researcher: row 97 round 2 build packet (#1591)

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

agents/researcher/work/queue-97/r2/build.md

@@ -0,0 +1,57 @@
+# Row 97 (#1591) round 2: the board timer gets a positive control
+
+Owner: Researcher. Reviewers: Darkwing and Filbert. Round 1 change request: Darkwing, comment 28083 (rev 865).
+
+Base 6b0215aa (rebased from 3bd3f85c). Nothing under `packages/webui` changed between the two, so
+the three candidate paths carry over; `browser-edge.test.mjs` is byte-identical to round 1.
+`build.patch` is `git diff 6b0215aa` of the three paths (sha256
+efe65504a1b890ae575b0d49ae3c00f09c9f229897ba61d2cac44e1e1b56bf19). `candidate-manifest.sha256` has
+repo-relative paths and verifies against the candidate worktree
+`/mnt/storage/scratch/researcher/q97` (checked after the runs). No source under `src/` changes.
+
+## The finding
+
+The stub filed a 10 s timer as 'board' only when `f.name === 'refresh'`. Test 1's three
+`boardPolls() === 0` checks had no positive control, so a board timer armed through an arrow
+callback (the shape bus.js uses) was filed as 'bus' and the checks passed.
+
+## Both fixes, as Darkwing allowed
+
+1. **Classification.** The stub files a timer by the script that armed it: `new Error().stack`
+   matched against `/app.js`, else 'bus'. The callback's shape no longer matters.
+2. **Positive control.** Test 1 waits for `window.__polls.includes('board')` once the toggle is on,
+   so the zero checks sit between a proven armed timer and a proven absent one.
+
+The README sentence says the timers are filed by the script that armed them and that one is armed
+once the toggle is on.
+
+## Mutants (`mutants.py`, outputs `mut-base.txt`, `mut-candidate.txt`)
+
+Baseline = 6b0215aa unmodified (round 1's tests are not the baseline here; Darkwing's 13a was alive
+on them).
+
+| mutant | baseline | candidate |
+|---|---|---|
+| 13 schedule polls with the toggle off | ALIVE | killed |
+| 13a Darkwing: 13 with `setTimeout(() => refresh(), 10000)` | ALIVE | killed |
+| 13b the arrow callback alone | ALIVE | ALIVE (equivalent: same behaviour) |
+| 27 bus schedule polls signed out | ALIVE | killed |
+| 12 refresh reads with the toggle off | killed | killed |
+| 26 bus render without its signed-out guard | killed | killed |
+| 19 toggle on reads nothing | STALLED (30 s) | killed (5.8 s) |
+| 14 refresh without its signed-out guard | ALIVE | ALIVE (equivalent: `api()` already returns when signed out) |
+
+Candidate 6 of 8, the two survivors equivalent; baseline 2 of 8. Sources restored byte for byte
+(sha256 checked at the end of both runs).
+
+## Suites
+
+- webui `node --test tests/`: 101 pass, 0 fail, on the real `app.js` and `bus.js` (`suites.txt`).
+- Not rerun: control-board and the `scripts/test-*.sh` suites. Only webui test files and its README
+  changed, and nothing outside `packages/webui` moved between 3bd3f85c and 6b0215aa on these paths.
+  Round 1's results for those stand; Sage's gate covers a rerun.
+- After the runs no process had a cwd under my scratch (/proc check, 0 found).
+
+## Secrets scan
+
+`build.patch` and `mutants.py`: no token, secret, password, api key or private key marker.

agents/researcher/work/queue-97/r2/build.patch

@@ -0,0 +1,177 @@
+diff --git a/packages/webui/README.md b/packages/webui/README.md
+index 6656f1ff..ce56d59f 100644
+--- a/packages/webui/README.md
++++ b/packages/webui/README.md
+@@ -566,11 +566,19 @@ claim that has an address and one that has none. They check the held roles
+ with the machine sessions off by default and no `/api/board` request, the
+ toggle on and off and its persistence across reloads, Message addressing
+ the send form to each role, no action without send, and that every fetch
+-asked for `redirect: 'manual'`. Then a redirect answer on a board read, an
+-`agents` read and a send: one line, no NetworkError, nothing stale left, no
+-further request on a route change, Refresh or the toggle, and the draft
+-kept. `send-browser.test.mjs` also checks the sign-out after its redirect
+-receipt and that a reload works again. Since the held roles lead the Board,
++asked for `redirect: 'manual'`. The tests record every 10 s poll timer the
++page arms, filed by the script that armed it, and assert that none is armed
++for the board while the toggle is off, even across Pause and Resume, and that
++one is armed once it is on. Then a redirect answer on each of six
++requests, the board read (`/api/board`), the `agents` read, `/api/conversations`,
++`/api/seen`, `/api/queue` and the send: one line, no NetworkError, nothing
++stale left, and no further request or new poll timer on a route change,
++Refresh, the toggle, a forced Pause click or a poll tick, and the draft kept. In
++that test the poll timers fire after 150 ms, so the check spans several
++ticks. `browser-edge.test.mjs` fails in 5 s, and closes the browser and
++servers, if the toggle makes no `/api/board` request. `send-browser.test.mjs`
++also checks the sign-out after its redirect receipt and that a reload works
++again. Since the held roles lead the Board,
+ the shell test's no-scroll check at 1440px uses a 1000px-tall window.
+ 
+ Bus tests (`bus.test.mjs`, `bus-browser.test.mjs`) run the routes and the
+diff --git a/packages/webui/tests/board-held.test.mjs b/packages/webui/tests/board-held.test.mjs
+index 19b875a8..6815a4bf 100644
+--- a/packages/webui/tests/board-held.test.mjs
++++ b/packages/webui/tests/board-held.test.mjs
+@@ -5,7 +5,11 @@
+ // through its reader session. A script added before the page loads records every fetch the page
+ // makes, answers /api/config with send on (unless the URL has ?nosend), answers POST
+ // /api/bus/message from a queue, and answers any path in window.__redirect the way fetch does
+-// under redirect: 'manual' (type opaqueredirect, status 0).
++// under redirect: 'manual' (type opaqueredirect, status 0). It also records every 10 s poll timer
++// the page arms (window.__polls, 'board' for app.js and 'bus' for bus.js). The second test has the
++// timers fire after 150 ms, so its "no further request" check spans real poll ticks, and both
++// tests assert on the timers themselves: a guard that only the fetch behind it backs up would
++// otherwise change nothing the page sends.
+ import { test } from 'node:test';
+ import assert from 'node:assert/strict';
+ import { mkdtempSync, rmSync } from 'node:fs';
+@@ -29,9 +33,17 @@ const businesses = {
+   },
+ };
+ 
+-const STUB = `(() => {
++const stub = fast => `(() => {
+   const real = window.fetch.bind(window);
+-  window.__reqs = []; window.__answers = []; window.__redirect = [];
++  window.__reqs = []; window.__answers = []; window.__redirect = []; window.__polls = [];
++  // app.js and bus.js both poll with a 10000 delay. A timer is filed by the script that armed it, read
++  // from the call stack, not from the shape of the callback, which either script may change.
++  const timeout = window.setTimeout.bind(window);
++  window.setTimeout = (f, ms, ...rest) => {
++    if (ms !== 10000) return timeout(f, ms, ...rest);
++    window.__polls.push(/\\/app\\.js[:?]/.test(new Error().stack) ? 'board' : 'bus');
++    return timeout(f, ${fast} ? 150 : ms, ...rest);
++  };
+   const redirect = () => ({ type: 'opaqueredirect', ok: false, status: 0, json: async () => { throw new SyntaxError('no body'); } });
+   window.fetch = async (url, init = {}) => {
+     const u = String(url), path = u.split('?')[0];
+@@ -49,7 +61,7 @@ const STUB = `(() => {
+   };
+ })();`;
+ 
+-async function setup(machine) {
++async function setup(machine, fast = false) {
+   const root = mkdtempSync(join(tmpdir(), 'webui-board-held-'));
+   const store = new Store(root), b0 = new Broker({ store, businesses });
+   // pm's run has an address; coder's claim has none, which the Board still lists.
+@@ -59,22 +71,24 @@ async function setup(machine) {
+   const bus = views({ call: async (verb, args = {}) => structuredClone(b0.request(reader, { verb, args })) });
+   const f = await fixture(), b = await browser({ machine });
+   const web = await startServer({ port: 0, board: f.boardURL, bus }), base = `http://127.0.0.1:${web.address().port}/`;
+-  await b.call('Page.addScriptToEvaluateOnNewDocument', { source: STUB });
++  await b.call('Page.addScriptToEvaluateOnNewDocument', { source: stub(fast) });
+   await b.viewport(1280);
+   const wait = expr => b.evaluate(`(async()=>{for(let i=0;i<100;i++){if(${expr})return true;await new Promise(r=>setTimeout(r,50))}throw new Error('Condition timed out: '+${JSON.stringify(expr)})})()`);
+   const text = sel => b.evaluate(`document.querySelector(${JSON.stringify(sel)})?.textContent ?? null`);
+   const shown = sel => b.evaluate(`(()=>{const e=document.querySelector(${JSON.stringify(sel)});return !!e && !e.hidden && getComputedStyle(e).display !== 'none' && !!e.getClientRects().length})()`);
+   const reqs = () => b.evaluate('window.__reqs');
++  const polls = () => b.evaluate('window.__polls');
+   const settle = () => new Promise(r => setTimeout(r, 600));
+   const done = async () => { await b.close(); await close(web); await f.close(); store.close(); rmSync(root, { recursive: true, force: true }); };
+-  return { b, base, wait, text, shown, reqs, settle, done };
++  return { b, base, wait, text, shown, reqs, polls, settle, done };
+ }
+ 
+ const HELD = 'document.querySelectorAll("#held table tbody tr").length === 2';
+ const SESSIONS = 'document.querySelector("table.sessions [data-open=\'proj/agent1\']")';
+ 
+ test('the Board: held roles with no machine sessions, a claim with no address, the toggle kept across a reload, and Message addressed to the role', { timeout: 120000 }, async () => {
+-  const { b, base, wait, text, shown, reqs, settle, done } = await setup(false);
++  const { b, base, wait, text, shown, reqs, polls, settle, done } = await setup(false);
++  const boardPolls = async () => (await polls()).filter(p => p === 'board').length;
+   try {
+     // Off by default: the held roles show, the machine scan is never read.
+     await b.navigate(base);
+@@ -95,11 +109,19 @@ test('the Board: held roles with no machine sessions, a claim with no address, t
+     await settle();
+     assert.equal((await reqs()).some(r => r.path === '/api/board'), false);
+     assert.equal(await b.evaluate(`!!${SESSIONS}`), false);
++    // The board is not polled either: no timer for it, even after Pause and Resume, which re-arm one.
++    assert.equal(await boardPolls(), 0);
++    assert.ok((await polls()).includes('bus'), 'the bus views still poll');
++    await b.evaluate('document.querySelector("#pause").click(), document.querySelector("#pause").click(), true');
++    await settle();
++    assert.equal(await boardPolls(), 0);
+ 
+     // On: the machine sessions are read and shown, and the choice is kept for the next load.
+     await b.evaluate('document.querySelector("#show-machine").click(), true');
+     await wait(SESSIONS);
+     assert.equal(await shown('#board-view'), true);
++    // Positive control for the zero checks above and below: with the toggle on, the board does arm its timer.
++    await wait("window.__polls.includes('board')");
+     assert.equal(await b.evaluate('localStorage.getItem("mosaic-console-board")'), '{"machine":true}');
+     await b.navigate(`${base}?on`);
+     await wait(SESSIONS);
+@@ -117,6 +139,7 @@ test('the Board: held roles with no machine sessions, a claim with no address, t
+     assert.equal(await b.evaluate('document.querySelector("#show-machine").checked'), false);
+     await settle();
+     assert.equal((await reqs()).some(r => r.path === '/api/board'), false);
++    assert.equal(await boardPolls(), 0);
+ 
+     // Message opens the send form addressed to that role, with the caret in the message.
+     for (const role of ['coder', 'pm']) {
+@@ -138,7 +161,8 @@ test('the Board: held roles with no machine sessions, a claim with no address, t
+ });
+ 
+ test('a redirect answer on a Board read, a bus read, a conversation read, a board POST, a module read and a send signs the page out: one line, nothing stale, no further reads', { timeout: 120000 }, async () => {
+-  const { b, base, wait, text, shown, reqs, settle, done } = await setup(true);
++  // The poll timers fire after 150 ms here, so the 600 ms settle below spans several ticks.
++  const { b, base, wait, text, shown, reqs, polls, settle, done } = await setup(true, true);
+   const LINE = 'Signed out. Reload the page to sign in.';
+   const out = async () => {
+     await wait('!document.querySelector("#signed-out").hidden');
+@@ -153,12 +177,17 @@ test('a redirect answer on a Board read, a bus read, a conversation read, a boar
+     assert.equal(await b.evaluate('document.querySelector("#held").children.length'), 0);
+     for (const sel of ['#held-section', '#board-view', '#board-head', '#error']) assert.equal(await shown(sel), false, sel);
+     assert.equal(await b.evaluate('document.querySelector("#refresh").disabled'), true);
+-    // No further Console fetch: not on a route change, a Refresh, the toggle or the poll.
+-    const n = (await reqs()).length;
++    // No further Console fetch and no new poll timer: not on a route change, a Refresh, the toggle
++    // or a poll tick (a timer armed before the sign-out may still fire, and finds nothing to read).
++    const n = (await reqs()).length, armed = (await polls()).length;
+     await b.evaluate(`location.hash = location.hash === '#/agents' ? '#/inbox' : '#/agents', true`);
+     await b.evaluate('document.querySelector("#refresh").click(), document.querySelector("#show-machine").click(), document.querySelector("#show-machine").click(), true');
++    // Pause is disabled when signed out. Enabled by force for a pause and a resume, its handlers must still
++    // arm nothing: the guard in each schedule() is the only thing between the resume and a timer.
++    await b.evaluate('(() => { const p = document.querySelector("#pause"); p.disabled = false; p.click(); p.click(); p.disabled = true; return true; })()');
+     await settle();
+     assert.equal((await reqs()).length, n);
++    assert.equal((await polls()).length, armed);
+     assert.equal(await b.evaluate('!!document.querySelector("#bus-view h1")'), false);
+     assert.deepEqual((await reqs()).filter(r => r.redirect !== 'manual'), []);
+   };
+diff --git a/packages/webui/tests/browser-edge.test.mjs b/packages/webui/tests/browser-edge.test.mjs
+index 446c9cc2..efb261ff 100644
+--- a/packages/webui/tests/browser-edge.test.mjs
++++ b/packages/webui/tests/browser-edge.test.mjs
+@@ -26,7 +26,10 @@ test('browser edge states: loading, empty, malformed, stale, hostile/long values
+     assert.match(await b.evaluate('document.querySelector("#status").textContent'), /Machine sessions hidden/);
+     await b.evaluate('document.querySelector("#show-machine").click()');
+     assert.match(await b.evaluate('document.querySelector("#status").textContent'), /Loading/);
+-    await boardRequested;
++    // Fails in 5 s if the toggle never reads the board; the finally below then closes the browser and servers.
++    let timer;
++    try { await Promise.race([boardRequested, new Promise((_, reject) => { timer = setTimeout(() => reject(new Error('Show machine sessions made no /api/board request in 5 s')), 5000); })]); }
++    finally { clearTimeout(timer); }
+     holdBoard = false; boardResponse.end(JSON.stringify(index));
+     await wait('document.querySelector("#sessions").textContent.includes("No sessions")');
+     assert.equal(await b.evaluate('document.querySelector("#waiting-count").textContent'), '0');

agents/researcher/work/queue-97/r2/candidate-manifest.sha256

@@ -0,0 +1,3 @@
+01213bffd915afa42dc0c531c8605364e19466686701ddadad514346bbf5851a  packages/webui/README.md
+32d01fbac94fff0fed337aa47182f78a623eaf3ad5845a880a3dccfbe1c70586  packages/webui/tests/board-held.test.mjs
+fc3cc8fa7388a6bae669bbaf60c21947f803620ff140af7e33d798a326716fc4  packages/webui/tests/browser-edge.test.mjs

agents/researcher/work/queue-97/r2/mut-base.txt

@@ -0,0 +1,11 @@
+worktree /mnt/storage/scratch/researcher/q97-base
+start 2026-10-13T12:36:46Z
+13 schedule polls with the toggle off: ALIVE (exit 0, 8.7s, board-held.test.mjs browser-edge.test.mjs)
+13a Darkwing: schedule polls with the toggle off, arrow callback: ALIVE (exit 0, 8.0s, board-held.test.mjs browser-edge.test.mjs)
+13b arrow callback alone (equivalent): ALIVE (exit 0, 8.1s, board-held.test.mjs browser-edge.test.mjs)
+27 bus schedule polls signed out: ALIVE (exit 0, 8.1s, board-held.test.mjs browser-edge.test.mjs)
+12 refresh reads with the toggle off: killed (exit 1, 6.9s, board-held.test.mjs browser-edge.test.mjs)
+14 refresh without its signed-out guard: ALIVE (exit 0, 8.0s, board-held.test.mjs browser-edge.test.mjs)
+26 bus render without its signed-out guard: killed (exit 1, 4.2s, board-held.test.mjs browser-edge.test.mjs)
+19 toggle on reads nothing: STALLED (exit timeout, 30.0s, browser-edge.test.mjs)
+killed 2 of 8; sources restored byte for byte: True

agents/researcher/work/queue-97/r2/mut-candidate.txt

@@ -0,0 +1,11 @@
+worktree /mnt/storage/scratch/researcher/q97
+start 2026-10-13T12:35:43Z
+13 schedule polls with the toggle off: killed (exit 1, 7.6s, board-held.test.mjs browser-edge.test.mjs)
+13a Darkwing: schedule polls with the toggle off, arrow callback: killed (exit 1, 7.4s, board-held.test.mjs browser-edge.test.mjs)
+13b arrow callback alone (equivalent): ALIVE (exit 0, 8.7s, board-held.test.mjs browser-edge.test.mjs)
+27 bus schedule polls signed out: killed (exit 1, 4.8s, board-held.test.mjs browser-edge.test.mjs)
+12 refresh reads with the toggle off: killed (exit 1, 6.9s, board-held.test.mjs browser-edge.test.mjs)
+14 refresh without its signed-out guard: ALIVE (exit 0, 8.6s, board-held.test.mjs browser-edge.test.mjs)
+26 bus render without its signed-out guard: killed (exit 1, 4.9s, board-held.test.mjs browser-edge.test.mjs)
+19 toggle on reads nothing: killed (exit 1, 5.8s, browser-edge.test.mjs)
+killed 6 of 8; sources restored byte for byte: True

agents/researcher/work/queue-97/r2/mutants.py

@@ -0,0 +1,72 @@
+#!/usr/bin/env python3
+# Row 97 (#1591) round 2 mutants: round 1's list plus Darkwing's (13 with the arrow callback, comment 28083)
+# and the arrow shape alone, an equivalent mutant. Each edit must match exactly once. Every run starts in its own process
+# group and the group is killed when the run ends or times out, so no node or chromium is left
+# behind. The mutated file is restored from the bytes read before the edit, and every touched
+# file's sha256 is checked against its start at the end.
+# Usage: mutants.py <worktree> <out.txt> [timeout-seconds]
+import hashlib, os, signal, subprocess, sys, time
+
+WT, OUT = sys.argv[1], sys.argv[2]
+TIMEOUT = int(sys.argv[3]) if len(sys.argv) > 3 else 120
+A = 'packages/webui/src/public/app.js'
+B = 'packages/webui/src/public/bus.js'
+BOTH = ['tests/board-held.test.mjs', 'tests/browser-edge.test.mjs']
+EDGE = ['tests/browser-edge.test.mjs']
+
+# (name, file, old, new, tests, per-run timeout)
+M = [
+  ('13 schedule polls with the toggle off', A, 'if (!paused && machine && !out()) timer', 'if (!paused && !out()) timer', BOTH, TIMEOUT),
+  ('13a Darkwing: schedule polls with the toggle off, arrow callback', A, 'if (!paused && machine && !out()) timer = setTimeout(refresh, 10000)', 'if (!paused && !out()) timer = setTimeout(() => refresh(), 10000)', BOTH, TIMEOUT),
+  ('13b arrow callback alone (equivalent)', A, 'timer = setTimeout(refresh, 10000)', 'timer = setTimeout(() => refresh(), 10000)', BOTH, TIMEOUT),
+  ('27 bus schedule polls signed out', B, 'if (!paused() && !out()) timer', 'if (!paused()) timer', BOTH, TIMEOUT),
+  ('12 refresh reads with the toggle off', A, 'if (busy || !machine || out()) return;', 'if (busy || out()) return;', BOTH, TIMEOUT),
+  ('14 refresh without its signed-out guard', A, 'if (busy || !machine || out()) return;', 'if (busy || !machine) return;', BOTH, TIMEOUT),
+  ('26 bus render without its signed-out guard', B, "    clearTimeout(timer);\n    if (out()) return;\n", "    clearTimeout(timer);\n", BOTH, TIMEOUT),
+  ('19 toggle on reads nothing', A, "fresh(); refresh(); } else hideMachine();", "fresh(); } else hideMachine();", EDGE, 30),
+]
+
+env = dict(os.environ, TMPDIR=os.environ.get('TMPDIR', '/mnt/storage/scratch/tmp'))
+sha = lambda p: hashlib.sha256(open(os.path.join(WT, p), 'rb').read()).hexdigest()
+start = {p: sha(p) for p in (A, B)}
+
+def run(tests, limit):
+  t0 = time.time()
+  p = subprocess.Popen(['node', '--test', *tests], cwd=os.path.join(WT, 'packages/webui'), env=env,
+                       stdout=subprocess.PIPE, stderr=subprocess.STDOUT, start_new_session=True, text=True)
+  try:
+    out, _ = p.communicate(timeout=limit)
+    code = p.returncode
+  except subprocess.TimeoutExpired:
+    code, out = 'timeout', ''
+  finally:
+    try: os.killpg(p.pid, signal.SIGKILL)
+    except ProcessLookupError: pass
+    try: p.communicate(timeout=5)
+    except Exception: pass
+  return code, round(time.time() - t0, 1)
+
+with open(OUT, 'w') as f:
+  def say(s): print(s); f.write(s + '\n'); f.flush()
+  say(f'worktree {WT}')
+  say(f'start {time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())}')
+  killed = 0
+  for name, path, old, new, tests, limit in M:
+    full = os.path.join(WT, path)
+    orig = open(full, 'rb').read()
+    src = orig.decode()
+    if src.count(old) != 1:
+      say(f'{name}: NO MATCH ({src.count(old)}), skipped'); continue
+    try:
+      open(full, 'w').write(src.replace(old, new))
+      code, secs = run(tests, limit)
+    finally:
+      open(full, 'wb').write(orig)
+    # A timeout is a stall, not a kill: the file never exited by itself.
+    dead = code not in (0, 'timeout')
+    killed += dead
+    verdict = 'killed' if dead else 'STALLED' if code == 'timeout' else 'ALIVE'
+    say(f'{name}: {verdict} (exit {code}, {secs}s, {" ".join(t.split("/")[-1] for t in tests)})')
+  ok = all(sha(p) == start[p] for p in (A, B))
+  say(f'killed {killed} of {len(M)}; sources restored byte for byte: {ok}')
+  sys.exit(0 if ok else 1)

agents/researcher/work/queue-97/r2/suites.txt

@@ -0,0 +1,5 @@
+== webui 12:34:42Z
+ℹ tests 101
+ℹ pass 101
+ℹ fail 0
+exit 0