@@ -22,7 +22,18 @@ If that prints meetings, you are connected. If it says no server, tell the perso
 overrule review . --playbook changes --cycles 1 --json
 ```
 
-The council reads your working tree, finds the change with `git status` and `git diff`, and judges that change, not the project. The exit status is the answer: **0** approved, **2** objections remain, **3** checks failed, **4** verification incomplete, **1** something went wrong.
+The council reads your working tree, finds the change with `git status` and `git diff`, and judges that change, not the project. The exit status is the answer:
+
+| Status | Meaning |
+| --- | --- |
+| 0 | Approved. Every voter approved and nothing is open. |
+| 2 | A member voted against the candidate. Real dissent; read the objections. |
+| 3 | A command the council ran on the candidate failed. |
+| 4 | Verification incomplete: there is a candidate, but the vote did not finish. |
+| 5 | Unsettled: the meeting ran out of turns or stopped moving, with nobody dissenting. Buy it more turns rather than treating it as disagreement. |
+| 1 | Something went wrong. Read the message. |
+
+Nothing is printed on standard output until the verdict; progress goes to standard error. Do not pipe the output into a parser before the command exits.
 
 **Ask a question that has no single right answer.** Design choices, migration plans, tradeoffs:
 
@@ -39,8 +50,11 @@ Read the whole report, not just the verdict. The value is in the objections, whi
 - **Approved.** Say so, and name anything the council flagged anyway.
 - **Objections remain.** Do not treat this as failure. Report the objections to the person in their own words, say which ones you agree with, and fix what is worth fixing.
 - **Checks failed.** A command the council ran on the candidate failed. That is a fact, not an opinion; act on it.
+- **Unsettled.** The clock ran out, not the argument. `overrule resume <id> --cycles 3` carries on with the same council and record instead of paying for the openings again, and `--note "<what to settle>"` goes in as the owner's own message.
 - **Error.** Report what the message said. Do not retry in a loop.
 
+With `--json` you get the record, not only the prose: `verdict` (with `dissent`, `stopReason`, and the vote counts), `ballots[]`, `objections[]` with each one's `resolvingCondition`, the `candidate` text, `members`, `metrics`, and the meeting `url`. Quote from those fields rather than scraping the Markdown.
+
 Never present the council's findings as your own verification. Say where they came from. If a finding contradicts what you know about the code, check it yourself before passing it on: the members read the tree, but they can be wrong, and two of them agreeing does not make a thing true.
 
 ## Rules
@@ -56,12 +70,14 @@ Never present the council's findings as your own verification. Say where they ca
 | Option | What it does |
 | --- | --- |
 | `--playbook <id>` | Start from a report template. `overrule playbooks` lists them: security, changes, bugs, dependencies, architecture, tests, performance, readiness. |
-| `--members <names>` | Pick the council by connection name. The default seats everything ready, up to eight. |
+| `--members <names>` | Pick the council by connection name. **The default seats everything ready, up to eight, which can include members billed per token on the owner's API accounts.** The command prints the seats and which of them bill before the first call. |
 | `--cycles <n>` | Turns per member. 1 is a quick read, 2 is a real argument, more is a long meeting. |
 | `--deep-research` | Members that can browse research the subject before taking a position, citing sources. Slower and more expensive. |
-| `--attach <file>` | Give the council a document: a plan, a diff, a spec. Repeatable. |
+| `--attach <file>` | Give the council a document: a plan, a diff, a spec. Repeatable. Documents are put to the members as material to weigh on its own terms, separate from the brief, which is the owner's framing. Attach the source rather than your summary of it. |
+| `--exclude <path>` | Tell members a path is out of scope, so their positions are not anchored on your notes. An instruction to the council, not a boundary the machine enforces. Repeatable. |
 | `--no-wait` | Print the meeting id and return immediately; pick it up later with `overrule watch <id>`. |
-| `--json` | One object: `{ id, status, verdict, final, error, url }`. Use this, and read `final`. |
+| `--json` | The whole record as one object: verdict, ballots, objections, candidate, members, metrics, url. Use this. |
+| `--turn-limit <s>` | Seconds one member may take. Raise it above the 180 second default when research is on, or a slow member loses its seat. |
 
 ## If you are working on Overrule itself
 
 
@@ -87,7 +87,7 @@ node bin/overrule.mjs ask "Is this migration plan sound?" --deep-research --json
 node bin/overrule.mjs list
 ```
 
-`review` attaches a folder read-only and takes a report template; `ask` holds a plain meeting. Both wait for the verdict, print the report on standard output, and set the exit status: **0** approved, **2** objections remain, **3** checks failed, **4** verification incomplete, **1** error. With `--json` the output is one object with the verdict, the final text, and the meeting id, which is what an agent should read. `--no-wait` prints the id instead of waiting, and `watch` or `show` picks it up later. `--attach` sends a file, `--members` and `--drafter` pick the council, and `--level workspace-write` lets the drafter implement rather than only report.
+`review` attaches a folder read-only and takes a report template; `ask` holds a plain meeting. Both wait for the verdict, print the report on standard output, and set the exit status: **0** approved, **2** objections remain because a member voted against, **3** checks failed, **4** verification incomplete, **5** unsettled because the meeting ran out of turns with nobody dissenting, **1** error. `resume <id> --cycles 3` buys an unsettled meeting more turns without paying for the openings again, and `reconvene <id> "<what next>"` opens a new session on the same record. With `--json` the output is one object holding the verdict (with the vote counts, whether anyone dissented, and why the meeting closed), the ballots, the objections with the condition that resolves each one, the candidate text, the members, and the meeting id, which is what an agent should read. `--no-wait` prints the id instead of waiting, and `watch` or `show` picks it up later. `--attach` sends a file, `--members` and `--drafter` pick the council, and `--level workspace-write` lets the drafter implement rather than only report.
 
 A coding agent can therefore finish a change and hand it straight to the council:
 
 
@@ -10,6 +10,8 @@ const USAGE = `overrule — convene a council of models from the command line.
 
   overrule ask "<brief>" [options]        Hold a meeting about a question.
   overrule review [path] [options]        Review a workspace, usually with a report template.
+  overrule resume <id> [options]          Pick a stopped meeting up, optionally with more turns.
+  overrule reconvene <id> "<what next>"   Reopen a closed meeting with a new instruction.
   overrule watch <id>                     Follow a meeting that is already running.
   overrule show <id> [--json]             Print a finished meeting's report.
   overrule list [--json]                  List recent meetings.
@@ -26,6 +28,8 @@ Options
   --research             Members that can browse may search the web.
   --deep-research        They research the subject before taking a position. Implies --research.
   --attach <file>        Attach a document. Repeatable.
+  --exclude <path>       Tell members a path is out of scope. Repeatable. An instruction to
+                         the council, not a boundary the machine enforces.
   --check <command>      A command run on the candidate commit at a write level. Repeatable.
   --network              Allow the network for the implementer and the checks at a write level.
   --implement-limit <s>  Seconds the implementer may take at a write level.
@@ -34,20 +38,23 @@ Options
   --call-limit <n>       Model calls the whole session may make.
   --acknowledge          Accept the warnings this meeting needs (secrets in the tree, uncommitted work).
   --acknowledge-full-access   Required for --level full-access.
+  --note <text>          A message to the council, delivered with resume.
   --no-wait              Print the meeting id and exit instead of waiting.
   --json                 Print one JSON object instead of a report.
   --quiet                Print the report only, with no progress.
   --url <address>        Server to use. Default ${DEFAULT_URL}.
   --code <pairing code>  Pair first; needed only when the server is not on this machine.
 
-Exit status: 0 approved, 2 objections remain, 3 checks failed, 4 verification incomplete, 1 error.`;
+Exit status: 0 approved, 2 objections remain (a member voted against), 3 checks failed,
+4 verification incomplete, 5 unsettled (out of turns or stalled, with nobody dissenting;
+resume with more cycles), 1 error.`;
 
-const VERDICT_EXIT = { approved: 0, objections: 2, 'checks-failed': 3, incomplete: 4 };
+const VERDICT_EXIT = { approved: 0, objections: 2, 'checks-failed': 3, incomplete: 4, unsettled: 5 };
 const FLAGS = new Set(['research', 'deep-research', 'network', 'json', 'quiet', 'no-wait', 'acknowledge', 'acknowledge-full-access', 'help', 'version']);
-const MANY = new Set(['attach', 'check']);
+const MANY = new Set(['attach', 'check', 'exclude']);
 
 export function parseArgs(argv) {
-  const options = { _: [], attach: [], check: [] };
+  const options = { _: [], attach: [], check: [], exclude: [] };
   for (let i = 0; i < argv.length; i++) {
     const arg = argv[i];
     if (!arg.startsWith('--')) { options._.push(arg); continue; }
@@ -114,7 +121,7 @@ function chooseMembers(bootstrap, options) {
   chosen = chosen.slice(0, 8);
   const drafter = options.drafter ? chosen.find(p => p.name.toLowerCase() === options.drafter.toLowerCase() || p.id === options.drafter) : chosen[0];
   if (!drafter) fail(`The drafter "${options.drafter}" is not one of the members.`);
-  return { participantIds: chosen.map(p => p.id), drafterId: drafter.id, names: chosen.map(p => p.name) };
+  return { participantIds: chosen.map(p => p.id), drafterId: drafter.id, names: chosen.map(p => p.name), seats: chosen };
 }
 
 async function attachDocuments(client, files, log) {
@@ -137,7 +144,7 @@ function workspacePayload(options) {
   if (!level) return undefined;
   if (level === 'full-access' && !options['acknowledge-full-access']) fail('Full access needs --acknowledge-full-access: every member with tools can do anything your account can do on that machine.');
   return {
-    path, level, checks: options.check, network: options.network === true || level === 'full-access',
+    path, level, checks: options.check, exclude: options.exclude, network: options.network === true || level === 'full-access',
     acknowledgeSecrets: Boolean(options.acknowledge), acknowledgeDirty: Boolean(options.acknowledge), acknowledgeFullAccess: Boolean(options['acknowledge-full-access']),
     ...(options['implement-limit'] ? { implementTimeout: Number(options['implement-limit']) } : {}),
   };
@@ -155,6 +162,28 @@ async function waitFor(client, id, log) {
   }
 }
 
+// Everything a calling agent would otherwise have to scrape out of the Markdown: the ballots, the objections with the
+// condition that resolves each one, the candidate text, and why the meeting closed.
+function machineReadable(run, url) {
+  const record = run.record || {};
+  const draft = [...(run.entries || [])].reverse().find(e => e.phase === 'draft' && e.status === 'complete' && !e.superseded && e.candidateVersion === run.candidateVersion);
+  const name = id => (run.participants || []).find(p => p.id === id)?.name || id;
+  return {
+    id: run.id, status: run.status, url: `${url}/#${run.id}`,
+    verdict: record.verdict ? { ...record.verdict, votes: record.votes || null, voters: record.voters ?? null, missingBallots: record.missingBallots ?? null } : null,
+    stopReason: run.stopReason || null, cycles: run.cycles, session: run.session || 1,
+    final: run.final || '', candidate: draft?.text || '', candidateVersion: run.candidateVersion,
+    ballots: (run.entries || []).filter(e => e.phase === 'ratify' && e.status === 'complete' && !e.superseded && e.candidateVersion === run.candidateVersion)
+      .map(e => ({ member: e.name, vote: e.fields?.vote || 'abstain', reason: e.fields?.reason || '', objections: (e.fields?.objections || []).map(o => ({ claim: o.claim, resolvingCondition: o.condition })) })),
+    objections: (run.issues || []).map(issue => ({ id: issue.id, raisedBy: name(issue.raisedBy), against: name(issue.against), claim: issue.claim, resolvingCondition: issue.condition, status: issue.status })),
+    openObjections: (run.issues || []).filter(i => i.status === 'open').length,
+    members: (run.participants || []).map(p => p.name), drafter: name(run.drafterId),
+    workspace: run.workspace ? { path: run.workspace.path, level: run.workspace.level, origin: run.workspace.origin || null, head: run.workspace.head || null } : null,
+    research: Boolean(run.research), deepResearch: Boolean(run.deepResearch),
+    metrics: record.metrics || null, error: run.error || null,
+  };
+}
+
 function report(run) {
   const verdict = run.record?.verdict;
   return [
@@ -181,12 +210,36 @@ export async function main(argv, { out = console.log, err = console.error } = {}
     for (const run of runs.slice(0, 20)) out(`${run.id.slice(0, 8)}  ${String(run.status).padEnd(11)} ${new Date(run.createdAt).toLocaleString()}  ${(run.prompt || '').replace(/\s+/g, ' ').slice(0, 70)}`);
     return 0;
   }
+  if (command === 'resume' || command === 'reconvene') {
+    const id = options._[1] || fail(`overrule ${command} needs a meeting id.`);
+    await client.connect();
+    const found = (await client.call('/api/runs')).find(run => run.id === id || run.id.startsWith(id)) || fail(`No meeting starts with ${id}.`);
+    const body = {
+      ...(options.cycles ? { cycles: number(options.cycles, 'cycles') } : {}),
+      ...(options.revisions === undefined ? {} : { revisions: number(options.revisions, 'revisions') }),
+      ...(options['turn-limit'] ? { timeoutSeconds: number(options['turn-limit'], 'turn-limit') } : {}),
+      ...(options['call-limit'] ? { maxCalls: number(options['call-limit'], 'call-limit') } : {}),
+      ...(options['time-limit'] ? { maxDurationSeconds: number(options['time-limit'], 'time-limit') * 60 } : {}),
+      ...(options.research || options['deep-research'] ? { research: true } : {}), ...(options['deep-research'] ? { deepResearch: true } : {}),
+    };
+    if (command === 'reconvene') {
+      const text = options._.slice(2).join(' ').trim() || options.note || fail('Say what the council should take up next: overrule reconvene <id> "…".');
+      await client.call(`/api/runs/${found.id}/reconvene`, 'POST', { text, cycles: body.cycles ?? 2, maxCalls: body.maxCalls, maxDurationSeconds: body.maxDurationSeconds });
+    } else {
+      await client.call(`/api/runs/${found.id}/resume`, 'POST', { ...body, note: options.note || '' });
+    }
+    log(`${command === 'resume' ? 'Resumed' : 'Reconvened'} ${found.id.slice(0, 8)}.`);
+    if (options['no-wait']) { out(options.json ? JSON.stringify({ id: found.id, status: 'running' }, null, 2) : found.id); return 0; }
+    const finished = await waitFor(client, found.id, log);
+    out(options.json ? JSON.stringify(machineReadable(finished, client.url), null, 2) : report(finished));
+    return VERDICT_EXIT[finished.record?.verdict?.code] ?? (finished.status === 'complete' ? 0 : 1);
+  }
   if (command === 'show' || command === 'watch') {
     const id = options._[1] || fail(`overrule ${command} needs a meeting id.`);
     await client.connect();
     const found = (await client.call('/api/runs')).find(run => run.id === id || run.id.startsWith(id)) || fail(`No meeting starts with ${id}.`);
     const run = command === 'watch' ? await waitFor(client, found.id, log) : await client.call(`/api/runs/${found.id}`);
-    out(options.json ? JSON.stringify({ id: run.id, status: run.status, verdict: run.record?.verdict || null, final: run.final || '', error: run.error || null }, null, 2) : report(run));
+    out(options.json ? JSON.stringify(machineReadable(run, client.url), null, 2) : report(run));
     return VERDICT_EXIT[run.record?.verdict?.code] ?? (run.status === 'complete' ? 0 : 1);
   }
   if (command !== 'ask' && command !== 'review') fail(`Unknown command "${command}". Run overrule --help.`);
@@ -205,9 +258,12 @@ export async function main(argv, { out = console.log, err = console.error } = {}
   if (!prompt.trim()) fail(command === 'review' ? 'Give a brief, or pick one with --playbook.' : 'Give a brief: overrule ask "your question".');
 
   const attachmentIds = options.attach.length ? await attachDocuments(client, options.attach, log) : [];
-  log(`Convening ${members.names.join(', ')}${workspace ? ` on ${workspace.path} (${workspace.level})` : ''}`);
+  const billed = members.seats.filter(p => !bootstrap.types[p.type]?.cli);
+  const seating = `Seating ${members.seats.map(p => `${p.name}${bootstrap.types[p.type]?.cli ? '' : ' (billed per token)'}`).join(', ')}${workspace ? ` on ${workspace.path} (${workspace.level})` : ''}`;
+  if (!options.quiet) err(seating + (billed.length && !options.members ? `\n${billed.length} of these bill your API accounts per token. Pass --members to choose the council yourself.` : ''));
+  const { seats, names, ...meeting } = members;
   const run = await client.call('/api/runs', 'POST', {
-    prompt, ...members, cycles: options.cycles ? number(options.cycles, 'cycles') : 2,
+    prompt, ...meeting, cycles: options.cycles ? number(options.cycles, 'cycles') : 2,
     revisions: options.revisions === undefined ? 1 : number(options.revisions, 'revisions'),
     research: Boolean(options.research) || Boolean(options['deep-research']), deepResearch: Boolean(options['deep-research']),
     ...(options['turn-limit'] ? { timeoutSeconds: number(options['turn-limit'], 'turn-limit') } : {}),
@@ -218,12 +274,13 @@ export async function main(argv, { out = console.log, err = console.error } = {}
   log(`Meeting ${run.id.slice(0, 8)} started. Watch it at ${client.url}`);
   if (options['no-wait']) { out(options.json ? JSON.stringify({ id: run.id, status: run.status }, null, 2) : run.id); return 0; }
   const finished = await waitFor(client, run.id, log);
-  out(options.json ? JSON.stringify({ id: finished.id, status: finished.status, verdict: finished.record?.verdict || null, final: finished.final || '', error: finished.error || null, url: `${client.url}/#${finished.id}` }, null, 2) : report(finished));
+  out(options.json ? JSON.stringify(machineReadable(finished, client.url), null, 2) : report(finished));
   return VERDICT_EXIT[finished.record?.verdict?.code] ?? (finished.status === 'complete' ? 0 : 1);
 }
 
 if (import.meta.url === `file://${process.argv[1]}`) {
+  // process.exit truncates a piped stdout: set the status and let the process end once the report has been written.
   main(process.argv.slice(2))
-    .then(code => process.exit(code))
-    .catch(error => { console.error(error.expected ? error.message : `overrule: ${error.stack}`); process.exit(1); });
+    .then(code => { process.exitCode = code; })
+    .catch(error => { console.error(error.expected ? error.message : `overrule: ${error.stack}`); process.exitCode = 1; });
 }