docs/swarm/review-classifier.md

@@ -232,7 +232,7 @@ key.
 
 A dedicated forge user, separate from swarm-controller's site-admin account
 and from every agent's account. It's the reviewer someone requests, it
-authors every classifier review, and it pushes the branches the try-out section
+authors every classifier review, and it pushes the branches the try-out page
 creates.
 
 `swarm-controller` creates it as `review-classifier`, with the email
@@ -256,7 +256,7 @@ collaborator:
   allowlist, the account counts only once it's on that list, and read access
   is then enough for reviews.
 
-The try-out's **open a PR** pushes a branch to the repo itself, so it
+The try-out page's **open a PR** pushes a branch to the repo itself, so it
 needs **write** whatever the branch protection says.
 
 The repo also needs a rules file on its default branch (see

docs/swarm/services.md

@@ -311,5 +311,5 @@ with no restart. Every run spends money on that key.
 It needs neither the forge nor a hive on its host. It reaches the queue and the
 secret store by name, under its own `review-classifier` leaf. On a host
 without the store, that leaf is the one file to copy, as for the CI runner.
-The swarm UI's try-out drives it through swarm-controller; what a run is
+The swarm UI's try-out page drives it through swarm-controller; what a run is
 and what it costs are in [`review-classifier.md`](review-classifier.md).

docs/swarm/ui.md

@@ -17,7 +17,8 @@ domain, covers host-level detail for one hive.
 | `/jobs`                                        | the controller's job graph — where agent creation and credential mints show progress                                                   |
 | `/schedules`                                   | the scheduled prompts: add, edit, pause, resume and remove                                                                             |
 | `/issues`                                      | a cross-repo issue report                                                                                                              |
-| `/review`                                      | the review classifier: its recent runs, a try-out on a real PR, and its settings                                                       |
+| `/review-runs`                                 | the review classifier's recent runs: which PRs, what it answered, which failed                                                         |
+| `/review-tryout`                               | try a review classification against a real PR                                                                                          |
 
 Everything it shows comes from [`swarm-controller`](../../swarm-controller/README.md).
 An agent created here or with `swarmctl agent create` starts `paused`; set it
@@ -56,18 +57,9 @@ page. Selecting outside it, or Escape, closes it. You can stop or pause the
 agent, or link an account, without leaving the terminal. A subagent's tab opens
 its agent's details.
 
-### Review
+### Review runs
 
-`/review` is the one page for the [review classifier](review-classifier.md),
+`/review-runs` lists what the [review classifier](review-classifier.md) has
-with three sections in a column. **try a classification** and **classifier
-settings** come first, each folded to a single bar: the bar of the settings
-says what they're now, and what you typed in a section stays when you fold it.
-**recent runs** is under them and always open. The sections share nothing and
-ask for their own data.
-
-#### Recent runs
-
-The **recent runs** section lists what the [review classifier](review-classifier.md) has
 run, newest first, one card per run. A card shows the repo and PR, the head
 commit, whether the run was a try-out or a requested review, how long ago it
 started, how long it took, how many classifications ran, and the outcome as a
@@ -92,9 +84,9 @@ does. An empty list says there are no runs yet, and a failed load shows the reas
 controller gave. The page shows no token counts, cache counts or cost, and
 stores and posts nothing.
 
-#### Try a classification
+### Review try-out
 
-The **try a classification** section shows how the [review classifier](review-classifier.md)
+`/review-tryout` shows how the [review classifier](review-classifier.md)
 answers a classification on a real PR, before you put it in a repo's config.
 Give it a repo as `owner/name` and a PR number, then write a classification: a
 name, a description, a polarity (positive or negative) and a threshold from 0
@@ -138,9 +130,9 @@ first. If the classifier's account may not push to the repo, the page says so:
 a repo admin adds it as a collaborator with write access
 ([the service account](review-classifier.md#the-service-account)).
 
-#### Classifier settings
+### Classifier settings
 
-The **classifier settings** section shows the runtime, model and
+Under the try-out form, **classifier settings** shows the runtime, model and
 effort the [review classifier](review-classifier.md) runs with and how many
 requested reviews run at once, and lets you change them. The runtime is **API
 key** (claude on the Anthropic key) or **ACP** (opencode on a provider). On the

frontend/packages/swarm-ui/src/App.tsx

@@ -8,7 +8,8 @@ import { ComponentsPage } from "./pages/ComponentsPage.js";
 import { JobsPage } from "./pages/JobsPage.js";
 import { HivesPage } from "./pages/HivesPage.js";
 import { IssueReportPage } from "./pages/IssueReportPage.js";
-import { ReviewPage } from "./pages/review/ReviewPage.js";
+import { ReviewRunsPage } from "./pages/review-runs/ReviewRunsPage.js";
+import { ReviewTryOutPage } from "./pages/review-tryout/ReviewTryOutPage.js";
 import { SchedulesPage } from "./pages/schedules/SchedulesPage.js";
 import { StartPage } from "./pages/StartPage.js";
 import { Panel } from "./ui/panel/Panel.js";
@@ -36,7 +37,8 @@ export function App() {
         <Route path="/jobs" component={JobsPage} />
         <Route path="/schedules" component={SchedulesPage} />
         <Route path="/issues" component={IssueReportPage} />
-        <Route path="/review" component={ReviewPage} />
+        <Route path="/review-runs" component={ReviewRunsPage} />
+        <Route path="/review-tryout" component={ReviewTryOutPage} />
         <Route path="/components" component={ComponentsPage} />
         <Route component={NotFound} />
       </Switch>

frontend/packages/swarm-ui/src/pages/review-runs/ReviewRunsPage.css

@@ -0,0 +1,47 @@
+/* The run list: one card per run, the head row a button that opens it. */
+.ui-runs-list {
+  display: flex;
+  flex-direction: column;
+  gap: 0.75em;
+  margin-top: 1em;
+}
+.ui-runs-head {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: center;
+  gap: 0.5em 0.75em;
+  padding: 0;
+  border: 0;
+  background: none;
+  color: inherit;
+  font: inherit;
+  text-align: left;
+  cursor: pointer;
+}
+.ui-runs-head:focus-visible {
+  outline: 2px solid var(--purple);
+  outline-offset: 2px;
+}
+.ui-runs-title {
+  display: inline-flex;
+  align-items: baseline;
+  gap: 0.5em;
+}
+.ui-runs-meta {
+  color: var(--muted);
+}
+.ui-runs-head .ui-runs-meta {
+  flex-basis: 100%;
+}
+.ui-runs-body {
+  display: flex;
+  flex-direction: column;
+  gap: 0.6em;
+  margin-top: 0.5em;
+}
+.ui-runs-body p {
+  margin: 0;
+}
+.ui-runs-empty {
+  color: var(--muted);
+}

frontend/packages/swarm-ui/src/pages/review-runs/ReviewRunsPage.tsx

@@ -0,0 +1,162 @@
+// <ReviewRunsPage> — the review classifier's recent runs, newest first
+// (`docs/swarm/review-classifier.md`, "Recent runs"): which PRs it ran on,
+// what it answered, and which runs failed or were turned away. One card per
+// run; selecting a card expands it to the same result table the try-out page
+// shows. Reads `GET /api/review-classifier/runs`. The page stores nothing and
+// posts nothing, and shows no token counts, cache counts or cost.
+import { useState } from "preact/hooks";
+import { ApiErrorPanel } from "@hive/shared/api-error-panel.js";
+import { readApiError, type ProblemDetails } from "@hive/shared/api-error.js";
+import { Badge } from "@hive/shared/badge.js";
+import { Card } from "../../ui/card/Card.js";
+import { Panel } from "../../ui/panel/Panel.js";
+import {
+  RefreshIntervalPicker,
+  useRefreshInterval,
+  type RefreshIntervalMs,
+} from "../../ui/refresh-interval/RefreshInterval.js";
+import { RelativeTime } from "../../ui/relative-time/RelativeTime.js";
+import { TryOutResult } from "../review-tryout/TryOutResult.js";
+import {
+  OUTCOME_LABEL,
+  OUTCOME_TONE,
+  TRIGGER_LABEL,
+  classificationCount,
+  durationLabel,
+  type ClassifierRun,
+} from "./RunsTypes.js";
+import "./ReviewRunsPage.css";
+
+const RUNS = "/api/review-classifier/runs";
+const DEFAULT_INTERVAL_MS: RefreshIntervalMs = 30_000;
+
+function RunCard({
+  run,
+  open,
+  onToggle,
+}: {
+  run: ClassifierRun;
+  open: boolean;
+  onToggle: () => void;
+}) {
+  const bodyId = `review-run-${run.id}`;
+  return (
+    <Card class="ui-runs-card">
+      <button
+        type="button"
+        class="ui-runs-head"
+        aria-expanded={open}
+        aria-controls={bodyId}
+        onClick={onToggle}
+      >
+        <span class="ui-runs-title">
+          <strong>
+            {run.repo} #{run.pr}
+          </strong>
+          {run.head_sha ? <code>{run.head_sha.slice(0, 7)}</code> : null}
+        </span>
+        <Badge
+          tone={OUTCOME_TONE[run.outcome]}
+          value={OUTCOME_LABEL[run.outcome]}
+        />
+        <span class="ui-runs-meta">
+          {TRIGGER_LABEL[run.trigger]} ·{" "}
+          <RelativeTime epochMs={run.started_unix * 1000} /> · took{" "}
+          {durationLabel(run.started_unix, run.finished_unix)} ·{" "}
+          {classificationCount(run)}
+        </span>
+      </button>
+      {open ? (
+        <div class="ui-runs-body" id={bodyId}>
+          {run.detail ? <p class="ui-runs-detail">{run.detail}</p> : null}
+          {run.rows && run.rows.length > 0 ? (
+            <TryOutResult
+              result={run}
+              filterKey="swarm-ui:review-runs:table-filters"
+            />
+          ) : null}
+          {run.review_url ? (
+            <p class="ui-runs-link">
+              <a href={run.review_url} target="_blank" rel="noreferrer">
+                the posted review
+              </a>
+            </p>
+          ) : null}
+          {run.requester ? (
+            <p class="ui-runs-meta">Requested by {run.requester}.</p>
+          ) : null}
+        </div>
+      ) : null}
+    </Card>
+  );
+}
+
+export function ReviewRunsPage() {
+  const [intervalMs, setIntervalMs] =
+    useState<RefreshIntervalMs>(DEFAULT_INTERVAL_MS);
+  const [runs, setRuns] = useState<ClassifierRun[] | null>(null);
+  const [error, setError] = useState<ProblemDetails | null>(null);
+  // Which cards are expanded. Survives a refresh, so a list that updates
+  // under you does not fold the card you are reading.
+  const [open, setOpen] = useState<ReadonlySet<string>>(new Set());
+
+  async function load() {
+    try {
+      const res = await fetch(RUNS);
+      if (!res.ok) {
+        setError(await readApiError(res));
+        return;
+      }
+      setRuns((await res.json()) as ClassifierRun[]);
+      setError(null);
+    } catch (e: unknown) {
+      setError({ detail: String(e) });
+    }
+  }
+
+  useRefreshInterval(intervalMs, () => void load());
+
+  function toggle(id: string) {
+    setOpen((cur) => {
+      const next = new Set(cur);
+      if (!next.delete(id)) next.add(id);
+      return next;
+    });
+  }
+
+  return (
+    <Panel
+      title="review runs"
+      icon="📜"
+      actions={
+        <RefreshIntervalPicker
+          id="review-runs-refresh"
+          value={intervalMs}
+          onChange={setIntervalMs}
+        />
+      }
+    >
+      <p>
+        The PRs the review classifier ran on, newest first: what it answered,
+        and which runs failed or were turned away. Select a run to see its
+        result table.
+      </p>
+      {error ? (
+        <ApiErrorPanel context="the runs could not be loaded" problem={error} />
+      ) : null}
+      {runs !== null && runs.length === 0 && !error ? (
+        <p class="ui-runs-empty">No runs yet.</p>
+      ) : null}
+      <div class="ui-runs-list">
+        {(runs ?? []).map((run) => (
+          <RunCard
+            key={run.id}
+            run={run}
+            open={open.has(run.id)}
+            onToggle={() => toggle(run.id)}
+          />
+        ))}
+      </div>
+    </Panel>
+  );
+}

frontend/packages/swarm-ui/src/pages/review-settings/ClassifierSettings.tsx

@@ -10,7 +10,7 @@ import { useEffect, useState } from "preact/hooks";
 import { ApiErrorPanel } from "@hive/shared/api-error-panel.js";
 import { readApiError, type ProblemDetails } from "@hive/shared/api-error.js";
 import { Button } from "../../ui/button/Button.js";
-import { Disclosure } from "../../ui/disclosure/Disclosure.js";
+import { Panel } from "../../ui/panel/Panel.js";
 import { SelectField } from "../../ui/select-field/SelectField.js";
 import { TextField } from "../../ui/text-field/TextField.js";
 import {
@@ -20,7 +20,6 @@ import {
   changed,
   draftOf,
   settingsProblem,
-  settingsSummary,
   toSettings,
   type ClassifierSettingsBody,
   type Runtime,
@@ -55,9 +54,6 @@ export function ClassifierSettings() {
   const [saveError, setSaveError] = useState<ProblemDetails | null>(null);
   const [saving, setSaving] = useState(false);
   const [justSaved, setJustSaved] = useState(false);
-  // Folded until wanted: the settings are changed now and then, and the
-  // folded header says what they are.
-  const [open, setOpen] = useState(false);
 
   useEffect(() => {
     let cancelled = false;
@@ -118,13 +114,7 @@ export function ClassifierSettings() {
   const problem = draft === null ? null : settingsProblem(draft);
 
   return (
-    <Disclosure
+    <Panel title="classifier settings" icon="⚙️">
-      title="classifier settings"
-      icon="⚙️"
-      hint={saved ? settingsSummary(saved) : undefined}
-      open={open}
-      onToggle={setOpen}
-    >
       <p>
         The runtime, model and effort the review classifier runs with, and how
         many requested reviews run at once. The API key runtime runs claude on
@@ -244,6 +234,6 @@ export function ClassifierSettings() {
           ) : null}
         </form>
       ) : null}
-    </Disclosure>
+    </Panel>
   );
 }

frontend/packages/swarm-ui/src/pages/review-settings/SettingsTypes.ts

@@ -182,15 +182,3 @@ export function changed(
       d.priceCacheWrite.trim() !== now.priceCacheWrite)
   );
 }
-
-/** The settings in force as one short line, for a folded view of them: the
- * runtime and the model it runs, the effort when one is set, and how many
- * reviews run at once. The model is the provider's under ACP. */
-export function settingsSummary(s: ClassifierSettingsBody): string {
-  const model = s.runtime === "acp" ? (s.acp?.model ?? "") : s.model;
-  const parts = [s.runtime, model, s.effort ? `${s.effort} effort` : ""].filter(
-    (p) => p !== "",
-  );
-  parts.push(`${s.max_concurrent_reviews} at once`);
-  return parts.join(" · ");
-}

frontend/packages/swarm-ui/src/pages/review-tryout/ClassificationEditor.tsx

@@ -6,7 +6,7 @@ import { SelectField } from "../../ui/select-field/SelectField.js";
 import { TextAreaField } from "../../ui/text-area-field/TextAreaField.js";
 import { TextField } from "../../ui/text-field/TextField.js";
 import type { ClassificationDraft, Polarity } from "./TryOutTypes.js";
-import "./ReviewTryOut.css";
+import "./ReviewTryOutPage.css";
 
 const POLARITY_OPTIONS = [
   { value: "positive", label: "positive — the PR should match" },

frontend/packages/swarm-ui/src/pages/review-tryout/ReviewTryOutPage.css

@@ -0,0 +1,64 @@
+/* The form panel and the result panel under it, apart from each other. */
+.ui-tryout-page {
+  display: flex;
+  flex-direction: column;
+  gap: 1em;
+}
+/* The try-out form: the repo and PR on one line, then one block per
+   classification, then the actions. */
+.ui-tryout-form {
+  display: flex;
+  flex-direction: column;
+  gap: 1em;
+  margin-top: 1em;
+}
+.ui-tryout-target {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 0.75em;
+}
+.ui-tryout-target > * {
+  flex: 1 1 14em;
+}
+.ui-tryout-target > button {
+  flex: none;
+  align-self: flex-end;
+}
+.ui-tryout-loaded {
+  margin: 0;
+  color: var(--muted);
+}
+.ui-tryout-classification {
+  display: flex;
+  flex-direction: column;
+  gap: 0.6em;
+  padding: 0.75em;
+  border: 1px solid var(--border);
+  border-radius: 0.5em;
+}
+.ui-tryout-classification-row {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: flex-end;
+  gap: 0.75em;
+}
+.ui-tryout-classification-row > * {
+  flex: 1 1 12em;
+}
+.ui-tryout-classification-row > button {
+  flex: none;
+}
+.ui-tryout-actions {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: center;
+  gap: 0.75em;
+}
+.ui-tryout-hint {
+  margin: 0;
+  color: var(--muted);
+}
+.ui-tryout-running {
+  margin: 0;
+  color: var(--muted);
+}

frontend/packages/swarm-ui/src/pages/review-tryout/ReviewTryOutPage.tsx

@@ -0,0 +1,468 @@
+// <ReviewTryOutPage> — write a review classification and see how the model
+// answers it on a real PR, before it goes into a repo's config
+// (`docs/swarm/review-classifier.md`, "Try-out"). Pick a repo and a PR, write
+// one or more classifications, run: the page posts them to
+// swarm-controller's `POST /api/review-classifier/try-out`, which answers
+// `202 {id}`, and polls `GET .../try-out/{id}` until the run is `done` or
+// `failed`. A run takes a minute or more, so the page says it is running.
+//
+// The repo and PR fields offer the repos that have an open PR and the open PRs
+// of the repo (`GET /api/pulls/repos`, `GET /api/pulls?repo=…`), and stay free
+// text: the run also works on a merged or closed PR.
+//
+// **load the repo's rules** fills the rows from the rules on the repo's default
+// branch (`GET /api/review-classifier/rules`), to run as they are or to edit.
+//
+// **open a PR** posts the rows to `POST /api/review-classifier/open-pr`, which
+// has the classifier's forge account push a branch adding them to the repo's
+// rules file, with the operator as commit author, and links to the forge's page
+// for opening the PR from that branch.
+//
+// Under the form, the classifier's own settings: its model, effort and review
+// concurrency (`ClassifierSettings`).
+//
+// The inputs stay as typed after a run, so the operator can edit and run
+// again. Apart from the settings, the page keeps nothing you typed and no
+// result, and a run posts nothing to the forge; the only thing the browser
+// remembers is the result table's column filters, as for every table in the UI.
+import { useEffect, useState } from "preact/hooks";
+import { ApiErrorPanel } from "@hive/shared/api-error-panel.js";
+import { readApiError, type ProblemDetails } from "@hive/shared/api-error.js";
+import { Button } from "../../ui/button/Button.js";
+import { ConfirmDialog } from "../../ui/confirm-dialog/ConfirmDialog.js";
+import { Panel } from "../../ui/panel/Panel.js";
+import { TextField } from "../../ui/text-field/TextField.js";
+import { ClassifierSettings } from "../review-settings/ClassifierSettings.js";
+import { ClassificationEditor } from "./ClassificationEditor.js";
+import { TryOutResult } from "./TryOutResult.js";
+import {
+  draftFromRule,
+  hasTypedText,
+  prNumber,
+  repoProblem,
+  rowsProblem,
+  runProblem,
+  toBody,
+  type ClassificationDraft,
+  type OpenPrAnswer,
+  type OpenPull,
+  type RulesAnswer,
+  type TryOutResult as Result,
+} from "./TryOutTypes.js";
+import "./ReviewTryOutPage.css";
+
+const TRY_OUT = "/api/review-classifier/try-out";
+const RULES = "/api/review-classifier/rules";
+const OPEN_PR = "/api/review-classifier/open-pr";
+const PULL_REPOS = "/api/pulls/repos";
+const PULLS = "/api/pulls";
+// How long the repo field rests before its PRs are fetched, so typing a repo
+// does not ask for the PRs of each prefix of it.
+const PICK_DELAY_MS = 300;
+const POLL_MS = 2000;
+// More classifications than this are not offered a row. It is the try-out
+// endpoint's own cap, which answers a longer list with a 400 that the page
+// shows as it is.
+const MAX_CLASSIFICATIONS = 10;
+
+let nextKey = 0;
+function blankDraft(): ClassificationDraft {
+  return {
+    key: nextKey++,
+    name: "",
+    description: "",
+    polarity: "positive",
+    threshold: "80",
+  };
+}
+
+export function ReviewTryOutPage() {
+  const [repo, setRepo] = useState("");
+  const [pr, setPr] = useState("");
+  const [drafts, setDrafts] = useState<ClassificationDraft[]>(() => [
+    blankDraft(),
+  ]);
+  // The id of the run being polled; null when none is in flight.
+  const [runId, setRunId] = useState<string | null>(null);
+  const [starting, setStarting] = useState(false);
+  const [result, setResult] = useState<Result | null>(null);
+  const [error, setError] = useState<ProblemDetails | null>(null);
+  // Whole seconds since the run started, so a long wait shows it is alive.
+  const [elapsed, setElapsed] = useState(0);
+
+  const [loadingRules, setLoadingRules] = useState(false);
+  const [rulesError, setRulesError] = useState<ProblemDetails | null>(null);
+  // Where the rows came from, until the next load replaces it.
+  const [loaded, setLoaded] = useState<(RulesAnswer & { repo: string }) | null>(
+    null,
+  );
+  const [confirmLoad, setConfirmLoad] = useState(false);
+
+  const [opening, setOpening] = useState(false);
+  const [opened, setOpened] = useState<
+    (OpenPrAnswer & { repo: string }) | null
+  >(null);
+  const [openError, setOpenError] = useState<ProblemDetails | null>(null);
+
+  // What the two fields offer to complete to. The fields stay free text, so a
+  // repo or a PR the lists lack, a merged one say, can still be typed.
+  const [repoChoices, setRepoChoices] = useState<string[]>([]);
+  const [repoChoicesError, setRepoChoicesError] =
+    useState<ProblemDetails | null>(null);
+  const [pulls, setPulls] = useState<OpenPull[]>([]);
+  const [pullsError, setPullsError] = useState<ProblemDetails | null>(null);
+
+  const running = starting || runId !== null;
+
+  // The repos with an open PR, read once when the page opens.
+  useEffect(() => {
+    let cancelled = false;
+    void (async () => {
+      try {
+        const res = await fetch(PULL_REPOS);
+        if (cancelled) return;
+        if (!res.ok) {
+          setRepoChoicesError(await readApiError(res));
+          return;
+        }
+        setRepoChoices((await res.json()) as string[]);
+      } catch (e: unknown) {
+        if (!cancelled) setRepoChoicesError({ detail: String(e) });
+      }
+    })();
+    return () => {
+      cancelled = true;
+    };
+  }, []);
+
+  // The open PRs of the repo once it is typed or picked and has stopped
+  // changing. A repo that is not valid yet has no list.
+  useEffect(() => {
+    setPulls([]);
+    setPullsError(null);
+    const target = repo.trim();
+    if (repoProblem(target) !== null) return;
+    let cancelled = false;
+    const timer = setTimeout(() => {
+      void (async () => {
+        try {
+          const res = await fetch(
+            `${PULLS}?repo=${encodeURIComponent(target)}`,
+          );
+          if (cancelled) return;
+          if (!res.ok) {
+            setPullsError(await readApiError(res));
+            return;
+          }
+          setPulls((await res.json()) as OpenPull[]);
+        } catch (e: unknown) {
+          if (!cancelled) setPullsError({ detail: String(e) });
+        }
+      })();
+    }, PICK_DELAY_MS);
+    return () => {
+      cancelled = true;
+      clearTimeout(timer);
+    };
+  }, [repo]);
+
+  useEffect(() => {
+    if (!running) return;
+    setElapsed(0);
+    const t = setInterval(() => setElapsed((s) => s + 1), 1000);
+    return () => clearInterval(t);
+  }, [running]);
+
+  // Polls the run until it leaves `running`. Cancelled by the next run, or by
+  // leaving the page.
+  useEffect(() => {
+    if (runId === null) return;
+    let cancelled = false;
+    let timer: ReturnType<typeof setTimeout> | undefined;
+    const stop = (problem: ProblemDetails) => {
+      setError(problem);
+      setRunId(null);
+    };
+    async function poll() {
+      try {
+        const res = await fetch(
+          `${TRY_OUT}/${encodeURIComponent(runId ?? "")}`,
+        );
+        if (cancelled) return;
+        if (res.status === 404) {
+          stop({
+            detail:
+              "the result is gone, because it expired or the controller restarted: run it again",
+          });
+          return;
+        }
+        if (!res.ok) {
+          const problem = await readApiError(res);
+          if (!cancelled) stop(problem);
+          return;
+        }
+        const body = (await res.json()) as Result;
+        if (cancelled) return;
+        setResult(body);
+        if (body.state === "running")
+          timer = setTimeout(() => void poll(), POLL_MS);
+        else setRunId(null);
+      } catch (e: unknown) {
+        if (!cancelled) stop({ detail: String(e) });
+      }
+    }
+    void poll();
+    return () => {
+      cancelled = true;
+      if (timer !== undefined) clearTimeout(timer);
+    };
+  }, [runId]);
+
+  const openPrProblem = rowsProblem(repo, drafts);
+  const problem = runProblem(repo, pr, drafts);
+
+  async function run(e: Event) {
+    e.preventDefault();
+    const number = prNumber(pr);
+    if (problem !== null || number === null) return;
+    setStarting(true);
+    setError(null);
+    setResult(null);
+    try {
+      const res = await fetch(TRY_OUT, {
+        method: "POST",
+        headers: { "content-type": "application/json" },
+        body: JSON.stringify({
+          repo: repo.trim(),
+          pr: number,
+          classifications: drafts.map(toBody),
+        }),
+      });
+      if (!res.ok) {
+        setError(await readApiError(res));
+        return;
+      }
+      setRunId(((await res.json()) as { id: string }).id);
+    } catch (err: unknown) {
+      setError({ detail: String(err) });
+    } finally {
+      setStarting(false);
+    }
+  }
+
+  // Reads the repo's rules from its default branch into the rows. A failed
+  // load leaves the rows as they were.
+  async function loadRules() {
+    const target = repo.trim();
+    setConfirmLoad(false);
+    setLoadingRules(true);
+    setRulesError(null);
+    try {
+      const res = await fetch(`${RULES}?repo=${encodeURIComponent(target)}`);
+      if (!res.ok) {
+        setRulesError(await readApiError(res));
+        return;
+      }
+      const answer = (await res.json()) as RulesAnswer;
+      setDrafts(answer.classifications.map((c) => draftFromRule(c, nextKey++)));
+      setLoaded({ ...answer, repo: target });
+    } catch (err: unknown) {
+      setRulesError({ detail: String(err) });
+    } finally {
+      setLoadingRules(false);
+    }
+  }
+
+  // Pushes the rows on a new branch of the repo and keeps the link to open
+  // the PR from it.
+  async function openPr() {
+    const target = repo.trim();
+    setOpening(true);
+    setOpened(null);
+    setOpenError(null);
+    try {
+      const res = await fetch(OPEN_PR, {
+        method: "POST",
+        headers: { "content-type": "application/json" },
+        body: JSON.stringify({
+          repo: target,
+          classifications: drafts.map(toBody),
+        }),
+      });
+      if (!res.ok) {
+        setOpenError(await readApiError(res));
+        return;
+      }
+      setOpened({ ...((await res.json()) as OpenPrAnswer), repo: target });
+    } catch (err: unknown) {
+      setOpenError({ detail: String(err) });
+    } finally {
+      setOpening(false);
+    }
+  }
+
+  // Typed rows are the operator's work, so replacing them is asked first.
+  function askLoadRules() {
+    if (hasTypedText(drafts)) setConfirmLoad(true);
+    else void loadRules();
+  }
+
+  function changeDraft(next: ClassificationDraft) {
+    setDrafts((ds) => ds.map((d) => (d.key === next.key ? next : d)));
+  }
+
+  return (
+    <div class="ui-tryout-page">
+      <Panel title="review try-out" icon="🧪">
+        <p>
+          Run a review classification against a real PR and see how the model
+          answers it. A run posts nothing to the forge, and nothing is saved:
+          change the classification and run it again until it answers the way
+          you want. Then open a PR to add it to the repo's rules.
+        </p>
+        <form class="ui-tryout-form" onSubmit={(e) => void run(e)}>
+          <div class="ui-tryout-target">
+            <TextField
+              id="tryout-repo"
+              label="repo"
+              value={repo}
+              placeholder="owner/name"
+              onInput={setRepo}
+              suggestions={repoChoices.map((value) => ({ value }))}
+            />
+            <TextField
+              id="tryout-pr"
+              label="PR number"
+              value={pr}
+              pattern="[0-9]+"
+              onInput={setPr}
+              suggestions={pulls.map((p) => ({
+                value: String(p.number),
+                label: p.title,
+              }))}
+            />
+            <Button
+              onClick={askLoadRules}
+              disabled={loadingRules || running || repoProblem(repo) !== null}
+            >
+              {loadingRules ? "loading…" : "load the repo's rules"}
+            </Button>
+          </div>
+          {repoChoicesError ? (
+            <ApiErrorPanel
+              context="the repos with an open PR could not be listed"
+              problem={repoChoicesError}
+            />
+          ) : null}
+          {pullsError ? (
+            <ApiErrorPanel
+              context="the repo's open PRs could not be listed"
+              problem={pullsError}
+            />
+          ) : null}
+          {rulesError ? (
+            <ApiErrorPanel
+              context="the repo's rules could not be loaded"
+              problem={rulesError}
+            />
+          ) : null}
+          {loaded ? (
+            <p class="ui-tryout-loaded">
+              Loaded {loaded.classifications.length} from{" "}
+              <code>{loaded.path}</code> on <code>{loaded.branch}</code> of{" "}
+              {loaded.repo}, read at commit{" "}
+              <code>{loaded.commit.slice(0, 12)}</code>. Edit them below, then
+              run.
+            </p>
+          ) : null}
+          {drafts.map((d) => (
+            <ClassificationEditor
+              key={d.key}
+              draft={d}
+              onChange={changeDraft}
+              onRemove={
+                drafts.length > 1
+                  ? () => setDrafts((ds) => ds.filter((x) => x.key !== d.key))
+                  : undefined
+              }
+            />
+          ))}
+          <div class="ui-tryout-actions">
+            <Button
+              onClick={() => setDrafts((ds) => [...ds, blankDraft()])}
+              disabled={drafts.length >= MAX_CLASSIFICATIONS}
+            >
+              add a classification
+            </Button>
+            <Button
+              variant="primary"
+              type="submit"
+              disabled={running || problem !== null}
+            >
+              {running ? "running…" : "run"}
+            </Button>
+            <Button
+              onClick={() => void openPr()}
+              disabled={opening || openPrProblem !== null}
+            >
+              {opening ? "pushing…" : "open a PR"}
+            </Button>
+          </div>
+          {problem !== null ? <p class="ui-tryout-hint">{problem}</p> : null}
+          {openError ? (
+            <ApiErrorPanel
+              context="the branch for the PR could not be pushed"
+              problem={openError}
+            />
+          ) : null}
+          {opened ? (
+            <p class="ui-tryout-loaded">
+              Pushed <code>{opened.branch}</code> to {opened.repo}, changing{" "}
+              <code>{opened.path}</code> with you as the author.{" "}
+              <a href={opened.compare_url} target="_blank" rel="noopener">
+                Open the PR on the forge
+              </a>
+              .
+            </p>
+          ) : null}
+        </form>
+      </Panel>
+      <ConfirmDialog
+        open={confirmLoad}
+        label="replace the classifications"
+        confirmLabel="replace them"
+        onCancel={() => setConfirmLoad(false)}
+        onConfirm={() => void loadRules()}
+      >
+        <p>
+          Loading the rules of {repo.trim()} replaces the classifications you
+          have written below.
+        </p>
+      </ConfirmDialog>
+      {running || result || error ? (
+        <Panel title="result" icon="📋">
+          {running ? (
+            <p class="ui-tryout-running" role="status">
+              Running — a run takes a minute or more, and up to half an hour for
+              a long list ({elapsed}s so far).
+            </p>
+          ) : null}
+          {error ? (
+            <ApiErrorPanel context="the try-out failed" problem={error} />
+          ) : null}
+          {result?.state === "failed" ? (
+            <ApiErrorPanel
+              context="the run failed"
+              problem={{
+                title: "the classifier could not finish",
+                detail: result.error ?? "no reason given",
+              }}
+            />
+          ) : null}
+          {result?.state === "done" ? <TryOutResult result={result} /> : null}
+        </Panel>
+      ) : null}
+      <ClassifierSettings />
+    </div>
+  );
+}

frontend/packages/swarm-ui/src/pages/review/ReviewPage.css

@@ -1,6 +0,0 @@
-/* The folded bars, then the runs: one column, one gap between the parts. */
-.ui-review-page {
-  display: flex;
-  flex-direction: column;
-  gap: 1em;
-}

frontend/packages/swarm-ui/src/pages/review/ReviewPage.tsx

@@ -1,20 +0,0 @@
-// <ReviewPage> — the review classifier on one page (`docs/swarm/ui.md`,
-// "Review"): the way to try a classification on a PR, the classifier's own
-// settings, and the runs it has done. The first two fold, since they are
-// used now and then and a folded header says what is in them; the runs are
-// what the page is for, so they are always in view, under the two bars.
-// Each part keeps its own data and requests: this page only lays them out.
-import { ClassifierSettings } from "../review-settings/ClassifierSettings.js";
-import { ReviewRuns } from "../review-runs/ReviewRuns.js";
-import { ReviewTryOut } from "../review-tryout/ReviewTryOut.js";
-import "./ReviewPage.css";
-
-export function ReviewPage() {
-  return (
-    <div class="ui-review-page">
-      <ReviewTryOut />
-      <ClassifierSettings />
-      <ReviewRuns />
-    </div>
-  );
-}

frontend/packages/swarm-ui/src/shell/Shell.tsx

@@ -54,7 +54,8 @@ const NAV_ITEMS: { href: string; label: string; accent: string }[] = [
   { href: "/jobs", label: "jobs", accent: "var(--pink)" },
   { href: "/schedules", label: "schedules", accent: "var(--cyan)" },
   { href: "/issues", label: "issues", accent: "var(--yellow)" },
-  { href: "/review", label: "review", accent: "var(--amber)" },
+  { href: "/review-runs", label: "review runs", accent: "var(--amber)" },
+  { href: "/review-tryout", label: "review try-out", accent: "var(--red)" },
   { href: "/components", label: "components", accent: "var(--blue)" },
 ];
 

frontend/packages/swarm-ui/src/shell/routes.ts

@@ -9,7 +9,12 @@
 // modifier alongside the base class, not a replacement, so every other
 // route's layout is untouched. `/agents` joined for the same reason —
 // its own table view (plus the list+detail split) hits the same cap.
-const WIDE_BODY_ROUTES = new Set(["/issues", "/agents", "/review"]);
+const WIDE_BODY_ROUTES = new Set([
+  "/issues",
+  "/agents",
+  "/review-runs",
+  "/review-tryout",
+]);
 
 // `/agents/:name/terminal` (and a subagent's, under
 // `/agents/:name/subagents/:subagent/`) is full-bleed: the terminal fills

frontend/packages/swarm-ui/src/ui/disclosure/Disclosure.css

@@ -1,47 +0,0 @@
-/* A shut disclosure is only its header row, so it has no divider to draw and
-   no body to space; the divider shows once it is open. */
-.ui-disclosure-summary {
-  cursor: pointer;
-  list-style: none;
-  border-bottom-color: transparent;
-}
-.ui-disclosure-summary::-webkit-details-marker {
-  display: none;
-}
-.ui-disclosure[open] > .ui-disclosure-summary {
-  border-bottom-color: var(--border);
-}
-.ui-disclosure-summary:hover .ui-panel-title {
-  text-decoration: underline;
-}
-.ui-disclosure-summary:focus-visible {
-  outline: 2px solid var(--blue);
-  outline-offset: -2px;
-  border-radius: 0.5em;
-}
-.ui-disclosure-hint {
-  min-width: 0;
-  overflow: hidden;
-  color: var(--muted);
-  font-size: 0.9em;
-  text-overflow: ellipsis;
-  white-space: nowrap;
-}
-.ui-disclosure-chevron {
-  margin-left: auto;
-  width: 0.55em;
-  height: 0.55em;
-  flex: none;
-  border-right: 2px solid var(--muted);
-  border-bottom: 2px solid var(--muted);
-  transform: rotate(-45deg);
-  transition: transform 0.15s ease;
-}
-.ui-disclosure[open] > .ui-disclosure-summary .ui-disclosure-chevron {
-  transform: rotate(45deg);
-}
-@media (prefers-reduced-motion: reduce) {
-  .ui-disclosure-chevron {
-    transition: none;
-  }
-}

frontend/packages/swarm-ui/src/ui/disclosure/Disclosure.tsx

@@ -1,50 +0,0 @@
-// <Disclosure> — a Panel that folds: the header row is the control, and the
-// body shows only while it is open. For a section a page keeps within reach
-// but that is not the point of the page, such as a form used now and then.
-// It is a native `<details>`, so the keyboard and assistive tech get the
-// open/closed state for free, and it wears the Panel's own classes, so a
-// Disclosure and a Panel on one page read as the same kind of surface.
-//
-// Controlled: `open` and `onToggle`, so a page can open it for the user (a
-// run started, say). The body stays in the DOM while it is shut, so what was
-// typed into it survives a fold. `hint` is a short line beside the title that
-// says what is in there without opening it.
-import type { ComponentChildren } from "preact";
-import "../panel/Panel.css";
-import "./Disclosure.css";
-
-export function Disclosure({
-  title,
-  icon,
-  hint,
-  open,
-  onToggle,
-  children,
-}: {
-  title: string;
-  icon?: string;
-  hint?: ComponentChildren;
-  open: boolean;
-  onToggle: (open: boolean) => void;
-  children: ComponentChildren;
-}) {
-  return (
-    <details
-      class="ui-panel ui-disclosure"
-      open={open}
-      onToggle={(e) => onToggle((e.currentTarget as HTMLDetailsElement).open)}
-    >
-      <summary class="ui-panel-header ui-disclosure-summary">
-        {icon ? (
-          <span class="ui-panel-icon" aria-hidden="true">
-            {icon}
-          </span>
-        ) : null}
-        <h2 class="ui-panel-title">{title}</h2>
-        {hint ? <span class="ui-disclosure-hint">{hint}</span> : null}
-        <span class="ui-disclosure-chevron" aria-hidden="true" />
-      </summary>
-      <div class="ui-panel-body">{children}</div>
-    </details>
-  );
-}

frontend/packages/swarm-ui/test/routes.test.mjs

@@ -33,7 +33,7 @@ test("a terminal path with extra segments is not full-bleed", () => {
 });
 
 test("the wide-body routes are exactly the table pages", () => {
-  for (const path of ["/issues", "/agents", "/review"]) {
+  for (const path of ["/issues", "/agents", "/review-runs", "/review-tryout"]) {
     assert.equal(isWideBodyRoute(path), true, path);
   }
   for (const path of ["/", "/jobs", "/schedules", "/components"]) {

frontend/packages/swarm-ui/test/settings-types.test.mjs

@@ -10,7 +10,6 @@ import {
   maxValue,
   priceValue,
   settingsProblem,
-  settingsSummary,
   toSettings,
 } from "../src/pages/review-settings/SettingsTypes.ts";
 
@@ -118,29 +117,3 @@ test("the reviews limit and a price parse as the form needs", () => {
   assert.equal(priceValue("-1"), null);
   assert.equal(priceValue("1e3"), null);
 });
-
-test("the summary names the runtime, model and how many run at once", () => {
-  assert.equal(settingsSummary(apiKey), "api-key · haiku · 2 at once");
-});
-
-test("the summary names an effort when one is set", () => {
-  assert.equal(
-    settingsSummary({ ...apiKey, effort: "high" }),
-    "api-key · haiku · high effort · 2 at once",
-  );
-});
-
-test("under ACP the summary names the provider's model, not the claude one", () => {
-  assert.equal(
-    settingsSummary({
-      ...apiKey,
-      runtime: "acp",
-      acp: {
-        base_url: "http://x",
-        model: "glm",
-        prices: { input: 1, output: 2 },
-      },
-    }),
-    "acp · glm · 2 at once",
-  );
-});