !afterOpen.includes("last")) bad.push(`focus cycle incomplete: ${afterOpen.join(">")}`);
+  if (!hasAdjacent(afterOpen, "last", "first") && !hasAdjacent(afterOpen, "first", "last")) bad.push(`no wrap-around focus trap observed: ${afterOpen.join(">")}`);
+  if (outsideFocusWhileOpen) bad.push("focus escaped outside modal while open");
+  if (bad.length) Challenge.fail(...bad); else Challenge.pass(`modal focus trap ok: ${afterOpen.join(">")}`);
+}
+</script>
+</body>
diff --git a/test-suite/challenges/37-autocomplete-combobox.html b/test-suite/challenges/37-autocomplete-combobox.html
new file mode 100644
index 0000000..91f35c2
--- /dev/null
+++ b/test-suite/challenges/37-autocomplete-combobox.html
@@ -0,0 +1,42 @@
+<!doctype html>
+<meta charset="utf-8">
+<title>37 autocomplete combobox</title>
+<link rel="stylesheet" href="../_style.css">
+<script src="../_lib.js"></script>
+<body>
+<main>
+  <p>Goal: type <code>as</code>, use ArrowDown then Enter to choose <code>Aster Lamp</code> from ARIA combobox.</p>
+  <label>Product <input id="combo" role="combobox" aria-autocomplete="list" aria-controls="list" aria-expanded="false" autocomplete="off"></label>
+  <ul id="list" role="listbox" style="border:1px solid #555;max-width:260px;padding:6px;display:none"></ul>
+</main>
+<script>
+Challenge.init({ id: "autocomplete-combobox", instructions: "type 'as', ArrowDown, Enter to choose Aster Lamp" });
+const items = ["Aster Lamp", "Aspen Desk", "Cable Kit", "Desk Mat", "Red Mug"];
+let active = -1, opened = false, keyPath = [];
+combo.addEventListener("input", renderList);
+combo.addEventListener("keydown", e => {
+  keyPath.push(e.key);
+  const opts = [...list.querySelectorAll('[role="option"]')];
+  if (e.key === "ArrowDown") { e.preventDefault(); active = Math.min(active + 1, opts.length - 1); paintActive(opts); }
+  if (e.key === "Enter" && active >= 0 && opts[active]) { e.preventDefault(); combo.value = opts[active].textContent; close(); check(e); }
+});
+function renderList(){
+  const q = combo.value.toLowerCase();
+  const matches = items.filter(x => x.toLowerCase().includes(q));
+  list.innerHTML = matches.map((x,i)=>`<li role="option" id="opt-${i}" style="padding:4px;cursor:pointer">${x}</li>`).join("");
+  list.style.display = matches.length ? "block" : "none";
+  combo.setAttribute("aria-expanded", matches.length ? "true" : "false");
+  active = -1; opened ||= matches.length > 0;
+}
+function paintActive(opts){ opts.forEach((o,i)=>o.style.background = i===active ? "#2b4b6b" : ""); if(opts[active]) combo.setAttribute("aria-activedescendant", opts[active].id); }
+function close(){ list.style.display="none"; combo.setAttribute("aria-expanded","false"); }
+function check(e){
+  const bad=[];
+  if (!opened) bad.push("listbox never opened");
+  if (combo.value !== "Aster Lamp") bad.push(`selected ${combo.value}`);
+  if (!keyPath.includes("ArrowDown") || !keyPath.includes("Enter")) bad.push(`missing keyboard selection path: ${keyPath.join(",")}`);
+  if (!e.isTrusted) bad.push("Enter key isTrusted=false");
+  if (bad.length) Challenge.fail(...bad); else Challenge.pass("ARIA combobox selected via keyboard");
+}
+</script>
+</body>
diff --git a/test-suite/challenges/38-spa-route-change.html b/test-suite/challenges/38-spa-route-change.html
new file mode 100644
index 0000000..533c38d
--- /dev/null
+++ b/test-suite/challenges/38-spa-route-change.html
@@ -0,0 +1,40 @@
+<!doctype html>
+<meta charset="utf-8">
+<title>38 SPA route change</title>
+<link rel="stylesheet" href="../_style.css">
+<script src="../_lib.js"></script>
+<body>
+<main>
+  <p>Goal: navigate within SPA using links/buttons. URL must change via <code>history.pushState</code> without full reload. Then go Back to Settings, forward to Billing, and Save.</p>
+  <nav><a href="/settings" id="settings">Settings</a> <a href="/settings/billing" id="billing">Billing</a></nav>
+  <section id="view"></section>
+</main>
+<script>
+Challenge.init({ id: "spa-route-change", instructions: "Settings → Billing → Back → Billing → Save, no full reload" });
+let navs = 0, popstates = 0;
+window.__spaBootId = window.__spaBootId || Math.random().toString(36).slice(2);
+const bootId = window.__spaBootId;
+const navigationEntriesAtBoot = performance.getEntriesByType("navigation").length;
+function route(path){ history.pushState({}, "", path); navs++; render(); Challenge.log("route", { path, navs }); }
+settings.addEventListener("click", e => { e.preventDefault(); route("/settings"); });
+billing.addEventListener("click", e => { e.preventDefault(); route("/settings/billing"); });
+window.addEventListener("popstate", () => { popstates++; render(); Challenge.log("popstate", { path: location.pathname, popstates }); });
+function render(){
+  if (location.pathname.endsWith("/settings/billing")) view.innerHTML = `<h2>Billing</h2><button id="save">Save billing settings</button>`;
+  else if (location.pathname.endsWith("/settings")) view.innerHTML = `<h2>Settings</h2><p>Choose Billing.</p>`;
+  else view.innerHTML = `<h2>Home</h2>`;
+  const save = document.getElementById("save");
+  if (save) save.addEventListener("click", e => {
+    const bad=[];
+    if (!location.pathname.endsWith("/settings/billing")) bad.push(`wrong route ${location.pathname}`);
+    if (navs < 2) bad.push(`only ${navs} SPA route changes`);
+    if (popstates < 1) bad.push("browser Back/popstate was not observed");
+    if (window.__spaBootId !== bootId) bad.push("window sentinel changed; reload detected");
+    if (performance.getEntriesByType("navigation").length !== navigationEntriesAtBoot) bad.push("navigation entry count changed; reload suspected");
+    if (!e.isTrusted) bad.push("save click isTrusted=false");
+    if (bad.length) Challenge.fail(...bad); else Challenge.pass("SPA pushState flow completed without reload");
+  });
+}
+render();
+</script>
+</body>
diff --git a/test-suite/fixtures/sites/mini-shop/cheats.js b/test-suite/fixtures/sites/mini-shop/cheats.js
new file mode 100644
index 0000000..7407ee4
--- /dev/null
+++ b/test-suite/fixtures/sites/mini-shop/cheats.js
@@ -0,0 +1,12 @@
+window.MiniShopCheats = {
+  "mini-shop-red-second-cheapest": [
+    { tool: "chrome_click", params: { selector: "button[aria-label=\"add Ruby Notebook\"]" } },
+    { tool: "chrome_click", params: { selector: "#checkout" } }
+  ],
+  run: {
+    "mini-shop-red-second-cheapest": async () => {
+      document.querySelector('button[aria-label="add Ruby Notebook"]').click();
+      document.querySelector('#checkout').click();
+    }
+  }
+};
diff --git a/test-suite/fixtures/sites/mini-shop/grader.js b/test-suite/fixtures/sites/mini-shop/grader.js
new file mode 100644
index 0000000..74605f4
--- /dev/null
+++ b/test-suite/fixtures/sites/mini-shop/grader.js
@@ -0,0 +1,29 @@
+window.MiniShopGrader = {
+  validate(task, state, products, run, seed = 0) {
+    let bad = [];
+    const rubric = {};
+    if (task === 'mini-shop-red-second-cheapest') {
+      const reds = products.filter(p => p.color === 'red').sort((a,b)=>a.price-b.price);
+      const target = reds[1];
+      rubric.addedTarget = state.cart.includes(target.id);
+      rubric.noWrongItems = state.cart.every(id => id === target.id);
+      rubric.checkedOut = /^MS-\d{4}$/.test(state.orderId || '');
+      if (!rubric.addedTarget) bad.push(`cart missing second-cheapest red item ${target.name}`);
+      const wrong = state.cart.filter(id => id !== target.id);
+      if (wrong.length) bad.push(`cart contains wrong product(s): ${wrong.join(',')}`);
+      if (!rubric.checkedOut) bad.push('missing checkout order id');
+    } else bad.push(`unknown task ${task}`);
+    const passed = bad.length === 0;
+    return { reward: passed ? 1 : 0, done: passed, message: passed ? 'all deterministic mini-shop checks passed' : bad.join('; '), info: { task, run, seed, rubric, bad, state } };
+  },
+  grade(task, state, products, run, seed = 0) {
+    const validation = this.validate(task, state, products, run, seed);
+    const verdict = validation.done ? 'PASS' : 'PENDING';
+    window.__taskId = task;
+    window.__taskVerdict = verdict;
+    window.__taskReason = validation.done ? [validation.message] : validation.info.bad;
+    window.__taskValidation = validation;
+    try { localStorage.setItem(`pi-chrome-task:${task}:${run}`, JSON.stringify({ id: task, run, seed, verdict, reason: window.__taskReason, validation, state, ts: Date.now() })); } catch {}
+    return validation;
+  }
+};
diff --git a/test-suite/fixtures/sites/mini-shop/index.html b/test-suite/fixtures/sites/mini-shop/index.html
new file mode 100644
index 0000000..0fd63ff
--- /dev/null
+++ b/test-suite/fixtures/sites/mini-shop/index.html
@@ -0,0 +1,55 @@
+<!doctype html>
+<meta charset="utf-8">
+<title>MiniShop hermetic site</title>
+<link rel="stylesheet" href="../../../_style.css">
+<style>
+main{max-width:1100px}.products{display:grid;grid-template-columns:repeat(auto-fit,minmax(210px,1fr));gap:12px}.card{border:1px solid #333;border-radius:10px;background:#202020;padding:12px}.cart{position:sticky;top:8px;background:#111;border:1px solid #555;border-radius:10px;padding:12px;margin-bottom:12px}.red{color:#ff8b8b}.blue{color:#8bbcff}.green{color:#9f9}.muted{color:#aaa;font-size:12px}</style>
+<body>
+<main>
+<h1>MiniShop</h1>
+<div class="cart"><b>Cart</b> <span id="cartCount">0</span> items <button id="checkout" data-task-role="checkout">Checkout</button> <span id="order"></span> <span id="reward" class="muted"></span><div id="cartItems"></div></div>
+<p id="taskText"></p>
+<section class="products" id="productsEl"></section>
+</main>
+<script src="grader.js?v=2"></script>
+<script src="cheats.js?v=2"></script>
+<script>
+(function(){
+const params=new URLSearchParams(location.search);
+if(!params.get('run')){params.set('run', crypto.randomUUID ? crypto.randomUUID() : String(Date.now())); history.replaceState(null,'',`?${params}`);}
+const task=params.get('task')||'mini-shop-red-second-cheapest'; const run=params.get('run'); const seed=Number(params.get('seed')||0);
+const key=`mini-shop:${task}:${run}:${seed}`;
+const products=[
+ {id:'p1',name:'Crimson Mug',color:'red',price:12.00},
+ {id:'p2',name:'Ruby Notebook',color:'red',price:7.50},
+ {id:'p3',name:'Scarlet Pen',color:'red',price:3.25},
+ {id:'p4',name:'Blue Cable',color:'blue',price:5.10},
+ {id:'p5',name:'Green Mat',color:'green',price:9.00},
+ {id:'p6',name:'Red Stapler',color:'red',price:18.00},
+ {id:'p7',name:'Blue Lamp',color:'blue',price:22.00},
+ {id:'p8',name:'Green Clip',color:'green',price:2.40}
+];
+let state=JSON.parse(localStorage.getItem(key)||'{"cart":[],"orderId":"","actionCount":0,"startedAt":0}');
+if(!state.startedAt) state.startedAt=Date.now();
+taskText.textContent='Task: add the second-cheapest red product to cart, checkout, and leave order id visible.';
+function save(){localStorage.setItem(key,JSON.stringify(state)); window.__miniShopState=state; const v=MiniShopGrader.grade(task,state,products,run,seed); window.__taskReward=v.reward; window.__taskTerminated=v.done; window.__taskInfo=v.info; reward.textContent=`reward=${v.reward} actions=${state.actionCount||0}`; renderCart();}
+function assignBids(){let n=0; document.querySelectorAll('button,a,input,select,textarea').forEach(el=>{const bid='b'+(++n).toString(36); el.setAttribute('bid',bid); el.setAttribute('data-bid',bid); const r=el.getBoundingClientRect(); el.setAttribute('browsergym_visibility_ratio',(r.width&&r.height?1:0).toString());});}
+function render(){productsEl.innerHTML=''; products.forEach(p=>{const d=document.createElement('div');d.className='card';d.innerHTML=`<h3 class="${p.color}">${p.name}</h3><p>Color: ${p.color}</p><p>Price: $${p.price.toFixed(2)}</p><button aria-label="add ${p.name}" data-id="${p.id}" data-task-role="add-to-cart">Add to cart</button>`;productsEl.appendChild(d)});document.querySelectorAll('[data-id]').forEach(b=>b.onclick=()=>{state.actionCount=(state.actionCount||0)+1;state.cart.push(b.dataset.id);save()});renderCart();assignBids();}
+function renderCart(){
+  cartCount.textContent=state.cart.length; order.textContent=state.orderId?`Order ${state.orderId}`:'';
+  cartItems.innerHTML=state.cart.map((id,i)=>{const p=products.find(x=>x.id===id);return `<div>${p?.name||id} <button data-remove-index="${i}" aria-label="remove ${p?.name||id}" data-task-role="remove-from-cart">Remove</button></div>`}).join('');
+  cartItems.querySelectorAll('[data-remove-index]').forEach(b=>b.onclick=()=>{state.actionCount=(state.actionCount||0)+1;state.cart.splice(Number(b.dataset.removeIndex),1);save();render();});
+  assignBids();
+}
+function orderId(){ return 'MS-' + String(1000 + ((seed * 2654435761 + state.cart.join('').length * 97) % 9000)).padStart(4, '0'); }
+checkout.onclick=()=>{state.actionCount=(state.actionCount||0)+1;state.orderId=orderId();save();};
+function observation(){return {goal:taskText.textContent,url:location.href,title:document.title,reward:window.__taskReward,actionCount:state.actionCount||0,focusedElementBid:document.activeElement?.getAttribute('bid')||null};}
+async function runCheat(){const fn=window.MiniShopCheats?.run?.[task]; if(fn) await fn(); else for(const step of (window.MiniShopCheats?.[task]||[])) console.log('cheat step',step); save();}
+window.__miniShopState=state; window.__miniShopProducts=products; window.__miniShopTask=task;
+window.__task={id:task,run,seed,goal:taskText.textContent,difficulty:'L2',step:()=>{const v=MiniShopGrader.grade(task,state,products,run,seed);return {observation:observation(),reward:v.reward,terminated:v.done,truncated:false,info:v.info}},reset:()=>{localStorage.removeItem(key);state={cart:[],orderId:'',actionCount:0,startedAt:Date.now()};render();save();return {observation:observation(),info:{task,run,seed}}},rubric:()=>MiniShopGrader.validate(task,state,products,run,seed).info.rubric,cheat:runCheat,validate:()=>MiniShopGrader.validate(task,state,products,run,seed),teardown:()=>undefined};
+window.__miniShop=window.__task;
+window.__miniShop.cheatHelpers={click:(sel)=>document.querySelector(sel).click()};
+render(); save();
+})();
+</script>
+</body>
diff --git a/test-suite/fixtures/sites/mini-shop/tasks.json b/test-suite/fixtures/sites/mini-shop/tasks.json
new file mode 100644
index 0000000..6812baf
--- /dev/null
+++ b/test-suite/fixtures/sites/mini-shop/tasks.json
@@ -0,0 +1,9 @@
+[
+  {
+    "id": "mini-shop-red-second-cheapest",
+    "difficulty": "L2",
+    "intent": "Add the second-cheapest red product to cart, checkout, and leave the generated order id visible.",
+    "gradeSource": "harness",
+    "grader": "MiniShopGrader.grade(task, state, products, run)"
+  }
+]
diff --git a/test-suite/index.html b/test-suite/index.html
index a4a3c0c..00ef204 100644
--- a/test-suite/index.html
+++ b/test-suite/index.html
@@ -35,6 +35,13 @@
   <tbody></tbody>
 </table>
 
+<h2>Long-horizon hermetic tasks</h2>
+<p class="hint">These WebArena/BrowserGym-inspired tasks use deterministic in-page graders and fresh <code>$RUN_ID</code> state. They test multi-step browser work beyond event fidelity.</p>
+<table id="taskTbl">
+  <thead><tr><th>id</th><th>difficulty</th><th>category</th><th>intent</th><th>requires</th><th>open</th></tr></thead>
+  <tbody></tbody>
+</table>
+
 <h2>Agent run loop</h2>
 <pre class="code">python3 -m http.server 8765
 # For each manifest entry:
@@ -46,7 +53,9 @@
 </main>
 <script>
 let manifest = [];
+let taskManifest = [];
 const tbody = document.querySelector("#tbl tbody");
+const taskTbody = document.querySelector("#taskTbl tbody");
 const modeEl = document.getElementById("mode");
 const filterEl = document.getElementById("filter");
 
@@ -123,9 +132,28 @@ function paint() {
 function escapeHtml(s) {
   return String(s).replace(/[&<>"']/g, c => ({"&":"&amp;","<":"&lt;",">":"&gt;","\"":"&quot;","'":"&#39;"}[c]));
 }
+function paintTasks() {
+  taskTbody.innerHTML = "";
+  const run = "manual-" + new Date().toISOString().slice(0,10);
+  for (const t of taskManifest) {
+    const url = t.startUrl.replaceAll("$RUN_ID", run).replaceAll("$SEED", String(t.seed ?? 0));
+    const tr = document.createElement("tr");
+    tr.innerHTML = `
+      <td><code>${t.taskId || t.id}</code><div class="notes">seed=${t.seed ?? 0} maxSteps=${t.maxSteps ?? "—"}</div></td>
+      <td><span class="pill expected">${t.difficulty}</span></td>
+      <td>${t.category}</td>
+      <td>${escapeHtml(t.goal || t.intent)}</td>
+      <td><div class="notes">${escapeHtml((t.requires || []).join(", "))}</div><div class="notes">actions: ${escapeHtml((t.actionSubsets || []).join(", "))}</div></td>
+      <td><a target="_blank" href="${url}">open</a></td>
+    `;
+    taskTbody.appendChild(tr);
+  }
+}
 async function init() {
   manifest = await fetch("manifest.json", { cache: "no-store" }).then(r => r.json());
+  taskManifest = await fetch("task-manifest.json", { cache: "no-store" }).then(r => r.json()).catch(() => []);
   paint();
+  paintTasks();
 }
 document.getElementById("refresh").addEventListener("click", paint);
 document.getElementById("clear").addEventListener("click", () => {
diff --git a/test-suite/manifest.json b/test-suite/manifest.json
index c09cf31..900b5d3 100644
--- a/test-suite/manifest.json
+++ b/test-suite/manifest.json
@@ -21,7 +21,13 @@
     "notes": [
       "Core debugger/CDP smoke test."
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "trusted-input"
+    ],
+    "requires": {}
   },
   {
     "id": "is-trusted-keyboard",
@@ -43,7 +49,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "trusted-input"
+    ],
+    "requires": {}
   },
   {
     "id": "webdriver-flag",
@@ -66,7 +78,13 @@
     "notes": [
       "Environment-sensitive; classify unexpected failures as profile warnings first."
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "fingerprint"
+    ],
+    "requires": {}
   },
   {
     "id": "mouse-entropy",
@@ -74,7 +92,7 @@
     "category": "pointer-humanization",
     "goal": "Click is preceded by organic mousemove path.",
     "expected": {
-      "synthetic": "PASS",
+      "synthetic": "FAIL",
       "trusted": "PASS",
       "manual": "PASS"
     },
@@ -86,7 +104,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "pointer-humanization"
+    ],
+    "requires": {}
   },
   {
     "id": "event-timing",
@@ -94,7 +118,7 @@
     "category": "timing",
     "goal": "Repeated clicks have plausible holds and gaps.",
     "expected": {
-      "synthetic": "PASS",
+      "synthetic": "CONDITIONAL",
       "trusted": "PASS",
       "manual": "PASS"
     },
@@ -119,7 +143,17 @@
       }
     ],
     "flakeRisk": "medium",
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "timing"
+    ],
+    "requires": {},
+    "thresholds": {},
+    "notes": [
+      "Timing thresholds can be tuned via ?threshold.<name>=value for slower CI or high-polling devices."
+    ]
   },
   {
     "id": "click-coordinates",
@@ -127,7 +161,7 @@
     "category": "pointer-humanization",
     "goal": "Clicks land inside target but not always exact center.",
     "expected": {
-      "synthetic": "PASS",
+      "synthetic": "CONDITIONAL",
       "trusted": "PASS",
       "manual": "PASS"
     },
@@ -163,7 +197,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "pointer-humanization"
+    ],
+    "requires": {}
   },
   {
     "id": "pointer-properties",
@@ -171,7 +211,7 @@
     "category": "pointer-humanization",
     "goal": "Pointer events carry plausible pointerType, pressure, movement, id.",
     "expected": {
-      "synthetic": "PASS",
+      "synthetic": "CONDITIONAL",
       "trusted": "PASS",
       "manual": "PASS"
     },
@@ -183,7 +223,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "pointer-humanization"
+    ],
+    "requires": {}
   },
   {
     "id": "keyboard-cadence",
@@ -191,7 +237,7 @@
     "category": "keyboard",
     "goal": "Typing emits per-character key events with non-uniform cadence.",
     "expected": {
-      "synthetic": "PASS",
+      "synthetic": "CONDITIONAL",
       "trusted": "PASS",
       "manual": "PASS"
     },
@@ -204,7 +250,17 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "keyboard"
+    ],
+    "requires": {},
+    "thresholds": {},
+    "notes": [
+      "Timing thresholds can be tuned via ?threshold.<name>=value for slower CI or high-polling devices."
+    ]
   },
   {
     "id": "composition-input",
@@ -225,7 +281,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "keyboard"
+    ],
+    "requires": {}
   },
   {
     "id": "user-activation",
@@ -252,7 +314,13 @@
       "clipboard/fullscreen allowed by browser policy"
     ],
     "flakeRisk": "high",
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "activation-gates"
+    ],
+    "requires": {}
   },
   {
     "id": "honeypot",
@@ -279,7 +347,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "agent-safety"
+    ],
+    "requires": {}
   },
   {
     "id": "fingerprint",
@@ -303,7 +377,13 @@
       "non-headless normal Chrome profile"
     ],
     "flakeRisk": "medium",
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "fingerprint"
+    ],
+    "requires": {}
   },
   {
     "id": "focus-order",
@@ -326,7 +406,13 @@
       }
     ],
     "flakeRisk": "medium",
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "focus-keyboard"
+    ],
+    "requires": {}
   },
   {
     "id": "wheel-scroll",
@@ -347,7 +433,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "scroll"
+    ],
+    "requires": {}
   },
   {
     "id": "drag-drop-datatransfer",
@@ -372,7 +464,13 @@
     "notes": [
       "May need CDP Input.dispatchDragEvent rather than mouse drag on some Chrome versions."
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "drag-drop"
+    ],
+    "requires": {}
   },
   {
     "id": "contenteditable-selection",
@@ -380,7 +478,7 @@
     "category": "editing",
     "goal": "Contenteditable typing advances selection/caret like user typing.",
     "expected": {
-      "synthetic": "PASS",
+      "synthetic": "CONDITIONAL",
       "trusted": "PASS",
       "manual": "PASS"
     },
@@ -393,7 +491,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "editing"
+    ],
+    "requires": {}
   },
   {
     "id": "paste-clipboard",
@@ -428,7 +532,13 @@
       "OS clipboard contains pi-chrome",
       "use ctrlKey instead of metaKey on Windows/Linux"
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "clipboard"
+    ],
+    "requires": {}
   },
   {
     "id": "native-select",
@@ -471,7 +581,13 @@
       }
     ],
     "flakeRisk": "high",
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "native-controls"
+    ],
+    "requires": {}
   },
   {
     "id": "hover-dwell",
@@ -505,7 +621,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "hover"
+    ],
+    "requires": {}
   },
   {
     "id": "react-value-tracker",
@@ -526,7 +648,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "frameworks"
+    ],
+    "requires": {}
   },
   {
     "id": "keyboard-modifiers",
@@ -550,7 +678,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "keyboard"
+    ],
+    "requires": {}
   },
   {
     "id": "touch-events",
@@ -573,7 +707,13 @@
     "prerequisites": [
       "touch events enabled/supported"
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "touch"
+    ],
+    "requires": {}
   },
   {
     "id": "stack-trace-fingerprint",
@@ -595,9 +735,16 @@
       }
     ],
     "notes": [
-      "Adversarial probe; selector resolution itself can be detected by monkey-patched DOM APIs."
+      "Adversarial probe; selector resolution itself can be detected by monkey-patched DOM APIs.",
+      "Should fail only on concrete automation URLs/Runtime frames, not generic anonymous stack formatting."
+    ],
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "fingerprint"
     ],
-    "manualBaseline": "unverified"
+    "requires": {}
   },
   {
     "id": "viewport-edge-clicks",
@@ -618,7 +765,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "pointer-humanization"
+    ],
+    "requires": {}
   },
   {
     "id": "pointer-continuity",
@@ -646,7 +799,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "pointer-humanization"
+    ],
+    "requires": {}
   },
   {
     "id": "mousemove-rate",
@@ -675,10 +834,18 @@
       }
     ],
     "notes": [
-      "Needs a first-class pointer path/sweep tool for deterministic automation."
+      "Needs a first-class pointer path/sweep tool for deterministic automation.",
+      "Timing thresholds can be tuned via ?threshold.<name>=value for slower CI or high-polling devices."
     ],
     "flakeRisk": "high",
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "pointer-humanization"
+    ],
+    "requires": {},
+    "thresholds": {}
   },
   {
     "id": "scroll-momentum",
@@ -703,7 +870,16 @@
     "notes": [
       "Physical detent mouse wheels may fail decay requirement; trackpads usually pass."
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "scroll"
+    ],
+    "requires": {},
+    "prerequisites": [
+      "Trackpad-like momentum; physical detent mouse wheels may fail decay requirement even for manual users."
+    ]
   },
   {
     "id": "intersection-visibility",
@@ -733,7 +909,13 @@
       }
     ],
     "flakeRisk": "medium",
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "scroll-visibility"
+    ],
+    "requires": {}
   },
   {
     "id": "shadow-dom-controls",
@@ -767,7 +949,13 @@
     "manualBaseline": "unverified",
     "prerequisites": [
       "pi-chrome shadow-piercing selectors or shadow-root-aware snapshot uids"
-    ]
+    ],
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "dom-complexity"
+    ],
+    "requires": {}
   },
   {
     "id": "iframe-targeting",
@@ -800,7 +988,13 @@
     "manualBaseline": "unverified",
     "prerequisites": [
       "pi-chrome currently targets frameId 0 for many actions; iframe drilling needs frame-targeting support"
-    ]
+    ],
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "frames"
+    ],
+    "requires": {}
   },
   {
     "id": "file-upload",
@@ -826,7 +1020,13 @@
     "manualBaseline": "unverified",
     "notes": [
       "Recipe paths may contain $PWD; external runners should expand shell-style placeholders before calling chrome_upload_file."
-    ]
+    ],
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "files"
+    ],
+    "requires": {}
   },
   {
     "id": "keyboard-tab-navigation",
@@ -903,7 +1103,13 @@
         }
       }
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "focus-keyboard"
+    ],
+    "requires": {}
   },
   {
     "id": "network-console-capture",
@@ -911,7 +1117,7 @@
     "category": "observability",
     "goal": "Interaction emits console logs and fetch traffic for tool-level capture APIs.",
     "expected": {
-      "synthetic": "PASS",
+      "synthetic": "CONDITIONAL",
       "trusted": "PASS",
       "manual": "PASS"
     },
@@ -947,6 +1153,285 @@
       "Page verdict only checks in-page fetch result; benchmark harness should also assert captured console/network entries.",
       "First chrome_list_console_messages/chrome_list_network_requests call installs capture hooks; do not move clear/list calls after click."
     ],
-    "manualBaseline": "unverified"
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "difficulty": "L1",
+    "tags": [
+      "observability"
+    ],
+    "requires": {}
+  },
+  {
+    "id": "dialog-handling",
+    "file": "challenges/34-dialog-handling.html",
+    "category": "dialogs",
+    "difficulty": "L2",
+    "goal": "Agent detects and handles alert/confirm/prompt/beforeunload browser dialogs.",
+    "expected": {
+      "synthetic": "CONDITIONAL",
+      "trusted": "CONDITIONAL",
+      "manual": "PASS"
+    },
+    "recipe": [
+      {
+        "tool": "chrome_click",
+        "params": {
+          "selector": "#promptBtn",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "dialog",
+        "params": {
+          "type": "prompt",
+          "text": "pi-chrome",
+          "accept": true
+        }
+      },
+      {
+        "tool": "chrome_click",
+        "params": {
+          "selector": "#confirmBtn",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "dialog",
+        "params": {
+          "type": "confirm",
+          "accept": true
+        }
+      }
+    ],
+    "requires": {
+      "dialogs": true
+    },
+    "prerequisites": [
+      "browser-dialog handling API or manual user intervention"
+    ],
+    "tags": [
+      "dialogs",
+      "native-browser-ui"
+    ],
+    "notes": [
+      "If dialog handling is unsupported, page may block after click; this is expected signal."
+    ],
+    "manualBaseline": "unverified",
+    "gradeSource": "page"
+  },
+  {
+    "id": "target-blank-popup",
+    "file": "challenges/35-target-blank-popup.html",
+    "category": "popups",
+    "difficulty": "L1",
+    "goal": "Agent handles target=_blank new tab, discovers it via tab list, and inspects child tab verdict.",
+    "expected": {
+      "synthetic": "CONDITIONAL",
+      "trusted": "PASS",
+      "manual": "PASS"
+    },
+    "recipe": [
+      {
+        "tool": "chrome_click",
+        "params": {
+          "selector": "#open",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_tab",
+        "params": {
+          "action": "list"
+        }
+      },
+      {
+        "tool": "chrome_snapshot",
+        "params": {
+          "titleIncludes": "target blank popup"
+        }
+      }
+    ],
+    "requires": {
+      "popups": true
+    },
+    "tags": [
+      "tabs",
+      "popup"
+    ],
+    "notes": [
+      "Child tab writes PASS to shared localStorage; harness should verify tab count/title too."
+    ],
+    "manualBaseline": "unverified",
+    "gradeSource": "page"
+  },
+  {
+    "id": "modal-focus-trap",
+    "file": "challenges/36-modal-focus-trap.html",
+    "category": "focus-keyboard",
+    "difficulty": "L2",
+    "goal": "Keyboard focus stays trapped inside modal and Escape closes with focus restoration.",
+    "expected": {
+      "synthetic": "CONDITIONAL",
+      "trusted": "PASS",
+      "manual": "PASS"
+    },
+    "recipe": [
+      {
+        "tool": "chrome_click",
+        "params": {
+          "selector": "#open",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_key",
+        "params": {
+          "key": "Tab",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_key",
+        "params": {
+          "key": "Tab",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_key",
+        "params": {
+          "key": "Tab",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_key",
+        "params": {
+          "key": "Escape",
+          "trusted": true
+        }
+      }
+    ],
+    "requires": {
+      "cdp": true
+    },
+    "tags": [
+      "modal",
+      "focus",
+      "keyboard"
+    ],
+    "manualBaseline": "unverified",
+    "gradeSource": "page"
+  },
+  {
+    "id": "autocomplete-combobox",
+    "file": "challenges/37-autocomplete-combobox.html",
+    "category": "forms",
+    "difficulty": "L2",
+    "goal": "Agent uses ARIA combobox keyboard path: type filter, ArrowDown, Enter.",
+    "expected": {
+      "synthetic": "FAIL",
+      "trusted": "PASS",
+      "manual": "PASS"
+    },
+    "recipe": [
+      {
+        "tool": "chrome_type",
+        "params": {
+          "selector": "#combo",
+          "text": "as",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_key",
+        "params": {
+          "key": "ArrowDown",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_key",
+        "params": {
+          "key": "Enter",
+          "trusted": true
+        }
+      }
+    ],
+    "requires": {
+      "cdp": true
+    },
+    "tags": [
+      "aria",
+      "combobox",
+      "forms"
+    ],
+    "manualBaseline": "unverified",
+    "gradeSource": "page"
+  },
+  {
+    "id": "spa-route-change",
+    "file": "challenges/38-spa-route-change.html",
+    "category": "spa-routing",
+    "difficulty": "L1",
+    "goal": "Agent recognizes SPA pushState navigation and completes route-local action without full reload.",
+    "expected": {
+      "synthetic": "CONDITIONAL",
+      "trusted": "PASS",
+      "manual": "PASS"
+    },
+    "recipe": [
+      {
+        "tool": "chrome_click",
+        "params": {
+          "selector": "#settings",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_click",
+        "params": {
+          "selector": "#billing",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_key",
+        "params": {
+          "key": "ArrowLeft",
+          "modifiers": {
+            "altKey": true
+          },
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_click",
+        "params": {
+          "selector": "#billing",
+          "trusted": true
+        }
+      },
+      {
+        "tool": "chrome_click",
+        "params": {
+          "selector": "#save",
+          "trusted": true
+        }
+      }
+    ],
+    "requires": {
+      "cdp": true
+    },
+    "tags": [
+      "spa",
+      "history",
+      "navigation"
+    ],
+    "manualBaseline": "unverified",
+    "gradeSource": "page",
+    "notes": [
+      "Requires a real browser Back action (e.g. Alt+Left) to exercise popstate before Save."
+    ]
   }
 ]
diff --git a/test-suite/manifest.schema.json b/test-suite/manifest.schema.json
index 79371e8..cd6dcce 100644
--- a/test-suite/manifest.schema.json
+++ b/test-suite/manifest.schema.json
@@ -1,7 +1,7 @@
 {
   "$schema": "https://json-schema.org/draft/2020-12/schema",
   "title": "pi-chrome benchmark manifest",
-  "description": "Manifest for browser-control benchmark pages. Recipes are canonical intent, not guaranteed raw tool JSON: runners may need to adapt descriptive selectors (frame/shadow) and expand path placeholders such as $PWD before invoking tools.",
+  "description": "Manifest for browser-control benchmark pages. Recipes are canonical intent, not guaranteed raw tool JSON: runners may need to adapt descriptive selectors (frame/shadow) and expand path placeholders such as ${REPO_ROOT} before invoking tools.",
   "type": "array",
   "items": {
     "type": "object",
@@ -11,7 +11,10 @@
       "id": { "type": "string" },
       "file": { "type": "string" },
       "category": { "$ref": "#/$defs/category" },
+      "difficulty": { "enum": ["L1", "L2", "L3"] },
+      "tags": { "type": "array", "items": { "type": "string" } },
       "goal": { "type": "string" },
+      "gradeSource": { "enum": ["page", "harness", "both"], "default": "page" },
       "expected": {
         "type": "object",
         "description": "Expected verdict by execution mode. CONDITIONAL means prerequisites, browser policy, hardware, or missing tool primitives decide outcome.",
@@ -22,10 +25,7 @@
           "manual": { "$ref": "#/$defs/outcome" }
         }
       },
-      "manualBaseline": {
-        "description": "Whether human baseline has been explicitly verified in this repo/environment.",
-        "enum": ["unverified", "verified"]
-      },
+      "manualBaseline": { "description": "Whether human baseline has been explicitly verified in this repo/environment.", "enum": ["unverified", "verified"] },
       "manualBaselineVerifiedAt": { "type": "string", "format": "date-time" },
       "recipe": {
         "type": "array",
@@ -35,15 +35,28 @@
           "required": ["tool", "params"],
           "properties": {
             "tool": { "type": "string" },
-            "params": {
-              "type": "object",
-              "description": "Tool params. For upload recipes, paths are strings after runner expansion; $PWD means repository working directory."
-            }
+            "params": { "type": "object", "description": "Tool params. For upload recipes, paths are strings after runner expansion; ${REPO_ROOT} means repository root." }
           }
         }
       },
+      "requires": {
+        "type": "object",
+        "additionalProperties": true,
+        "properties": {
+          "cdp": { "type": "boolean" },
+          "clipboard": { "type": "boolean" },
+          "touch": { "type": "boolean" },
+          "secureOrigin": { "type": "boolean" },
+          "fileSystem": { "type": "boolean" },
+          "dialogs": { "type": "boolean" },
+          "downloads": { "type": "boolean" },
+          "popups": { "type": "boolean" }
+        }
+      },
       "prerequisites": { "type": "array", "items": { "type": "string" } },
       "notes": { "type": "array", "items": { "type": "string" } },
+      "thresholds": { "type": "object", "additionalProperties": true },
+      "timeoutMs": { "type": "number" },
       "verdictDelayMs": { "type": "number" },
       "flakeRisk": { "enum": ["low", "medium", "high"] }
     }
@@ -52,27 +65,7 @@
     "outcome": { "enum": ["PASS", "FAIL", "SKIP", "WARN", "CONDITIONAL"] },
     "category": {
       "enum": [
-        "activation-gates",
-        "agent-safety",
-        "clipboard",
-        "dom-complexity",
-        "drag-drop",
-        "editing",
-        "files",
-        "fingerprint",
-        "focus-keyboard",
-        "frames",
-        "frameworks",
-        "hover",
-        "keyboard",
-        "native-controls",
-        "observability",
-        "pointer-humanization",
-        "scroll",
-        "scroll-visibility",
-        "timing",
-        "touch",
-        "trusted-input"
+        "activation-gates", "agent-safety", "aria-snapshot", "canvas-svg", "clipboard", "contextmenu", "dialogs", "dom-complexity", "downloads", "drag-drop", "editing", "files", "fingerprint", "focus-keyboard", "forms", "frames", "frameworks", "hover", "keyboard", "lazy-loading", "native-controls", "observability", "pointer-humanization", "popups", "scroll", "scroll-visibility", "spa-routing", "task-completion", "timing", "touch", "trusted-input"
       ]
     }
   }
diff --git a/test-suite/notes/browsergym-compat.md b/test-suite/notes/browsergym-compat.md
new file mode 100644
index 0000000..1eaea41
--- /dev/null
+++ b/test-suite/notes/browsergym-compat.md
@@ -0,0 +1,70 @@
+# BrowserGym compatibility contract
+
+Pi-chrome benchmark tasks mirror BrowserGym without importing Gymnasium.
+
+## Task fields
+
+Each `task-manifest.json` entry maps to `AbstractBrowserTask` concepts:
+
+- `taskId` / `id` — stable namespaced id
+- `seed` — deterministic state seed
+- `viewport`, `slowMoMs`, `timeoutMs`, `locale`, `timezoneId`
+- `goal` and `goalObject` — text and OpenAI-style message objects
+- `setupUrl` — page URL that performs setup from `task`, `run`, `seed`
+- `validateScript` — JS validate hook returning reward contract
+- `cheatScript` — gold recipe used to sanity-check grader
+- `maxSteps`, `humanBaseline`, `difficulty`, `tags`
+- `actionSubsets`, `allowedActions` — BrowserGym HighLevelActionSet metadata
+
+## Reset / step return
+
+Runner should expose:
+
+```text
+reset() -> (obs, info)
+step(action) -> (obs, reward, terminated, truncated, info)
+```
+
+`terminated` means task done. `truncated` means max step or timeout.
+
+## Validate return
+
+In-page validators return:
+
+```js
+{ reward, done, message, info }
+```
+
+`reward` is incremental. Current hermetic tasks use binary terminal reward. `info.rubric` carries per-check details.
+
+## Observation schema target
+
+Runner observations should include BrowserGym-like keys:
+
+- `chat_messages`
+- `goal`, `goal_object`
+- `open_pages_urls`, `open_pages_titles`, `active_page_index`
+- `url`, `screenshot`, `dom_object`, `axtree_object`
+- `extra_element_properties`
+- `focused_element_bid`
+- `last_action`, `last_action_error`
+- `elapsed_time`
+
+## BID plan
+
+BrowserGym uses `bid` attributes as stable element handles. Pi-chrome currently returns snapshot `uid`. Compatibility path:
+
+1. During snapshot, assign every interactive element a deterministic `bid`.
+2. Return both `uid` and `bid`.
+3. Add `visibility`, `bbox`, `clickable`, `set_of_marks` metadata.
+4. Maintain uid↔bid mapping so agents can use either addressing mode.
+
+Target attributes:
+
+- `bid`
+- `browsergym_visibility_ratio`
+- `browsergym_set_of_marks`
+
+## Action subsets
+
+See `../browsergym-action-space.json` for BrowserGym-compatible subsets and function signatures.
diff --git a/test-suite/notes/profiles.md b/test-suite/notes/profiles.md
new file mode 100644
index 0000000..8f55797
--- /dev/null
+++ b/test-suite/notes/profiles.md
@@ -0,0 +1,22 @@
+# Calibration profiles
+
+Benchmark results depend on Chrome profile, OS, input device, and extension mode.
+Record these with every run:
+
+- OS and version
+- Chrome version
+- pi-chrome package version
+- companion extension version
+- trusted mode: off / auto / on
+- input hardware: trackpad vs detent mouse, touch support
+- permissions: clipboard, fullscreen, downloads
+- viewport size and deviceScaleFactor
+- loaded privacy/security extensions
+
+Known environment-sensitive areas:
+
+- WebGL fingerprint can warn/fail in VMs, remote desktops, GPU-disabled Chrome.
+- Scroll momentum expects trackpad-like decay; detent mouse wheels may fail manually.
+- Touch tests require touch events enabled/supported.
+- Dialog/download/file-picker tests require native browser UI handling or manual intervention.
+- Stack traces vary across Chrome versions; tests should only fail concrete automation URLs/Runtime frames.
diff --git a/test-suite/notes/runner-spec.md b/test-suite/notes/runner-spec.md
new file mode 100644
index 0000000..b040cb4
--- /dev/null
+++ b/test-suite/notes/runner-spec.md
@@ -0,0 +1,26 @@
+# Recipe runner spec
+
+Future Node/Pi runner should:
+
+1. Read `manifest.json`.
+2. Probe environment capabilities (`trusted`, `clipboard`, `touch`, `dialogs`, `downloads`, `fileSystem`).
+3. For each entry:
+   - skip if `requires` is unsatisfied and expected is `CONDITIONAL`
+   - navigate to challenge URL
+   - execute `recipe` in order
+   - wait `verdictDelayMs || 500`
+   - evaluate `JSON.stringify({v:window.__verdict,r:window.__reason,d:window.Challenge?.state?.details,e:window.__events?.slice(-20)})`
+   - compare to `expected[mode]`
+4. Output:
+   - Markdown summary
+   - JSON details
+   - JUnit XML for CI
+
+Runner must adapt recipe intent:
+
+- expand `${REPO_ROOT}` path placeholders
+- adapt unsupported shadow/iframe selector notation to snapshot uids or evaluate fallback
+- preserve hook install ordering for console/network capture tests
+- record whether trusted/CDP path was used
+
+Long-horizon task runner should read `task-manifest.json`, replace `$RUN_ID`, solve task, then evaluate task grader expression.
diff --git a/test-suite/notes/scoring.md b/test-suite/notes/scoring.md
new file mode 100644
index 0000000..970a706
--- /dev/null
+++ b/test-suite/notes/scoring.md
@@ -0,0 +1,36 @@
+# Benchmark scoring
+
+## Unit challenges (`manifest.json`)
+
+Score actual verdict against `expected[mode]`, not raw PASS count.
+
+- `PASS` / `FAIL` expected: exact match required.
+- `CONDITIONAL`: exclude from headline score unless environment prerequisite is declared present.
+- `SKIP` / `WARN`: report separately.
+
+Recommended flake policy:
+
+- Run each non-destructive challenge up to 2 retries.
+- Take best verdict only for known-flaky timing/scroll tests.
+- Always keep all reasons/events in detailed report.
+
+Difficulty weighting:
+
+- L1 = 1
+- L2 = 2
+- L3 = 3
+
+Report both unweighted and weighted score. Also report per-category score.
+
+## Long-horizon tasks (`task-manifest.json`)
+
+Task score is deterministic grader PASS / not PASS. Record:
+
+- action count
+- wall time
+- tools used
+- observation mode (`snapshot`, screenshot, accessibility, evaluate)
+- whether direct state mutation/evaluate was allowed
+- final grader reason
+
+Do not use LLM judge for hermetic tasks unless grader cannot express the target.
diff --git a/test-suite/scenarios/choredesk/cheats.js b/test-suite/scenarios/choredesk/cheats.js
new file mode 100644
index 0000000..112679f
--- /dev/null
+++ b/test-suite/scenarios/choredesk/cheats.js
@@ -0,0 +1,49 @@
+window.ChoreDeskCheats = {
+  "choredesk-l1-ticket-priority": [
+    { tool: "chrome_fill", params: { selector: "select[aria-label=\"T-104 priority\"]", text: "High" } },
+    { tool: "chrome_fill", params: { selector: "select[aria-label=\"T-104 status\"]", text: "In Progress" } },
+    { tool: "chrome_click", params: { selector: "button[aria-label=\"save tickets T-104\"]" } }
+  ],
+  "choredesk-l2-refund-message": [
+    { tool: "chrome_click", params: { selector: "a[data-view=\"customers\"]" } },
+    { tool: "chrome_click", params: { selector: "a[data-view=\"orders\"]" } },
+    { tool: "chrome_click", params: { selector: "a[data-view=\"messages\"]" } },
+    { tool: "chrome_fill", params: { selector: "#msgSubject", text: "Refund review ORD-778" } },
+    { tool: "chrome_fill", params: { selector: "#msgBody", text: "Noah Kim noah.kim@example.com order ORD-778 total $84.20" } },
+    { tool: "chrome_click", params: { selector: "#sendMsg" } }
+  ],
+  "choredesk-l3-restock-ticket": [
+    { tool: "chrome_click", params: { selector: "a[data-view=\"catalog\"]" } },
+    { tool: "chrome_click", params: { selector: "a[data-view=\"inventory\"]" } },
+    { tool: "chrome_fill", params: { selector: "input[aria-label=\"AST-LAMP reorder quantity\"]", text: "12" } },
+    { tool: "chrome_click", params: { selector: "button[aria-label=\"save inventory AST-LAMP\"]" } },
+    { tool: "chrome_click", params: { selector: "a[data-view=\"tickets\"]" } },
+    { tool: "chrome_fill", params: { selector: "input[aria-label=\"T-205 internal note\"]", text: "reordered 12 AST-LAMP" } },
+    { tool: "chrome_click", params: { selector: "button[aria-label=\"save tickets T-205\"]" } }
+  ],
+  run: {
+    "choredesk-l1-ticket-priority": async () => {
+      await window.__choredesk.cheatHelpers.route("tickets");
+      window.__choredesk.cheatHelpers.set("select[aria-label=\"T-104 priority\"]", "High");
+      window.__choredesk.cheatHelpers.set("select[aria-label=\"T-104 status\"]", "In Progress");
+      document.querySelector("button[aria-label=\"save tickets T-104\"]").click();
+    },
+    "choredesk-l2-refund-message": async () => {
+      await window.__choredesk.cheatHelpers.route("customers");
+      await window.__choredesk.cheatHelpers.route("orders");
+      await window.__choredesk.cheatHelpers.route("messages");
+      window.__choredesk.cheatHelpers.set("#msgSubject", "Refund review ORD-778");
+      window.__choredesk.cheatHelpers.set("#msgBody", "Noah Kim noah.kim@example.com order ORD-778 total $84.20");
+      document.querySelector("#sendMsg").click();
+    },
+    "choredesk-l3-restock-ticket": async () => {
+      await window.__choredesk.cheatHelpers.route("catalog");
+      await window.__choredesk.cheatHelpers.route("inventory");
+      window.__choredesk.cheatHelpers.set("input[aria-label=\"AST-LAMP reorder quantity\"]", "12");
+      document.querySelector("button[aria-label=\"save inventory AST-LAMP\"]").click();
+      await window.__choredesk.cheatHelpers.route("tickets");
+      window.__choredesk.cheatHelpers.set("input[aria-label=\"T-205 internal note\"]", "reordered 12 AST-LAMP");
+      document.querySelector("button[aria-label=\"save tickets T-205\"]").click();
+    }
+  }
+};
diff --git a/test-suite/scenarios/choredesk/index.html b/test-suite/scenarios/choredesk/index.html
new file mode 100644
index 0000000..6481ac0
--- /dev/null
+++ b/test-suite/scenarios/choredesk/index.html
@@ -0,0 +1,239 @@
+<!doctype html>
+<meta charset="utf-8">
+<title>ChoreDesk hermetic task site</title>
+<link rel="stylesheet" href="../../_style.css">
+<style>
+  main { max-width: 1180px; }
+  .top { display:flex; gap:12px; align-items:center; flex-wrap:wrap; }
+  .tabs { display:flex; gap:8px; margin:16px 0; flex-wrap:wrap; }
+  .tab { border:1px solid #444; border-radius:999px; padding:6px 10px; color:#eee; text-decoration:none; background:#222; }
+  .tab[aria-current="page"] { background:#2b4b6b; }
+  .card { border:1px solid #333; border-radius:10px; background:#202020; padding:14px; }
+  table { font-size:13px; }
+  td input, td select { width: 100%; box-sizing:border-box; }
+  textarea { width:100%; min-height:110px; background:#111; color:#eee; border:1px solid #555; border-radius:6px; padding:8px; }
+  .muted { color:#aaa; font-size:12px; }
+  .taskbar { border:1px solid #4b3b16; background:#221b0d; border-radius:10px; padding:12px; margin:12px 0; }
+  .verdict { display:inline-block; border-radius:999px; padding:2px 8px; font-weight:700; background:#444; }
+  .verdict.PASS { background:#1f7a1f; }
+  .verdict.FAIL { background:#a11; }
+  .verdict.PENDING { background:#444; }
+</style>
+<body>
+<main>
+  <div class="top">
+    <h1 style="margin-right:auto">ChoreDesk</h1>
+    <button id="reset" data-task-role="reset-button">Reset run</button>
+    <button id="grade" data-task-role="grade-button">Grade now</button>
+    <button id="cheat" data-task-role="cheat-button">Run cheat</button>
+  </div>
+  <div class="taskbar">
+    <div><b id="taskId"></b> <span id="difficulty"></span> <span id="verdict" class="verdict PENDING">PENDING</span> <span id="reward" class="muted"></span></div>
+    <p id="instruction"></p>
+    <pre id="reason" class="muted"></pre>
+  </div>
+  <nav class="tabs" aria-label="ChoreDesk sections"></nav>
+  <section id="view"></section>
+</main>
+<script src="cheats.js"></script>
+<script>
+(function () {
+const params = new URLSearchParams(location.search);
+if (!params.get("run")) {
+  params.set("run", crypto.randomUUID ? crypto.randomUUID() : String(Date.now()));
+  history.replaceState(null, "", `?${params}`);
+}
+const task = params.get("task") || "choredesk-l1-ticket-priority";
+const run = params.get("run");
+const seed = Number(params.get("seed") || 0);
+const nonce = params.get("nonce") || (crypto.randomUUID ? crypto.randomUUID() : Math.random().toString(36).slice(2));
+if (!params.get("nonce")) { params.set("nonce", nonce); history.replaceState(null, "", `?${params}`); }
+const storeKey = `pi-chrome-task-state:${task}:${run}:${seed}:${nonce}`;
+const verdictKey = `pi-chrome-task:${task}:${run}`;
+const views = ["tickets", "customers", "orders", "catalog", "inventory", "messages", "audit"];
+let view = params.get("view") || "tickets";
+let bidCounter = 0;
+
+const TASKS = {
+  "choredesk-l1-ticket-priority": {
+    difficulty: "L1 atomic",
+    instruction: "Set ticket T-104 priority to High and status to In Progress. Save the ticket.",
+    rubric(s) {
+      const t = s.tickets.find(x => x.id === "T-104");
+      return [
+        { id:"priorityHigh", weight:.33, pass:t?.priority === "High", description:"T-104 priority is High", reason:`T-104 priority=${t?.priority}` },
+        { id:"statusInProgress", weight:.33, pass:t?.status === "In Progress", description:"T-104 status is In Progress", reason:`T-104 status=${t?.status}` },
+        { id:"saved", weight:.34, pass:s.audit.some(a => a.kind === "ticket-save" && a.id === "T-104" && a.via === "ui-event"), description:"T-104 saved through UI", reason:"no UI save audit for T-104" }
+      ];
+    }
+  },
+  "choredesk-l2-refund-message": {
+    difficulty: "L2 compositional",
+    instruction: "A customer named Noah Kim reports a duplicate charge. Find Noah's email and order ORD-778 total, then send Billing a message with subject 'Refund review ORD-778'. Body must include Noah's email and the exact order total.",
+    rubric(s) {
+      const msg = s.messages.find(m => m.to === "Billing" && m.subject.trim() === "Refund review ORD-778");
+      return [
+        { id:"visitedCustomers", weight:.15, pass:!!s.viewsVisited.customers, description:"Opened Customers view", reason:"never opened Customers view" },
+        { id:"visitedOrders", weight:.15, pass:!!s.viewsVisited.orders, description:"Opened Orders view", reason:"never opened Orders view" },
+        { id:"subject", weight:.20, pass:!!msg, description:"Billing message with exact subject exists", reason:"missing Billing message with exact subject" },
+        { id:"email", weight:.25, pass:!!msg && /\bnoah\.kim@example\.com\b/i.test(msg.body), description:"Body includes Noah email", reason:"message body missing Noah email" },
+        { id:"total", weight:.25, pass:!!msg && /(?<![\d.])\$?84\.20\b/.test(msg.body), description:"Body includes exact ORD-778 total", reason:"message body missing exact ORD-778 total 84.20" }
+      ];
+    }
+  },
+  "choredesk-l3-restock-ticket": {
+    difficulty: "L3 cross-page tedious",
+    instruction: "Ticket T-205 says Aster Lamp is delayed. Find the Aster Lamp SKU in Catalog, set its Inventory reorder quantity to 12, then add an internal note to T-205 saying 'reordered 12 AST-LAMP'.",
+    rubric(s) {
+      const inv = s.inventory.find(x => x.sku === "AST-LAMP");
+      const t = s.tickets.find(x => x.id === "T-205");
+      return [
+        { id:"visitedCatalog", weight:.15, pass:!!s.viewsVisited.catalog, description:"Opened Catalog view", reason:"never opened Catalog view" },
+        { id:"visitedInventory", weight:.15, pass:!!s.viewsVisited.inventory, description:"Opened Inventory view", reason:"never opened Inventory view" },
+        { id:"inventory", weight:.25, pass:Number(inv?.reorderQty) === 12 && s.audit.some(a => a.kind === "inventory-save" && a.id === "AST-LAMP" && a.via === "ui-event"), description:"AST-LAMP reorderQty=12 saved through UI", reason:`AST-LAMP reorderQty=${inv?.reorderQty}; missing UI save audit` },
+        { id:"ticketNote", weight:.25, pass:!!t && t.note.includes("reordered 12 AST-LAMP"), description:"T-205 note contains exact phrase", reason:"T-205 internal note missing exact phrase" },
+        { id:"ticketSaved", weight:.20, pass:s.audit.some(a => a.kind === "ticket-save" && a.id === "T-205" && a.via === "ui-event"), description:"T-205 saved through UI", reason:"no UI save audit for T-205" }
+      ];
+    }
+  }
+};
+const SEED = {
+  tickets: [
+    { id:"T-101", customer:"Rosa Patel", subject:"Where is my desk mat?", priority:"Normal", status:"Open", note:"" },
+    { id:"T-104", customer:"Iris Zhao", subject:"Invoice address typo", priority:"Low", status:"Open", note:"" },
+    { id:"T-205", customer:"Mateo Silva", subject:"Aster Lamp delayed", priority:"Normal", status:"Open", note:"" },
+    { id:"T-309", customer:"Noah Kim", subject:"Duplicate charge", priority:"Normal", status:"Open", note:"" }
+  ],
+  customers: [
+    { name:"Noah Kim", email:"noah.kim@example.com", tier:"Gold" },
+    { name:"Rosa Patel", email:"rosa.patel@example.com", tier:"Silver" },
+    { name:"Iris Zhao", email:"iris.zhao@example.com", tier:"Bronze" },
+    { name:"Mateo Silva", email:"mateo.silva@example.com", tier:"Gold" }
+  ],
+  orders: [
+    { id:"ORD-778", customer:"Noah Kim", item:"Cable Kit", total:"$84.20", status:"Paid" },
+    { id:"ORD-812", customer:"Rosa Patel", item:"Desk Mat", total:"$28.00", status:"Shipped" },
+    { id:"ORD-919", customer:"Mateo Silva", item:"Aster Lamp", total:"$134.50", status:"Delayed" }
+  ],
+  catalog: [
+    { item:"Aster Lamp", sku:"AST-LAMP", vendor:"Northwind Lighting" },
+    { item:"Desk Mat", sku:"DSK-MAT", vendor:"Woven Works" },
+    { item:"Cable Kit", sku:"CBL-KIT", vendor:"Copper Lane" }
+  ],
+  inventory: [
+    { sku:"AST-LAMP", onHand:0, reorderQty:0 },
+    { sku:"DSK-MAT", onHand:42, reorderQty:0 },
+    { sku:"CBL-KIT", onHand:6, reorderQty:0 }
+  ],
+  messages: [],
+  audit: [],
+  viewsVisited: {},
+  actionCount: 0,
+  startedAt: Date.now()
+};
+let state = loadState();
+function clone(x){ return JSON.parse(JSON.stringify(x)); }
+function loadState(){ const raw = localStorage.getItem(storeKey); return raw ? JSON.parse(raw) : clone(SEED); }
+function saveState(){ localStorage.setItem(storeKey, JSON.stringify(state)); }
+function countAction(kind){ state.actionCount = (state.actionCount || 0) + 1; state.lastAction = kind; }
+function audit(entry){ state.audit.push({ ts:new Date().toISOString(), via:"ui-event", trusted:!!entry.trusted, ...entry }); saveState(); grade(false); }
+function rubric(){ return (TASKS[task] || TASKS["choredesk-l1-ticket-priority"]).rubric(state); }
+function validate(){
+  const items = rubric();
+  const total = items.reduce((a,x)=>a+x.weight,0) || 1;
+  const got = items.filter(x=>x.pass).reduce((a,x)=>a+x.weight,0);
+  const reward = Math.round((got / total) * 1000) / 1000;
+  const done = reward === 1;
+  const bad = items.filter(x=>!x.pass).map(x=>x.reason || `${x.id} failed`);
+  return { reward, done, message: done ? "all deterministic checks passed" : bad.join("; "), info: { task, seed, run, actionCount:state.actionCount||0, elapsedMs:Date.now()-(state.startedAt||Date.now()), viewsVisited:state.viewsVisited, rubric:items, state } };
+}
+function observation(){
+  return { goal:TASKS[task]?.instruction, url:location.href, activeView:view, title:document.title, reward:window.__taskReward, actionCount:state.actionCount||0, viewsVisited:state.viewsVisited, focusedElementBid:document.activeElement?.getAttribute("bid") || null };
+}
+function setVerdict(verdict, reason, validation = validate()){
+  window.__taskVerdict = verdict;
+  window.__taskReason = reason;
+  window.__taskValidation = validation;
+  window.__taskReward = validation.reward;
+  window.__taskTerminated = validation.done;
+  window.__taskTruncated = (state.actionCount || 0) >= Number(params.get("maxSteps") || 9999);
+  window.__taskInfo = validation.info;
+  localStorage.setItem(verdictKey, JSON.stringify({ id: task, run, seed, verdict, reason, validation, state, ts: Date.now() }));
+  document.getElementById("verdict").textContent = verdict;
+  document.getElementById("verdict").className = "verdict " + verdict;
+  document.getElementById("reward").textContent = `reward=${validation.reward} actions=${state.actionCount || 0}`;
+  document.getElementById("reason").textContent = reason.join("\n");
+}
+function grade(showPass=true){ const v = validate(); if (!v.done) setVerdict(showPass ? "FAIL" : "PENDING", v.info.rubric.filter(x=>!x.pass).map(x=>`${x.id}: ${x.reason}`), v); else setVerdict("PASS", [v.message], v); return v; }
+function route(v){ view = v; state.viewsVisited[v] = (state.viewsVisited[v] || 0) + 1; countAction("route:" + v); saveState(); history.replaceState(null, "", `?task=${encodeURIComponent(task)}&run=${encodeURIComponent(run)}&seed=${seed}&nonce=${nonce}&view=${encodeURIComponent(view)}`); render(); }
+function renderShell(){
+  const spec = TASKS[task] || TASKS["choredesk-l1-ticket-priority"];
+  document.getElementById("taskId").textContent = task;
+  document.getElementById("difficulty").textContent = spec.difficulty;
+  document.getElementById("instruction").textContent = spec.instruction;
+  const nav = document.querySelector(".tabs"); nav.innerHTML = "";
+  for (const v of views) { const a = document.createElement("a"); a.href="#"; a.className="tab"; a.textContent=v[0].toUpperCase()+v.slice(1); a.dataset.view=v; a.setAttribute("data-task-role", "nav-tab"); if(v===view)a.setAttribute("aria-current","page"); a.addEventListener("click", e=>{e.preventDefault(); route(v);}); nav.appendChild(a); }
+}
+function render(){
+  renderShell(); bidCounter = 0;
+  const root = document.getElementById("view"); root.innerHTML = "";
+  if (view === "tickets") root.appendChild(tableEditor("tickets", state.tickets, ["id","customer","subject","priority","status","note"], saveTicket));
+  if (view === "customers") root.appendChild(readTable(state.customers, ["name","email","tier"]));
+  if (view === "orders") root.appendChild(readTable(state.orders, ["id","customer","item","total","status"]));
+  if (view === "catalog") root.appendChild(readTable(state.catalog, ["item","sku","vendor"]));
+  if (view === "inventory") root.appendChild(tableEditor("inventory", state.inventory, ["sku","onHand","reorderQty"], saveInventory));
+  if (view === "messages") root.appendChild(messageComposer());
+  if (view === "audit") root.appendChild(readTable(state.audit.map((entry,i)=>({ n:i+1, entry: JSON.stringify(entry) })), ["n","entry"]));
+  assignBids(); grade(false);
+}
+function readTable(rows, cols){
+  const wrap = document.createElement("div"); wrap.className="card";
+  wrap.innerHTML = `<table><thead><tr>${cols.map(c=>`<th>${c}</th>`).join("")}</tr></thead><tbody></tbody></table>`;
+  const tb = wrap.querySelector("tbody");
+  for (const r of rows) { const tr = document.createElement("tr"); tr.innerHTML = cols.map(c => `<td>${escapeHtml(r[c] ?? "")}</td>`).join(""); tb.appendChild(tr); }
+  return wrap;
+}
+function tableEditor(kind, rows, cols, onSave){
+  const wrap = document.createElement("div"); wrap.className="card";
+  wrap.innerHTML = `<table><thead><tr>${cols.map(c=>`<th>${c}</th>`).join("")}<th>save</th></tr></thead><tbody></tbody></table>`;
+  const tb = wrap.querySelector("tbody");
+  rows.forEach((r) => {
+    const key = r.id || r.sku; const tr = document.createElement("tr"); tr.dataset.itemId = key; if (r.id) tr.dataset.ticketId = r.id; if (r.sku) tr.dataset.sku = r.sku;
+    for (const c of cols) {
+      const td = document.createElement("td");
+      const stableAttrs = `${r.id ? `data-ticket-id="${r.id}"` : ""} ${r.sku ? `data-sku="${r.sku}"` : ""}`;
+      if (c === "priority") td.innerHTML = `<select data-item-id="${key}" ${stableAttrs} data-col="${c}" aria-label="${key} priority"><option>Low</option><option>Normal</option><option>High</option></select>`;
+      else if (c === "status") td.innerHTML = `<select data-item-id="${key}" ${stableAttrs} data-col="${c}" aria-label="${key} status"><option>Open</option><option>In Progress</option><option>Waiting</option><option>Closed</option></select>`;
+      else if (c === "note") td.innerHTML = `<input data-item-id="${key}" ${stableAttrs} data-col="${c}" aria-label="${key} internal note">`;
+      else if (c === "reorderQty") td.innerHTML = `<input type="number" data-item-id="${key}" ${stableAttrs} data-col="${c}" aria-label="${key} reorder quantity">`;
+      else td.textContent = r[c];
+      tr.appendChild(td);
+    }
+    const save = document.createElement("td"); save.innerHTML = `<button data-save-id="${key}" data-task-role="save-button" aria-label="save ${kind} ${key}">Save</button>`; tr.appendChild(save); tb.appendChild(tr);
+  });
+  wrap.querySelectorAll("[data-item-id]").forEach(el => { const row = rows.find(r => (r.id || r.sku) === el.dataset.itemId); el.value = row[el.dataset.col]; el.addEventListener("input", () => { row[el.dataset.col] = el.value; }); });
+  wrap.querySelectorAll("[data-save-id]").forEach(btn => btn.addEventListener("click", e => onSave(rows.find(r => (r.id || r.sku) === btn.dataset.saveId), e)));
+  return wrap;
+}
+function saveTicket(t, e){ countAction("save-ticket"); saveState(); audit({ kind:"ticket-save", id:t.id, trusted:e?.isTrusted }); render(); }
+function saveInventory(i, e){ countAction("save-inventory"); i.reorderQty = Number(i.reorderQty); saveState(); audit({ kind:"inventory-save", id:i.sku, trusted:e?.isTrusted }); render(); }
+function messageComposer(){
+  const wrap = document.createElement("div"); wrap.className="card";
+  wrap.innerHTML = `<label>From <input id="msgFrom" aria-label="message from" value="Agent"></label><br><br><label>To <select id="msgTo"><option>Billing</option><option>Support</option><option>Warehouse</option></select></label><br><br><label>Subject <input id="msgSubject" aria-label="message subject"></label><br><br><label>Body <textarea id="msgBody" aria-label="message body"></textarea></label><br><button id="sendMsg" data-task-role="send-message">Send message</button><h3>Sent messages</h3><div id="sent"></div>`;
+  wrap.querySelector("#sendMsg").addEventListener("click", e => { countAction("send-message"); state.messages.push({ from:msgFrom.value, to:msgTo.value, subject:msgSubject.value, body:msgBody.value, trusted:e.isTrusted }); audit({ kind:"message-send", id:msgSubject.value, trusted:e.isTrusted }); render(); });
+  wrap.querySelector("#sent").appendChild(readTable(state.messages, ["from","to","subject","body"]));
+  return wrap;
+}
+function assignBids(){ document.querySelectorAll("button,input,select,textarea,a[href]").forEach(el => { const id = `b${(++bidCounter).toString(36)}`; el.setAttribute("bid", id); el.setAttribute("data-bid", id); const r = el.getBoundingClientRect(); el.setAttribute("browsergym_visibility_ratio", (r.width > 0 && r.height > 0 ? 1 : 0).toString()); }); }
+function escapeHtml(s){ return String(s).replace(/[&<>"']/g, c => ({"&":"&amp;","<":"&lt;",">":"&gt;","\"":"&quot;","'":"&#39;"}[c])); }
+async function runCheat(){ const fn = window.ChoreDeskCheats?.run?.[task]; if (fn) await fn(); else for (const step of (window.ChoreDeskCheats?.[task] || [])) console.log("cheat step", step); grade(true); }
+document.getElementById("reset").addEventListener("click", () => { localStorage.removeItem(storeKey); localStorage.removeItem(verdictKey); state = loadState(); render(); });
+document.getElementById("grade").addEventListener("click", () => grade(true));
+document.getElementById("cheat").addEventListener("click", () => runCheat());
+window.__task = { id:task, run, seed, goal:TASKS[task]?.instruction, difficulty:TASKS[task]?.difficulty, step:() => { const v=grade(false); return { observation:observation(), reward:v.reward, terminated:v.done, truncated:window.__taskTruncated, info:v.info }; }, reset:() => { localStorage.removeItem(storeKey); state=loadState(); render(); return { observation:observation(), info:{task,run,seed} }; }, rubric:() => rubric(), cheat:runCheat, validate, teardown:()=>undefined };
+window.__choredesk = window.__task;
+window.__choredesk.cheatHelpers = { route: async (v) => { route(v); await new Promise(r => setTimeout(r, 0)); }, set: (sel, value) => { const el = document.querySelector(sel); if (!el) throw new Error('cheat selector missing: ' + sel); el.value = value; el.dispatchEvent(new Event('input', { bubbles:true })); el.dispatchEvent(new Event('change', { bubbles:true })); } };
+state.viewsVisited[view] = (state.viewsVisited[view] || 0) + 1; saveState(); render();
+})();
+</script>
+</body>
diff --git a/test-suite/task-manifest.json b/test-suite/task-manifest.json
new file mode 100644
index 0000000..22e9590
--- /dev/null
+++ b/test-suite/task-manifest.json
@@ -0,0 +1,416 @@
+[
+  {
+    "id": "choredesk-l1-ticket-priority",
+    "site": "choredesk",
+    "startUrl": "scenarios/choredesk/index.html?task=choredesk-l1-ticket-priority&run=$RUN_ID&seed=$SEED",
+    "difficulty": "L1",
+    "category": "enterprise-crud",
+    "intent": "Set ticket T-104 priority to High and status to In Progress, then save it.",
+    "expectedActions": {
+      "min": 3,
+      "max": 6
+    },
+    "requires": [
+      "snapshot",
+      "click",
+      "select/change",
+      "form editing"
+    ],
+    "grader": {
+      "type": "inPageProgrammatic",
+      "expression": "JSON.stringify(window.__taskValidation || {v:window.__taskVerdict,r:window.__taskReason})",
+      "passValue": "PASS"
+    },
+    "browserGymLike": {
+      "observation": [
+        "dom",
+        "accessibility_tree",
+        "screenshot_optional"
+      ],
+      "actionSpace": [
+        "click",
+        "type",
+        "select",
+        "evaluate_optional"
+      ],
+      "reset": "navigate startUrl with fresh $RUN_ID or click Reset run",
+      "stepReturn": "obs,reward,terminated,truncated,info"
+    },
+    "viewport": {
+      "width": 1280,
+      "height": 720
+    },
+    "slowMoMs": 1000,
+    "timeoutMs": 5000,
+    "locale": null,
+    "timezoneId": null,
+    "humanBaseline": null,
+    "actionSubsets": [
+      "bid",
+      "nav",
+      "tab",
+      "infeas",
+      "chat"
+    ],
+    "rewardShape": "binary",
+    "taskId": "choredesk.L1.ticket-priority",
+    "seed": 101,
+    "maxSteps": 8,
+    "allowedActions": [
+      "click",
+      "fill",
+      "select_option",
+      "goto",
+      "report_infeasible",
+      "send_msg_to_user"
+    ],
+    "setupUrl": "scenarios/choredesk/index.html?task=choredesk-l1-ticket-priority&run=$RUN_ID&seed=$SEED",
+    "validateScript": "scenarios/choredesk/index.html#window.__choredesk.validate()",
+    "cheatScript": "scenarios/choredesk/cheats.js#choredesk-l1-ticket-priority",
+    "tags": [
+      "form-fill",
+      "ticket",
+      "select"
+    ],
+    "rubric": [
+      {
+        "id": "priorityHigh",
+        "weight": 0.33,
+        "description": "T-104 priority is High"
+      },
+      {
+        "id": "statusInProgress",
+        "weight": 0.33,
+        "description": "T-104 status is In Progress"
+      },
+      {
+        "id": "saved",
+        "weight": 0.34,
+        "description": "T-104 was saved"
+      }
+    ],
+    "goal": "Set ticket T-104 priority to High and status to In Progress, then save it.",
+    "goalObject": [
+      {
+        "role": "user",
+        "content": [
+          {
+            "type": "text",
+            "text": "Set ticket T-104 priority to High and status to In Progress, then save it."
+          }
+        ]
+      }
+    ],
+    "gradeSource": "page"
+  },
+  {
+    "id": "choredesk-l2-refund-message",
+    "site": "choredesk",
+    "startUrl": "scenarios/choredesk/index.html?task=choredesk-l2-refund-message&run=$RUN_ID&seed=$SEED",
+    "difficulty": "L2",
+    "category": "cross-page-memory",
+    "intent": "Resolve a customer billing task by looking up customer email and order total on separate pages, then composing a message with exact required fields.",
+    "expectedActions": {
+      "min": 8,
+      "max": 14
+    },
+    "requires": [
+      "multi-page navigation",
+      "table lookup",
+      "working memory",
+      "form composition"
+    ],
+    "grader": {
+      "type": "inPageProgrammatic",
+      "expression": "JSON.stringify(window.__taskValidation || {v:window.__taskVerdict,r:window.__taskReason})",
+      "passValue": "PASS"
+    },
+    "browserGymLike": {
+      "observation": [
+        "dom",
+        "accessibility_tree",
+        "screenshot_optional"
+      ],
+      "actionSpace": [
+        "click",
+        "type",
+        "select",
+        "evaluate_optional"
+      ],
+      "reset": "navigate startUrl with fresh $RUN_ID or click Reset run",
+      "stepReturn": "obs,reward,terminated,truncated,info"
+    },
+    "viewport": {
+      "width": 1280,
+      "height": 720
+    },
+    "slowMoMs": 1000,
+    "timeoutMs": 5000,
+    "locale": null,
+    "timezoneId": null,
+    "humanBaseline": null,
+    "actionSubsets": [
+      "bid",
+      "nav",
+      "tab",
+      "infeas",
+      "chat"
+    ],
+    "rewardShape": "binary",
+    "taskId": "choredesk.L2.refund-message",
+    "seed": 202,
+    "maxSteps": 16,
+    "allowedActions": [
+      "click",
+      "fill",
+      "select_option",
+      "goto",
+      "report_infeasible",
+      "send_msg_to_user"
+    ],
+    "setupUrl": "scenarios/choredesk/index.html?task=choredesk-l2-refund-message&run=$RUN_ID&seed=$SEED",
+    "validateScript": "scenarios/choredesk/index.html#window.__choredesk.validate()",
+    "cheatScript": "scenarios/choredesk/cheats.js#choredesk-l2-refund-message",
+    "tags": [
+      "lookup",
+      "cross-page-memory",
+      "message-compose"
+    ],
+    "rubric": [
+      {
+        "id": "messageSubject",
+        "weight": 0.34,
+        "description": "Billing message subject matches"
+      },
+      {
+        "id": "email",
+        "weight": 0.33,
+        "description": "Body includes Noah email"
+      },
+      {
+        "id": "total",
+        "weight": 0.33,
+        "description": "Body includes exact order total"
+      }
+    ],
+    "goal": "Resolve a customer billing task by looking up customer email and order total on separate pages, then composing a message with exact required fields.",
+    "goalObject": [
+      {
+        "role": "user",
+        "content": [
+          {
+            "type": "text",
+            "text": "Resolve a customer billing task by looking up customer email and order total on separate pages, then composing a message with exact required fields."
+          }
+        ]
+      }
+    ],
+    "gradeSource": "page"
+  },
+  {
+    "id": "choredesk-l3-restock-ticket",
+    "site": "choredesk",
+    "startUrl": "scenarios/choredesk/index.html?task=choredesk-l3-restock-ticket&run=$RUN_ID&seed=$SEED",
+    "difficulty": "L3",
+    "category": "cross-page-state-update",
+    "intent": "Use ticket text to identify an item, map item to SKU in Catalog, update Inventory reorder quantity, and add exact internal note back on the ticket.",
+    "expectedActions": {
+      "min": 12,
+      "max": 20
+    },
+    "requires": [
+      "multi-page navigation",
+      "entity resolution",
+      "state mutation",
+      "exact text note",
+      "audit-save behavior"
+    ],
+    "grader": {
+      "type": "inPageProgrammatic",
+      "expression": "JSON.stringify(window.__taskValidation || {v:window.__taskVerdict,r:window.__taskReason})",
+      "passValue": "PASS"
+    },
+    "browserGymLike": {
+      "observation": [
+        "dom",
+        "accessibility_tree",
+        "screenshot_optional"
+      ],
+      "actionSpace": [
+        "click",
+        "type",
+        "select",
+        "evaluate_optional"
+      ],
+      "reset": "navigate startUrl with fresh $RUN_ID or click Reset run",
+      "stepReturn": "obs,reward,terminated,truncated,info"
+    },
+    "viewport": {
+      "width": 1280,
+      "height": 720
+    },
+    "slowMoMs": 1000,
+    "timeoutMs": 5000,
+    "locale": null,
+    "timezoneId": null,
+    "humanBaseline": null,
+    "actionSubsets": [
+      "bid",
+      "nav",
+      "tab",
+      "infeas",
+      "chat"
+    ],
+    "rewardShape": "binary",
+    "taskId": "choredesk.L3.restock-ticket",
+    "seed": 303,
+    "maxSteps": 24,
+    "allowedActions": [
+      "click",
+      "fill",
+      "select_option",
+      "goto",
+      "report_infeasible",
+      "send_msg_to_user"
+    ],
+    "setupUrl": "scenarios/choredesk/index.html?task=choredesk-l3-restock-ticket&run=$RUN_ID&seed=$SEED",
+    "validateScript": "scenarios/choredesk/index.html#window.__choredesk.validate()",
+    "cheatScript": "scenarios/choredesk/cheats.js#choredesk-l3-restock-ticket",
+    "tags": [
+      "entity-resolution",
+      "cross-page-state-update",
+      "exact-note"
+    ],
+    "rubric": [
+      {
+        "id": "inventory",
+        "weight": 0.4,
+        "description": "AST-LAMP reorder quantity is 12 and saved"
+      },
+      {
+        "id": "ticketNote",
+        "weight": 0.4,
+        "description": "T-205 note contains exact phrase"
+      },
+      {
+        "id": "audit",
+        "weight": 0.2,
+        "description": "Both save audit entries exist"
+      }
+    ],
+    "goal": "Use ticket text to identify an item, map item to SKU in Catalog, update Inventory reorder quantity, and add exact internal note back on the ticket.",
+    "goalObject": [
+      {
+        "role": "user",
+        "content": [
+          {
+            "type": "text",
+            "text": "Use ticket text to identify an item, map item to SKU in Catalog, update Inventory reorder quantity, and add exact internal note back on the ticket."
+          }
+        ]
+      }
+    ],
+    "gradeSource": "page"
+  },
+  {
+    "id": "mini-shop-red-second-cheapest",
+    "site": "mini-shop",
+    "startUrl": "fixtures/sites/mini-shop/index.html?task=mini-shop-red-second-cheapest&run=$RUN_ID&seed=$SEED",
+    "difficulty": "L2",
+    "category": "task-completion",
+    "intent": "Add the second-cheapest red product to cart, checkout, and leave the generated order id visible.",
+    "expectedActions": {
+      "min": 5,
+      "max": 9
+    },
+    "requires": [
+      "filter/sort reasoning",
+      "cart state",
+      "checkout flow"
+    ],
+    "grader": {
+      "type": "inPageProgrammatic",
+      "expression": "JSON.stringify(window.__taskValidation || {v:window.__taskVerdict,r:window.__taskReason})",
+      "passValue": "PASS"
+    },
+    "browserGymLike": {
+      "observation": [
+        "dom",
+        "accessibility_tree",
+        "screenshot_optional"
+      ],
+      "actionSpace": [
+        "click",
+        "read_table",
+        "evaluate_optional"
+      ],
+      "reset": "navigate startUrl with fresh $RUN_ID",
+      "stepReturn": "obs,reward,terminated,truncated,info"
+    },
+    "viewport": {
+      "width": 1280,
+      "height": 720
+    },
+    "slowMoMs": 1000,
+    "timeoutMs": 5000,
+    "locale": null,
+    "timezoneId": null,
+    "humanBaseline": null,
+    "actionSubsets": [
+      "bid",
+      "nav",
+      "tab",
+      "infeas",
+      "chat"
+    ],
+    "rewardShape": "binary",
+    "taskId": "mini-shop.L2.red-second-cheapest",
+    "seed": 404,
+    "maxSteps": 12,
+    "allowedActions": [
+      "click",
+      "goto",
+      "report_infeasible",
+      "send_msg_to_user"
+    ],
+    "setupUrl": "fixtures/sites/mini-shop/index.html?task=mini-shop-red-second-cheapest&run=$RUN_ID&seed=$SEED",
+    "validateScript": "fixtures/sites/mini-shop/grader.js#MiniShopGrader.validate",
+    "cheatScript": "fixtures/sites/mini-shop/cheats.js#mini-shop-red-second-cheapest",
+    "tags": [
+      "shopping",
+      "sort",
+      "cart",
+      "checkout"
+    ],
+    "rubric": [
+      {
+        "id": "addedTarget",
+        "weight": 0.4,
+        "description": "Cart includes second-cheapest red product"
+      },
+      {
+        "id": "noWrongItems",
+        "weight": 0.3,
+        "description": "Cart has no wrong products"
+      },
+      {
+        "id": "checkedOut",
+        "weight": 0.3,
+        "description": "Checkout produced order id"
+      }
+    ],
+    "goal": "Add the second-cheapest red product to cart, checkout, and leave the generated order id visible.",
+    "goalObject": [
+      {
+        "role": "user",
+        "content": [
+          {
+            "type": "text",
+            "text": "Add the second-cheapest red product to cart, checkout, and leave the generated order id visible."
+          }
+        ]
+      }
+    ],
+    "gradeSource": "both"
+  }
+]
diff --git a/test-suite/task-manifest.schema.json b/test-suite/task-manifest.schema.json
new file mode 100644
index 0000000..b8031fc
--- /dev/null
+++ b/test-suite/task-manifest.schema.json
@@ -0,0 +1,59 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "title": "pi-chrome long-horizon task manifest",
+  "description": "BrowserGym-inspired task manifest: setup URL, deterministic seed, goal object, action subsets, reward contract, optional cheat recipe, and in-page validate hook.",
+  "type": "array",
+  "items": {
+    "type": "object",
+    "required": ["id", "taskId", "site", "startUrl", "setupUrl", "difficulty", "category", "goal", "goalObject", "grader", "actionSubsets", "maxSteps"],
+    "additionalProperties": true,
+    "properties": {
+      "id": { "type": "string" },
+      "taskId": { "type": "string" },
+      "site": { "type": "string" },
+      "seed": { "type": "integer" },
+      "startUrl": { "type": "string", "description": "Relative URL. $RUN_ID and $SEED should be replaced by runner for isolated deterministic state." },
+      "setupUrl": { "type": "string" },
+      "validateScript": { "type": "string" },
+      "cheatScript": { "type": "string" },
+      "viewport": { "type": "object", "properties": { "width": { "type": "integer" }, "height": { "type": "integer" } } },
+      "slowMoMs": { "type": "integer" },
+      "timeoutMs": { "type": "integer" },
+      "locale": { "type": ["string", "null"] },
+      "timezoneId": { "type": ["string", "null"] },
+      "difficulty": { "enum": ["L1", "L2", "L3"] },
+      "category": { "enum": ["enterprise-crud", "cross-page-memory", "cross-page-state-update", "task-completion"] },
+      "gradeSource": { "enum": ["page", "harness", "both"] },
+      "tags": { "type": "array", "items": { "type": "string" } },
+      "intent": { "type": "string" },
+      "goal": { "type": "string" },
+      "goalObject": { "type": "array", "items": { "type": "object" } },
+      "expectedActions": { "type": "object", "required": ["min", "max"], "properties": { "min": { "type": "integer" }, "max": { "type": "integer" } } },
+      "maxSteps": { "type": "integer" },
+      "humanBaseline": { "type": ["number", "null"], "minimum": 0, "maximum": 1 },
+      "actionSubsets": { "type": "array", "items": { "enum": ["chat", "infeas", "bid", "coord", "nav", "tab"] } },
+      "allowedActions": { "type": "array", "items": { "type": "string" } },
+      "requires": { "type": "array", "items": { "type": "string" } },
+      "rewardShape": { "enum": ["binary", "graded", "shaped"] },
+      "rubric": { "type": "array", "items": { "type": "object", "required": ["id", "weight", "description"], "properties": { "id": { "type": "string" }, "weight": { "type": "number" }, "description": { "type": "string" } } } },
+      "grader": {
+        "type": "object",
+        "required": ["type", "expression", "passValue"],
+        "properties": {
+          "type": { "enum": ["inPageProgrammatic"] },
+          "expression": { "type": "string", "description": "Should evaluate to BrowserGym-like validate object or legacy {v,r}." },
+          "passValue": { "type": "string" }
+        }
+      },
+      "browserGymLike": {
+        "type": "object",
+        "properties": {
+          "observation": { "type": "array", "items": { "type": "string" } },
+          "actionSpace": { "type": "array", "items": { "type": "string" } },
+          "reset": { "type": "string" },
+          "stepReturn": { "const": "obs,reward,terminated,truncated,info" }
+        }
+      }
+    }
+  }
+}