@iblai/iblai-js 1.26.2 → 1.26.4

package/dist/playwright/index.cjs

@@ -3165,7 +3165,7 @@ const PRIVACY_LABELS = {
  * reach DOM that Radix renders outside the dialog subtree (popovers,
  * select options, etc.).
  */
-function asPage$5(scope) {
+function asPage$6(scope) {
     return 'page' in scope ? scope.page() : scope;
 }
 /**
@@ -3226,7 +3226,7 @@ async function selectPrivacyAction(scope, action) {
     await trigger.click();
     // Radix Select renders options in a portal at the document root, so we
     // always look them up on the Page — never on the dialog Locator.
-    const option = asPage$5(scope).getByRole('option', {
+    const option = asPage$6(scope).getByRole('option', {
         name: PRIVACY_LABELS.actionOptions[action],
     });
     await test$1.expect(option).toBeVisible({ timeout: 5000 });
@@ -3338,7 +3338,7 @@ const CHAT_PRIVACY_LABELS = {
         rowLabel: 'Allow users to control chat privacy',
     },
 };
-function asPage$4(scope) {
+function asPage$5(scope) {
     return 'page' in scope ? scope.page() : scope;
 }
 // ──────────────────────────────────────────────────────────────────────
@@ -3410,7 +3410,7 @@ async function clickChatPrivacyToggle(scope) {
 // ── Mid-session confirm dialog ────────────────────────────────────────
 /** Locator for the AlertDialog content that the mid-session enable opens. */
 function getChatPrivacyConfirmDialog(scope) {
-    return asPage$4(scope).getByTestId('chat-privacy-confirm-dialog');
+    return asPage$5(scope).getByTestId('chat-privacy-confirm-dialog');
 }
 async function expectChatPrivacyConfirmDialogOpen(scope, open) {
     if (open) {
@@ -3429,7 +3429,7 @@ async function expectChatPrivacyConfirmDialogOpen(scope, open) {
  */
 async function confirmEnableChatPrivacyMidSession(scope) {
     await clickChatPrivacyToggle(scope);
-    const page = asPage$4(scope);
+    const page = asPage$5(scope);
     await expectChatPrivacyConfirmDialogOpen(page, true);
     const action = page.getByTestId('chat-privacy-confirm-action');
     await test$1.expect(action).toBeVisible({ timeout: 10000 });
@@ -3440,7 +3440,7 @@ async function confirmEnableChatPrivacyMidSession(scope) {
 }
 /** Cancel the confirm dialog without enabling. */
 async function cancelEnableChatPrivacyMidSession(scope) {
-    const page = asPage$4(scope);
+    const page = asPage$5(scope);
     const cancel = page.getByTestId('chat-privacy-confirm-cancel');
     await test$1.expect(cancel).toBeVisible({ timeout: 10000 });
     await cancel.click();
@@ -3627,7 +3627,7 @@ const VOICE_LABELS = {
  * reach DOM that Radix renders outside the dialog subtree (popovers,
  * select options, etc.).
  */
-function asPage$3(scope) {
+function asPage$4(scope) {
     return 'page' in scope ? scope.page() : scope;
 }
 // ─── Tab navigation ──────────────────────────────────────────────────────
@@ -3709,7 +3709,7 @@ async function openMentorVoicePicker(scope) {
     const openBtn = scope.getByTestId('mentor-voice-trigger-open');
     await test$1.expect(openBtn).toBeVisible({ timeout: 10000 });
     await openBtn.click();
-    await test$1.expect(asPage$3(scope).getByTestId('voice-picker-modal')).toBeVisible({
+    await test$1.expect(asPage$4(scope).getByTestId('voice-picker-modal')).toBeVisible({
         timeout: 10000,
     });
 }
@@ -3721,7 +3721,7 @@ async function openCallConfigVoicePicker(scope) {
     const openBtn = scope.getByTestId('call-config-voice-trigger-open');
     await test$1.expect(openBtn).toBeVisible({ timeout: 10000 });
     await openBtn.click();
-    await test$1.expect(asPage$3(scope).getByTestId('voice-picker-modal')).toBeVisible({
+    await test$1.expect(asPage$4(scope).getByTestId('voice-picker-modal')).toBeVisible({
         timeout: 10000,
     });
 }
@@ -3834,7 +3834,7 @@ async function selectCallMode(scope, mode) {
     const trigger = scope.getByRole('combobox', { name: VOICE_LABELS.callConfigFields.mode });
     await test$1.expect(trigger).toBeVisible({ timeout: 10000 });
     await trigger.click();
-    const option = asPage$3(scope).getByRole('option', { name: VOICE_LABELS.modeOptions[mode] });
+    const option = asPage$4(scope).getByRole('option', { name: VOICE_LABELS.modeOptions[mode] });
     await test$1.expect(option).toBeVisible({ timeout: 5000 });
     await option.click();
     await test$1.expect(trigger).toHaveText(new RegExp(VOICE_LABELS.modeOptions[mode]));
@@ -3850,7 +3850,7 @@ async function selectFromCombobox(scope, triggerName, optionName) {
     const trigger = scope.getByRole('combobox', { name: triggerName });
     await test$1.expect(trigger).toBeVisible({ timeout: 10000 });
     await trigger.click();
-    const option = asPage$3(scope).getByRole('option', { name: optionName });
+    const option = asPage$4(scope).getByRole('option', { name: optionName });
     await test$1.expect(option).toBeVisible({ timeout: 5000 });
     await option.click();
 }
@@ -3947,7 +3947,7 @@ const SCREENSHARE_LABELS = {
     },
     saveButton: 'Save',
 };
-function asPage$2(scope) {
+function asPage$3(scope) {
     return 'page' in scope ? scope.page() : scope;
 }
 /**
@@ -3987,7 +3987,7 @@ async function openScreenSharePromptEditor(scope, field) {
     const btn = scope.getByRole('button', { name: `Edit ${label}` });
     await test$1.expect(btn).toBeVisible({ timeout: 10000 });
     await btn.click();
-    await test$1.expect(asPage$2(scope).getByText(`Edit ${label}`)).toBeVisible({ timeout: 10000 });
+    await test$1.expect(asPage$3(scope).getByText(`Edit ${label}`)).toBeVisible({ timeout: 10000 });
     logger.info(`Opened screen-share ${field} editor`);
 }
 /**
@@ -3997,7 +3997,7 @@ async function openScreenSharePromptEditor(scope, field) {
  */
 async function setScreenSharePrompt(scope, field, text) {
     await openScreenSharePromptEditor(scope, field);
-    const page = asPage$2(scope);
+    const page = asPage$3(scope);
     const editor = page.getByRole('dialog').locator('[contenteditable="true"]').first();
     await test$1.expect(editor).toBeVisible({ timeout: 10000 });
     await editor.click();
@@ -4006,7 +4006,10 @@ async function setScreenSharePrompt(scope, field, text) {
     await editor.pressSequentially(text);
     // The modal's Save button — scope to dialog so we don't grab the
     // outer tab's Save.
-    await page.getByRole('dialog').getByRole('button', { name: /^save$/i }).click();
+    await page
+        .getByRole('dialog')
+        .getByRole('button', { name: /^save$/i })
+        .click();
     await test$1.expect(page.getByRole('dialog')).toBeHidden({ timeout: 10000 });
     logger.info(`Set screen-share ${field}`);
 }
@@ -4325,7 +4328,7 @@ const TASKS_LABELS = {
  * reach DOM that Radix renders outside the dialog subtree (popovers,
  * select options, the schedule / delete / log-details dialogs, etc.).
  */
-function asPage$1(scope) {
+function asPage$2(scope) {
     return 'page' in scope ? scope.page() : scope;
 }
 // ── Tab navigation ─────────────────────────────────────────────────────
@@ -4435,7 +4438,7 @@ async function searchTasks(scope, query) {
  * shares an accessible name with the dialog's submit button.
  */
 function getScheduleTaskDialog(scope) {
-    return asPage$1(scope).getByRole('dialog', { name: TASKS_LABELS.scheduleDialog.dialogName });
+    return asPage$2(scope).getByRole('dialog', { name: TASKS_LABELS.scheduleDialog.dialogName });
 }
 /** Open the Schedule Task dialog and wait for it to be interactive. */
 async function openScheduleTaskDialog(scope) {
@@ -4454,7 +4457,7 @@ async function openScheduleTaskDialog(scope) {
  */
 async function scheduleTask(scope, opts) {
     await openScheduleTaskDialog(scope);
-    const page = asPage$1(scope);
+    const page = asPage$2(scope);
     // Scope every query to the dialog: the toolbar "Schedule Task" button shares
     // an accessible name with the dialog's submit button, so an unscoped
     // getByRole('button', { name: 'Schedule Task' }) matches two elements.
@@ -4489,13 +4492,13 @@ async function scheduleTask(scope, opts) {
 }
 /** Asserts the in-dialog past-time error is currently shown. */
 async function expectScheduleStartTimeInPastError(scope) {
-    await test$1.expect(asPage$1(scope).getByText(TASKS_LABELS.scheduleDialog.startTimeInPast)).toBeVisible({
+    await test$1.expect(asPage$2(scope).getByText(TASKS_LABELS.scheduleDialog.startTimeInPast)).toBeVisible({
         timeout: 5000,
     });
 }
 // ── Delete flow ────────────────────────────────────────────────────────
 function getDeleteTaskDialog(scope) {
-    return asPage$1(scope).locator(`[role="dialog"][aria-label="${TASKS_LABELS.deleteDialog.dialogName}"]`);
+    return asPage$2(scope).locator(`[role="dialog"][aria-label="${TASKS_LABELS.deleteDialog.dialogName}"]`);
 }
 /**
  * Click the trash icon on a task row and confirm the delete in the
@@ -4516,7 +4519,7 @@ async function deleteTask(scope, taskName) {
 }
 // ── Logs panel + log details ───────────────────────────────────────────
 function getLogDetailsDialog(scope) {
-    return asPage$1(scope).locator(`[role="dialog"][aria-label="${TASKS_LABELS.logDetails.dialogName}"]`);
+    return asPage$2(scope).locator(`[role="dialog"][aria-label="${TASKS_LABELS.logDetails.dialogName}"]`);
 }
 /**
  * Wait for the logs panel to finish loading and assert that the selected
@@ -4552,6 +4555,801 @@ async function expectLogDetailsStatus(scope, status) {
     });
 }
 
+/**
+ * Evals tab helpers — Playwright bindings for the `AgentEvaluationTab`
+ * component from `@iblai/web-containers` (edit-mentor modal > Evals tab),
+ * including the embedded tenant Benchmarks management dialog.
+ *
+ * All UI strings below mirror `AGENT_EVALUATION_TAB_LABELS` from
+ * `@iblai/web-containers/next` plus the `benchmarks*` i18n namespaces. If a
+ * consumer renames a label via the `labels` prop, override these constants
+ * per spec — don't edit this file.
+ *
+ * Every dialog spawned by this tab carries an explicit `aria-label` on its
+ * `DialogContent` (see `dialogName` entries). Helpers ALWAYS resolve the
+ * dialog once via `[role="dialog"][aria-label="…"]` — the attribute selector,
+ * not the role accessible name, because a visible `DialogTitle` makes Radix's
+ * `aria-labelledby` win the accessible-name computation — and then resolve
+ * every in-dialog control from that scoped locator. This matters here more
+ * than usual:
+ *   - the toolbar "New Evaluation" / "Manage benchmarks" buttons share their
+ *     text with the dialog titles they open;
+ *   - the LLM-judge dialog's title AND submit button both read "Evaluate";
+ *   - dialogs nest three deep (Manage benchmarks → benchmark items → add
+ *     Q&A), and the add-Q&A dialog has Manual / CSV Upload sub-tabs whose
+ *     footers each have their own Cancel button.
+ *
+ * The helpers assume the Edit Mentor dialog is already open. Call
+ * `switchToEvaluationTab(page)` first; from then on, every helper accepts
+ * either the `Page` or a scoping `Locator`.
+ */
+const EVALS_LABELS = {
+    tabName: 'Evals',
+    header: {
+        title: 'Evals',
+        description: 'Run this agent against a benchmark and score the responses.',
+    },
+    benchmarkPicker: {
+        placeholder: 'Select a benchmark',
+        searchPlaceholder: 'Search benchmarks...',
+        empty: 'No benchmarks found',
+        /** `data-testid` on the combobox trigger button (its text is the selected benchmark, so it has no stable accessible name). */
+        triggerTestId: 'benchmark-combobox',
+        /** `data-testid` on the dropdown panel (rendered inline, not in a portal). */
+        dropdownTestId: 'benchmark-combobox-dropdown',
+    },
+    actions: {
+        newExperiment: 'New Evaluation',
+        manageBenchmarks: 'Manage benchmarks',
+        viewResults: 'View results',
+        evaluate: 'New review',
+        exportCsv: 'Export CSV',
+        checkStatus: 'Check status',
+        delete: 'Delete',
+    },
+    checkStatusToast: {
+        pending: 'Evaluation is still running',
+        completed: 'Evaluation completed',
+        failed: 'Evaluation failed',
+    },
+    emptyBenchmarks: 'No benchmarks yet. Use "Manage benchmarks" above to create one.',
+    startDialog: {
+        /** `aria-label` on the dialog's `DialogContent` — scopes in-dialog locators. */
+        dialogName: 'start-evaluation',
+        experimentNameLabel: 'Evaluation name',
+        cancel: 'Cancel',
+        start: 'Start Evaluation',
+    },
+    manageBenchmarksDialog: {
+        /** `aria-label` on the dialog's `DialogContent` — scopes in-dialog locators. */
+        dialogName: 'manage-benchmarks',
+        searchAriaLabel: 'Search benchmarks',
+        newBenchmark: 'New Benchmark',
+        empty: 'No benchmarks found',
+    },
+    createBenchmarkDialog: {
+        /** `aria-label` on the dialog's `DialogContent` — scopes in-dialog locators. */
+        dialogName: 'create-benchmark',
+        nameLabel: 'Name',
+        descriptionLabel: 'Description',
+        cancel: 'Cancel',
+        create: 'Create',
+    },
+    benchmarkItemsDialog: {
+        /** `aria-label` on the dialog's `DialogContent` — scopes in-dialog locators. */
+        dialogName: 'benchmark-items',
+        addItems: 'Add Q&A',
+        deleteItemAriaLabel: 'Delete Q&A',
+        empty: 'No Q&A in this benchmark yet',
+    },
+    addItemsDialog: {
+        /** `aria-label` on the dialog's `DialogContent` — scopes in-dialog locators. */
+        dialogName: 'add-benchmark-items',
+        manualTab: 'Manual',
+        csvTab: 'CSV Upload',
+        questionLabel: 'Question',
+        answerLabel: 'Answer',
+        addRow: 'Add row',
+        removeRowAriaLabel: 'Remove row',
+        add: 'Add',
+        csvFileLabel: 'CSV file',
+        upload: 'Upload',
+        cancel: 'Cancel',
+    },
+    deleteItemDialog: {
+        /** `aria-label` on the dialog's `DialogContent` — scopes in-dialog locators. */
+        dialogName: 'delete-benchmark-item',
+        delete: 'Delete',
+        cancel: 'Cancel',
+    },
+    detailDialog: {
+        /** `aria-label` on the dialog's `DialogContent` — scopes in-dialog locators. */
+        dialogName: 'evaluation-detail',
+        refreshAriaLabel: 'Refresh',
+        newReview: 'New review',
+        waitingHint: 'Waiting for the evaluation to produce traces. This page auto-refreshes.',
+        noScoresHint: 'No scores yet',
+        scoreNameLabel: 'Score name',
+        scoreValueLabel: 'Value',
+        scoreCommentLabel: 'Comment',
+        addScore: 'Add',
+        removeScoreAriaLabel: 'Remove score',
+        reviewsHeading: 'Reviews',
+        reviewDetailsAriaLabel: 'Toggle review details',
+        reviewStatus: {
+            pending: 'Queued',
+            running: 'Running',
+            completed: 'Complete',
+            failed: 'Failed',
+        },
+    },
+    judgeDialog: {
+        /** `aria-label` on the dialog's `DialogContent` — scopes in-dialog locators. */
+        dialogName: 'llm-judge',
+        criteriaLabel: 'Criteria',
+        scoreNameLabel: 'Score name',
+        maxConcurrencyLabel: 'Max concurrency',
+        selectorAriaLabel: 'LLM Model Selector',
+        cancel: 'Cancel',
+        submit: 'Evaluate',
+    },
+    llmPickerDialog: {
+        /** `aria-label` on the dialog's `DialogContent` — scopes in-dialog locators. */
+        dialogName: 'evaluate-llm-picker',
+        searchPlaceholder: 'Search providers...',
+        /** Title of the follow-up model-selection modal (llm-tab's `LLMProviderModal`, no aria-label of its own). */
+        modelDialogTitle: 'LLM Selection',
+    },
+    deleteDialog: {
+        /** `aria-label` on the dialog's `DialogContent` — scopes in-dialog locators. */
+        dialogName: 'delete-evaluation',
+        cancel: 'Cancel',
+        confirm: 'Delete',
+    },
+};
+/**
+ * Resolve a Page from either a Page or a Locator. Used for DOM that Radix
+ * renders in a portal outside the current scope (dialogs, dropdown menus,
+ * toasts) — those queries must run on the page, never on the tab pane or a
+ * parent dialog.
+ */
+function asPage$1(scope) {
+    return 'page' in scope ? scope.page() : scope;
+}
+/** Escape a user-supplied string (run / benchmark names) for use in a RegExp. */
+function escapeRegExp(value) {
+    return value.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+}
+/** Attribute-based dialog locator — immune to title renames and to Radix's `aria-labelledby` name precedence. */
+function dialogByName(scope, dialogName) {
+    return asPage$1(scope).locator(`[role="dialog"][aria-label="${dialogName}"]`);
+}
+// ── Tab navigation ─────────────────────────────────────────────────────
+/**
+ * Check whether the Evals tab is rendered in the Edit Mentor dialog.
+ * Returns false instead of throwing so callers can guard permission-gated
+ * tabs.
+ */
+async function isEvaluationTabVisible(page) {
+    const tab = page.getByRole('tab', { name: EVALS_LABELS.tabName, exact: true });
+    try {
+        await test$1.expect(tab).toBeVisible({ timeout: 5000 });
+        return true;
+    }
+    catch (_a) {
+        return false;
+    }
+}
+/**
+ * Switch to the Evals tab. Assumes the Edit Mentor dialog is open. Waits
+ * for the toolbar "Manage benchmarks" button (unique to this pane) before
+ * returning, so callers can rely on the panel being interactive.
+ */
+async function switchToEvaluationTab(page) {
+    const tab = page.getByRole('tab', { name: EVALS_LABELS.tabName, exact: true });
+    await test$1.expect(tab).toBeVisible({ timeout: 10000 });
+    await tab.click();
+    await test$1.expect(page.getByRole('button', { name: EVALS_LABELS.actions.manageBenchmarks })).toBeVisible({ timeout: 10000 });
+    logger.info('Switched to Evals tab');
+}
+// ── Toolbar: benchmark picker + action buttons ─────────────────────────
+/** Locator for the benchmark combobox trigger. Its text is the selected benchmark name. */
+function getBenchmarkCombobox(scope) {
+    return scope.getByTestId(EVALS_LABELS.benchmarkPicker.triggerTestId);
+}
+/** Locator for the combobox dropdown panel (rendered inline under the trigger, not in a portal). */
+function getBenchmarkComboboxDropdown(scope) {
+    return scope.getByTestId(EVALS_LABELS.benchmarkPicker.dropdownTestId);
+}
+/** Locator for the toolbar "New Evaluation" button (opens the start dialog). */
+function getNewEvaluationButton(scope) {
+    return scope.getByRole('button', { name: EVALS_LABELS.actions.newExperiment, exact: true });
+}
+/** Locator for the toolbar "Manage benchmarks" button (opens the benchmarks dialog). */
+function getManageBenchmarksButton(scope) {
+    return scope.getByRole('button', { name: EVALS_LABELS.actions.manageBenchmarks, exact: true });
+}
+/** Assert the currently selected benchmark shown on the combobox trigger. */
+async function expectSelectedBenchmark(scope, benchmarkName) {
+    await test$1.expect(getBenchmarkCombobox(scope)).toContainText(benchmarkName, { timeout: 10000 });
+}
+/**
+ * Open the benchmark combobox, optionally filter via its search input, and
+ * pick a benchmark by exact name. Waits for the dropdown to close and the
+ * trigger to reflect the selection.
+ */
+async function selectBenchmark(scope, benchmarkName, opts = {}) {
+    const trigger = getBenchmarkCombobox(scope);
+    await test$1.expect(trigger).toBeEnabled({ timeout: 10000 });
+    await trigger.click();
+    const dropdown = getBenchmarkComboboxDropdown(scope);
+    await test$1.expect(dropdown).toBeVisible({ timeout: 5000 });
+    if (opts.search) {
+        const search = dropdown.getByPlaceholder(EVALS_LABELS.benchmarkPicker.searchPlaceholder);
+        await search.fill(benchmarkName);
+        await test$1.expect(search).toHaveValue(benchmarkName);
+    }
+    await dropdown
+        .getByRole('button', { name: new RegExp(`^${escapeRegExp(benchmarkName)}$`) })
+        .click();
+    await test$1.expect(dropdown).toBeHidden({ timeout: 5000 });
+    await expectSelectedBenchmark(scope, benchmarkName);
+    logger.info(`Selected benchmark: ${benchmarkName}`);
+}
+/** Assert the "no benchmarks yet" empty notice is shown in the tab body. */
+async function expectNoBenchmarksNotice(scope) {
+    await test$1.expect(scope.getByText(EVALS_LABELS.emptyBenchmarks)).toBeVisible({ timeout: 10000 });
+}
+// ── Runs table ─────────────────────────────────────────────────────────
+/**
+ * Locator for an evaluation run's table row, matched by an exact cell so a
+ * run whose name is a substring of another ("smoke" vs "smoke-2") can't
+ * cross-match.
+ */
+function getRunRow(scope, runName) {
+    return scope
+        .getByRole('row')
+        .filter({ has: asPage$1(scope).getByRole('cell', { name: runName, exact: true }) });
+}
+async function expectRunInTable(scope, runName) {
+    await test$1.expect(getRunRow(scope, runName)).toBeVisible({ timeout: 15000 });
+}
+async function expectRunNotInTable(scope, runName) {
+    await test$1.expect(getRunRow(scope, runName)).toBeHidden({ timeout: 15000 });
+}
+/** Assert the status badge on a run's row (badge text is upper-cased by the UI). */
+async function expectRunStatus(scope, runName, status) {
+    await test$1.expect(getRunRow(scope, runName).getByText(status, { exact: true })).toBeVisible({
+        timeout: 15000,
+    });
+}
+/** Assert the empty-state row shown when the selected benchmark has no runs for this agent. */
+async function expectRunsTableEmpty(scope, benchmarkName) {
+    await test$1.expect(scope.getByText(`No evaluations yet for this agent on benchmark "${benchmarkName}".`)).toBeVisible({ timeout: 15000 });
+}
+/**
+ * Open the three-dots actions menu on a run's row. The trigger carries
+ * `aria-label="Actions for <runName>"`, so it can't collide with other rows.
+ * Returns the menu locator (Radix portals it outside the table — page scope).
+ */
+async function openRunActionsMenu(scope, runName) {
+    await getRunRow(scope, runName)
+        .getByRole('button', { name: `Actions for ${runName}`, exact: true })
+        .click();
+    const menu = asPage$1(scope).getByRole('menu');
+    await test$1.expect(menu).toBeVisible({ timeout: 5000 });
+    return menu;
+}
+/**
+ * Row action: "View results" — opens the evaluation detail dialog.
+ * Only enabled once the run is COMPLETED. Returns the scoped detail dialog.
+ */
+async function openRunResults(scope, runName) {
+    const menu = await openRunActionsMenu(scope, runName);
+    const item = menu.getByRole('menuitem', { name: EVALS_LABELS.actions.viewResults });
+    await test$1.expect(item).toBeEnabled({ timeout: 5000 });
+    await item.click();
+    const dialog = getEvaluationDetailDialog(scope);
+    await test$1.expect(dialog).toBeVisible({ timeout: 10000 });
+    logger.info(`Opened results for run: ${runName}`);
+    return dialog;
+}
+/**
+ * Row action: "New review" — opens the LLM-judge dialog for the run.
+ * Only enabled once the run is COMPLETED. Returns the scoped judge dialog.
+ */
+async function openNewReviewForRun(scope, runName) {
+    const menu = await openRunActionsMenu(scope, runName);
+    const item = menu.getByRole('menuitem', { name: EVALS_LABELS.actions.evaluate });
+    await test$1.expect(item).toBeEnabled({ timeout: 5000 });
+    await item.click();
+    const dialog = getLlmJudgeDialog(scope);
+    await test$1.expect(dialog).toBeVisible({ timeout: 10000 });
+    return dialog;
+}
+/**
+ * Row action: "Check status" — triggers a fresh status fetch and waits for
+ * the outcome toast. Returns which toast appeared. The menu stays open by
+ * design (the item prevents default), so we close it with Escape.
+ */
+async function checkRunStatus(scope, runName) {
+    const page = asPage$1(scope);
+    const menu = await openRunActionsMenu(scope, runName);
+    await menu.getByRole('menuitem', { name: EVALS_LABELS.actions.checkStatus }).click();
+    const toasts = EVALS_LABELS.checkStatusToast;
+    const toast = page
+        .getByText(new RegExp(`${toasts.pending}|${toasts.completed}|${toasts.failed}`))
+        .first();
+    await test$1.expect(toast).toBeVisible({ timeout: 15000 });
+    const text = (await toast.innerText()).trim();
+    await page.keyboard.press('Escape');
+    await test$1.expect(menu).toBeHidden({ timeout: 5000 });
+    if (text.includes(toasts.completed))
+        return 'completed';
+    if (text.includes(toasts.failed))
+        return 'failed';
+    return 'pending';
+}
+/**
+ * Row action: "Export CSV" — waits for the browser download and returns its
+ * suggested filename (`<benchmark>_<run>_results.csv`).
+ */
+async function exportRunCsv(scope, runName) {
+    const page = asPage$1(scope);
+    const menu = await openRunActionsMenu(scope, runName);
+    const item = menu.getByRole('menuitem', { name: EVALS_LABELS.actions.exportCsv });
+    await test$1.expect(item).toBeEnabled({ timeout: 5000 });
+    const downloadPromise = page.waitForEvent('download', { timeout: 30000 });
+    await item.click();
+    const download = await downloadPromise;
+    logger.info(`Exported run CSV: ${download.suggestedFilename()}`);
+    return download.suggestedFilename();
+}
+// ── Start evaluation dialog ────────────────────────────────────────────
+/** Locator for the "New Evaluation" dialog. Scoped by aria-label — its title text collides with the toolbar button that opens it. */
+function getStartEvaluationDialog(scope) {
+    return dialogByName(scope, EVALS_LABELS.startDialog.dialogName);
+}
+/** Open the start dialog via the toolbar button and wait for it to be interactive. */
+async function openStartEvaluationDialog(scope) {
+    await getNewEvaluationButton(scope).click();
+    const dialog = getStartEvaluationDialog(scope);
+    await test$1.expect(dialog.getByLabel(EVALS_LABELS.startDialog.experimentNameLabel)).toBeVisible({
+        timeout: 10000,
+    });
+    return dialog;
+}
+/**
+ * End-to-end: open the "New Evaluation" dialog, optionally name the run,
+ * submit, and wait for the dialog to close. When `name` is omitted the
+ * backend auto-generates one.
+ */
+async function startEvaluation(scope, opts = {}) {
+    const dialog = await openStartEvaluationDialog(scope);
+    if (opts.name !== undefined) {
+        await dialog.getByLabel(EVALS_LABELS.startDialog.experimentNameLabel).fill(opts.name);
+    }
+    // Scoped to the dialog: "Start Evaluation" is unique today, but the toolbar
+    // "New Evaluation" trigger sits right behind the overlay — never query the page.
+    const submit = dialog.getByRole('button', { name: EVALS_LABELS.startDialog.start });
+    await test$1.expect(submit).toBeEnabled({ timeout: 5000 });
+    await submit.click();
+    await test$1.expect(dialog).toBeHidden({ timeout: 15000 });
+    logger.info(`Started evaluation${opts.name ? `: ${opts.name}` : ''}`);
+}
+/** Cancel out of the start dialog and wait for it to close. */
+async function cancelStartEvaluation(scope) {
+    const dialog = getStartEvaluationDialog(scope);
+    await dialog.getByRole('button', { name: EVALS_LABELS.startDialog.cancel }).click();
+    await test$1.expect(dialog).toBeHidden({ timeout: 5000 });
+}
+// ── Delete evaluation dialog ───────────────────────────────────────────
+/** Locator for the "Delete Evaluation" confirm dialog. */
+function getDeleteEvaluationDialog(scope) {
+    return dialogByName(scope, EVALS_LABELS.deleteDialog.dialogName);
+}
+/**
+ * Delete a run from its row actions menu and confirm in the follow-up
+ * dialog. The confirm button ("Delete") shares its name with the menu item
+ * that opened the dialog, so the confirm click MUST stay dialog-scoped.
+ */
+async function deleteEvaluation(scope, runName) {
+    const menu = await openRunActionsMenu(scope, runName);
+    await menu.getByRole('menuitem', { name: EVALS_LABELS.actions.delete }).click();
+    const dialog = getDeleteEvaluationDialog(scope);
+    const confirm = dialog.getByRole('button', { name: EVALS_LABELS.deleteDialog.confirm });
+    await test$1.expect(confirm).toBeVisible({ timeout: 5000 });
+    await confirm.click();
+    await test$1.expect(dialog).toBeHidden({ timeout: 15000 });
+    logger.info(`Deleted evaluation: ${runName}`);
+}
+/** Open the delete confirm from the row menu, then cancel out of it. */
+async function cancelDeleteEvaluation(scope, runName) {
+    const menu = await openRunActionsMenu(scope, runName);
+    await menu.getByRole('menuitem', { name: EVALS_LABELS.actions.delete }).click();
+    const dialog = getDeleteEvaluationDialog(scope);
+    await dialog.getByRole('button', { name: EVALS_LABELS.deleteDialog.cancel }).click();
+    await test$1.expect(dialog).toBeHidden({ timeout: 5000 });
+}
+// ── LLM-judge ("Evaluate") dialog ──────────────────────────────────────
+/**
+ * Locator for the LLM-judge dialog. Scoped by aria-label — its title AND its
+ * submit button both read "Evaluate", and it can be stacked on top of the
+ * evaluation-detail dialog.
+ */
+function getLlmJudgeDialog(scope) {
+    return dialogByName(scope, EVALS_LABELS.judgeDialog.dialogName);
+}
+/** Locator for the judge dialog's LLM provider-picker dialog (stacked above it). */
+function getLlmPickerDialog(scope) {
+    return dialogByName(scope, EVALS_LABELS.llmPickerDialog.dialogName);
+}
+/**
+ * Inside the judge dialog: pick the judging LLM. Opens the provider picker,
+ * optionally filters, clicks the provider card, then picks the model in the
+ * follow-up "LLM Selection" modal (llm-tab's `LLMProviderModal`; matched by
+ * title text since it has no aria-label). Both pickers close on selection.
+ */
+async function selectJudgeLlm(scope, providerName, modelName) {
+    const page = asPage$1(scope);
+    const judgeDialog = getLlmJudgeDialog(scope);
+    await judgeDialog
+        .getByRole('button', { name: EVALS_LABELS.judgeDialog.selectorAriaLabel })
+        .click();
+    const picker = getLlmPickerDialog(scope);
+    await test$1.expect(picker).toBeVisible({ timeout: 10000 });
+    await picker.getByPlaceholder(EVALS_LABELS.llmPickerDialog.searchPlaceholder).fill(providerName);
+    await picker.getByRole('button', { name: new RegExp(escapeRegExp(providerName), 'i') }).click();
+    // Model-selection modal is a separate stacked dialog outside the picker.
+    const modelDialog = page
+        .getByRole('dialog')
+        .filter({ hasText: EVALS_LABELS.llmPickerDialog.modelDialogTitle });
+    await test$1.expect(modelDialog).toBeVisible({ timeout: 10000 });
+    await modelDialog.getByRole('button', { name: new RegExp(escapeRegExp(modelName), 'i') }).click();
+    await test$1.expect(picker).toBeHidden({ timeout: 10000 });
+    logger.info(`Selected judge LLM ${providerName}/${modelName}`);
+}
+/**
+ * Fill and submit the LLM-judge dialog (assumes it is already open — see
+ * `openNewReviewForRun` / `openNewReviewFromDetail`). Waits for the dialog
+ * to close on success.
+ */
+async function submitLlmJudge(scope, opts) {
+    const dialog = getLlmJudgeDialog(scope);
+    await test$1.expect(dialog).toBeVisible({ timeout: 10000 });
+    await dialog.getByLabel(EVALS_LABELS.judgeDialog.criteriaLabel).fill(opts.criteria);
+    await dialog.getByLabel(EVALS_LABELS.judgeDialog.scoreNameLabel).fill(opts.scoreName);
+    if (opts.maxConcurrency !== undefined) {
+        await dialog
+            .getByLabel(EVALS_LABELS.judgeDialog.maxConcurrencyLabel)
+            .fill(String(opts.maxConcurrency));
+    }
+    await selectJudgeLlm(scope, opts.provider, opts.model);
+    // Dialog-scoped on purpose: the dialog title is also "Evaluate".
+    const submit = dialog.getByRole('button', { name: EVALS_LABELS.judgeDialog.submit, exact: true });
+    await test$1.expect(submit).toBeEnabled({ timeout: 5000 });
+    await submit.click();
+    await test$1.expect(dialog).toBeHidden({ timeout: 15000 });
+    logger.info(`Started LLM review "${opts.scoreName}"`);
+}
+/** Cancel out of the judge dialog and wait for it to close. */
+async function cancelLlmJudge(scope) {
+    const dialog = getLlmJudgeDialog(scope);
+    await dialog.getByRole('button', { name: EVALS_LABELS.judgeDialog.cancel }).click();
+    await test$1.expect(dialog).toBeHidden({ timeout: 5000 });
+}
+// ── Evaluation detail dialog ───────────────────────────────────────────
+/** Locator for the evaluation-detail dialog. Its title is the run name (dynamic), hence the aria-label hook. */
+function getEvaluationDetailDialog(scope) {
+    return dialogByName(scope, EVALS_LABELS.detailDialog.dialogName);
+}
+/** Click the detail dialog's refresh icon button (aria-label "Refresh"). */
+async function refreshEvaluationDetail(scope) {
+    const dialog = getEvaluationDetailDialog(scope);
+    const refresh = dialog.getByRole('button', {
+        name: EVALS_LABELS.detailDialog.refreshAriaLabel,
+        exact: true,
+    });
+    await refresh.click();
+    await test$1.expect(refresh).toBeEnabled({ timeout: 15000 });
+}
+/** Open the LLM-judge dialog from the detail dialog's "New review" CTA. Returns the judge dialog. */
+async function openNewReviewFromDetail(scope) {
+    const dialog = getEvaluationDetailDialog(scope);
+    const cta = dialog.getByRole('button', {
+        name: EVALS_LABELS.detailDialog.newReview,
+        exact: true,
+    });
+    await test$1.expect(cta).toBeEnabled({ timeout: 10000 });
+    await cta.click();
+    const judgeDialog = getLlmJudgeDialog(scope);
+    await test$1.expect(judgeDialog).toBeVisible({ timeout: 10000 });
+    return judgeDialog;
+}
+/** Assert the detail dialog is still waiting for traces (auto-refreshing hint). */
+async function expectDetailWaitingForTraces(scope) {
+    await test$1.expect(getEvaluationDetailDialog(scope).getByText(EVALS_LABELS.detailDialog.waitingHint)).toBeVisible({ timeout: 10000 });
+}
+/**
+ * Locator for a trace's expand/collapse row inside the detail dialog,
+ * matched by (part of) its question text.
+ */
+function getTraceRow(scope, questionText) {
+    return getEvaluationDetailDialog(scope)
+        .getByRole('button')
+        .filter({ hasText: questionText })
+        .first();
+}
+/** Expand a trace by its question text so its outputs and scores are visible. */
+async function expandTrace(scope, questionText) {
+    await getTraceRow(scope, questionText).click();
+    // The manual-score form only exists on the expanded trace.
+    await test$1.expect(getEvaluationDetailDialog(scope).getByLabel(EVALS_LABELS.detailDialog.scoreNameLabel)).toBeVisible({ timeout: 10000 });
+}
+/**
+ * Add a manual score to the currently expanded trace. `value` must be
+ * between 0 and 1 (the UI rejects anything else with an error toast).
+ */
+async function addManualScore(scope, opts) {
+    const dialog = getEvaluationDetailDialog(scope);
+    await dialog.getByLabel(EVALS_LABELS.detailDialog.scoreNameLabel).fill(opts.name);
+    await dialog.getByLabel(EVALS_LABELS.detailDialog.scoreValueLabel).fill(String(opts.value));
+    if (opts.comment !== undefined) {
+        await dialog.getByLabel(EVALS_LABELS.detailDialog.scoreCommentLabel).fill(opts.comment);
+    }
+    await dialog
+        .getByRole('button', { name: EVALS_LABELS.detailDialog.addScore, exact: true })
+        .click();
+    await expectTraceScore(scope, opts.name);
+    logger.info(`Added manual score: ${opts.name}=${opts.value}`);
+}
+/** Assert a score chip with the given name is visible on the expanded trace. */
+async function expectTraceScore(scope, scoreName) {
+    await test$1.expect(getEvaluationDetailDialog(scope).getByText(scoreName, { exact: true }).first()).toBeVisible({ timeout: 15000 });
+}
+/**
+ * Remove a score chip from the expanded trace. The chip's trash button
+ * carries `aria-label="Remove score"`; we scope through the chip matched by
+ * score name so multiple chips can't cross-match.
+ */
+async function removeTraceScore(scope, scoreName) {
+    const dialog = getEvaluationDetailDialog(scope);
+    const chip = dialog
+        .locator('div')
+        .filter({ has: asPage$1(scope).getByText(scoreName, { exact: true }) })
+        .filter({
+        has: asPage$1(scope).getByRole('button', {
+            name: EVALS_LABELS.detailDialog.removeScoreAriaLabel,
+        }),
+    })
+        .last();
+    await chip
+        .getByRole('button', { name: EVALS_LABELS.detailDialog.removeScoreAriaLabel })
+        .first()
+        .click();
+}
+/**
+ * Locator for a review row's expand toggle in the detail dialog's Reviews
+ * list. All toggles share the "Toggle review details" aria-label, so the
+ * row is disambiguated by the score name it displays.
+ */
+function getReviewRow(scope, scoreName) {
+    return getEvaluationDetailDialog(scope)
+        .getByRole('button', { name: EVALS_LABELS.detailDialog.reviewDetailsAriaLabel })
+        .filter({ hasText: scoreName })
+        .first();
+}
+/** Assert a review with the given score name appears in the Reviews list. */
+async function expectReviewInList(scope, scoreName) {
+    await test$1.expect(getReviewRow(scope, scoreName)).toBeVisible({ timeout: 15000 });
+}
+/** Assert a review row's status pill (Queued / Running / Complete / Failed). */
+async function expectReviewStatus(scope, scoreName, status) {
+    await test$1.expect(getReviewRow(scope, scoreName).getByText(EVALS_LABELS.detailDialog.reviewStatus[status], {
+        exact: true,
+    })).toBeVisible({ timeout: 15000 });
+}
+/** Expand a review row to show its criteria and produced scores. */
+async function expandReview(scope, scoreName) {
+    const row = getReviewRow(scope, scoreName);
+    await row.click();
+    await test$1.expect(row).toHaveAttribute('aria-expanded', 'true', { timeout: 5000 });
+}
+/** Close the detail dialog via Escape and wait for it to be gone. */
+async function closeEvaluationDetailDialog(scope) {
+    const dialog = getEvaluationDetailDialog(scope);
+    await asPage$1(scope).keyboard.press('Escape');
+    await test$1.expect(dialog).toBeHidden({ timeout: 10000 });
+}
+// ── Manage benchmarks dialog (embedded tenant Benchmarks) ──────────────
+/** Locator for the "Manage benchmarks" dialog. Scoped by aria-label — its title collides with the toolbar button that opens it. */
+function getManageBenchmarksDialog(scope) {
+    return dialogByName(scope, EVALS_LABELS.manageBenchmarksDialog.dialogName);
+}
+/** Open the Manage benchmarks dialog from the toolbar and wait for its table UI. Returns the scoped dialog. */
+async function openManageBenchmarksDialog(scope) {
+    await getManageBenchmarksButton(scope).click();
+    const dialog = getManageBenchmarksDialog(scope);
+    await test$1.expect(dialog.getByRole('button', { name: EVALS_LABELS.manageBenchmarksDialog.newBenchmark })).toBeVisible({ timeout: 10000 });
+    return dialog;
+}
+/** Type into the benchmarks search input (aria-label "Search benchmarks") inside the manage dialog. */
+async function searchBenchmarks(scope, query) {
+    const input = getManageBenchmarksDialog(scope).getByLabel(EVALS_LABELS.manageBenchmarksDialog.searchAriaLabel);
+    await input.fill(query);
+    await test$1.expect(input).toHaveValue(query);
+}
+/** Assert a benchmark appears in the manage dialog's table. */
+async function expectBenchmarkListed(scope, benchmarkName) {
+    await test$1.expect(getManageBenchmarksDialog(scope).getByText(benchmarkName, { exact: true }).first()).toBeVisible({ timeout: 15000 });
+}
+/** Locator for the "New Benchmark" dialog. Scoped by aria-label — its title collides with the button that opens it. */
+function getCreateBenchmarkDialog(scope) {
+    return dialogByName(scope, EVALS_LABELS.createBenchmarkDialog.dialogName);
+}
+/**
+ * Create a benchmark from inside the manage dialog: open "New Benchmark",
+ * fill the form, submit, and wait for both the dialog to close and the new
+ * row to appear.
+ */
+async function createBenchmark(scope, opts) {
+    const manageDialog = getManageBenchmarksDialog(scope);
+    await manageDialog
+        .getByRole('button', { name: EVALS_LABELS.manageBenchmarksDialog.newBenchmark })
+        .click();
+    const dialog = getCreateBenchmarkDialog(scope);
+    await dialog.getByLabel(EVALS_LABELS.createBenchmarkDialog.nameLabel).fill(opts.name);
+    if (opts.description !== undefined) {
+        await dialog
+            .getByLabel(EVALS_LABELS.createBenchmarkDialog.descriptionLabel)
+            .fill(opts.description);
+    }
+    const submit = dialog.getByRole('button', {
+        name: EVALS_LABELS.createBenchmarkDialog.create,
+        exact: true,
+    });
+    await test$1.expect(submit).toBeEnabled({ timeout: 5000 });
+    await submit.click();
+    await test$1.expect(dialog).toBeHidden({ timeout: 15000 });
+    await expectBenchmarkListed(scope, opts.name);
+    logger.info(`Created benchmark: ${opts.name}`);
+}
+/** Locator for a benchmark's items dialog (Q&A list). Its title is the benchmark name (dynamic), hence the aria-label hook. */
+function getBenchmarkItemsDialog(scope) {
+    return dialogByName(scope, EVALS_LABELS.benchmarkItemsDialog.dialogName);
+}
+/**
+ * Open a benchmark's Q&A items dialog from its row's eye button
+ * (`aria-label="View items in <name>"`). Returns the scoped items dialog.
+ */
+async function openBenchmarkItems(scope, benchmarkName) {
+    await getManageBenchmarksDialog(scope)
+        .getByRole('button', { name: `View items in ${benchmarkName}`, exact: true })
+        .click();
+    const dialog = getBenchmarkItemsDialog(scope);
+    await test$1.expect(dialog.getByRole('button', { name: EVALS_LABELS.benchmarkItemsDialog.addItems })).toBeVisible({ timeout: 10000 });
+    return dialog;
+}
+/** Assert a Q&A item (matched by its question text) is listed in the items dialog. */
+async function expectQaItemListed(scope, questionText) {
+    await test$1.expect(getBenchmarkItemsDialog(scope).getByText(questionText, { exact: false }).first()).toBeVisible({ timeout: 15000 });
+}
+/** Locator for the "Add Q&A" dialog (Manual / CSV Upload sub-tabs). */
+function getAddItemsDialog(scope) {
+    return dialogByName(scope, EVALS_LABELS.addItemsDialog.dialogName);
+}
+/** Open the "Add Q&A" dialog from the items dialog. Returns the scoped add dialog. */
+async function openAddItemsDialog(scope) {
+    await getBenchmarkItemsDialog(scope)
+        .getByRole('button', { name: EVALS_LABELS.benchmarkItemsDialog.addItems })
+        .click();
+    const dialog = getAddItemsDialog(scope);
+    // Both sub-tab triggers are unique to this dialog — a reliable ready signal.
+    await test$1.expect(dialog.getByRole('tab', { name: EVALS_LABELS.addItemsDialog.manualTab, exact: true })).toBeVisible({ timeout: 10000 });
+    return dialog;
+}
+/**
+ * Switch to a sub-tab INSIDE the Add Q&A dialog. The tab query must stay
+ * dialog-scoped: the edit-mentor modal behind it has its own `role="tab"`
+ * strip, and an unscoped `getByRole('tab')` can cross-match. Returns the
+ * active tabpanel so callers interact only with that sub-tab's controls
+ * (each panel has its own footer with its own Cancel/submit buttons).
+ */
+async function switchToAddItemsSubTab(scope, subTab) {
+    const dialog = getAddItemsDialog(scope);
+    const tabName = subTab === 'manual'
+        ? EVALS_LABELS.addItemsDialog.manualTab
+        : EVALS_LABELS.addItemsDialog.csvTab;
+    const tab = dialog.getByRole('tab', { name: tabName, exact: true });
+    await tab.click();
+    await test$1.expect(tab).toHaveAttribute('data-state', 'active', { timeout: 5000 });
+    const panel = dialog.getByRole('tabpanel');
+    await test$1.expect(panel).toBeVisible({ timeout: 5000 });
+    return panel;
+}
+/**
+ * Add Q&A pairs via the Manual sub-tab. Fills one row per pair (clicking
+ * "Add row" as needed), submits, and waits for the dialog to close. Row
+ * inputs are matched by label WITHIN each row container so the Question /
+ * Answer labels of sibling rows can't collide.
+ */
+async function addQaPairsManually(scope, pairs) {
+    const dialog = getAddItemsDialog(scope);
+    const panel = await switchToAddItemsSubTab(scope, 'manual');
+    for (let i = 0; i < pairs.length; i++) {
+        if (i > 0) {
+            await panel
+                .getByRole('button', { name: EVALS_LABELS.addItemsDialog.addRow, exact: true })
+                .click();
+        }
+        const question = panel.getByLabel(EVALS_LABELS.addItemsDialog.questionLabel).nth(i);
+        await test$1.expect(question).toBeVisible({ timeout: 5000 });
+        await question.fill(pairs[i].question);
+        if (pairs[i].answer !== undefined) {
+            await panel
+                .getByLabel(EVALS_LABELS.addItemsDialog.answerLabel)
+                .nth(i)
+                .fill(pairs[i].answer);
+        }
+    }
+    await panel.getByRole('button', { name: EVALS_LABELS.addItemsDialog.add, exact: true }).click();
+    await test$1.expect(dialog).toBeHidden({ timeout: 15000 });
+    logger.info(`Added ${pairs.length} Q&A pair(s) manually`);
+}
+/**
+ * Upload Q&A pairs via the CSV Upload sub-tab. `csvPath` is a path on the
+ * test machine; the file needs `input` (required) and `expected_output`
+ * (optional) columns. Waits for the dialog to close on success.
+ */
+async function uploadQaCsv(scope, csvPath) {
+    const dialog = getAddItemsDialog(scope);
+    const panel = await switchToAddItemsSubTab(scope, 'csv');
+    await panel.getByLabel(EVALS_LABELS.addItemsDialog.csvFileLabel).setInputFiles(csvPath);
+    await panel
+        .getByRole('button', { name: EVALS_LABELS.addItemsDialog.upload, exact: true })
+        .click();
+    await test$1.expect(dialog).toBeHidden({ timeout: 30000 });
+    logger.info(`Uploaded Q&A CSV: ${csvPath}`);
+}
+/** Locator for the Q&A delete-confirm dialog. */
+function getDeleteQaItemDialog(scope) {
+    return dialogByName(scope, EVALS_LABELS.deleteItemDialog.dialogName);
+}
+/**
+ * Delete a Q&A item by its question text: click the row's trash button
+ * (`aria-label="Delete Q&A"`) and confirm. The confirm "Delete" button must
+ * stay scoped to the confirm dialog — the items dialog's rows keep their
+ * own delete buttons mounted behind it.
+ */
+async function deleteQaItem(scope, questionText) {
+    const itemsDialog = getBenchmarkItemsDialog(scope);
+    const row = itemsDialog.getByRole('row').filter({ hasText: questionText }).first();
+    await row
+        .getByRole('button', { name: EVALS_LABELS.benchmarkItemsDialog.deleteItemAriaLabel })
+        .click();
+    const confirmDialog = getDeleteQaItemDialog(scope);
+    const confirm = confirmDialog.getByRole('button', {
+        name: EVALS_LABELS.deleteItemDialog.delete,
+        exact: true,
+    });
+    await test$1.expect(confirm).toBeVisible({ timeout: 5000 });
+    await confirm.click();
+    await test$1.expect(confirmDialog).toBeHidden({ timeout: 15000 });
+    logger.info(`Deleted Q&A item: ${questionText}`);
+}
+/** Close the benchmark items dialog via Escape and wait for it to be gone. */
+async function closeBenchmarkItemsDialog(scope) {
+    const dialog = getBenchmarkItemsDialog(scope);
+    await asPage$1(scope).keyboard.press('Escape');
+    await test$1.expect(dialog).toBeHidden({ timeout: 10000 });
+}
+/** Close the Manage benchmarks dialog via Escape and wait for it to be gone. */
+async function closeManageBenchmarksDialog(scope) {
+    const dialog = getManageBenchmarksDialog(scope);
+    await asPage$1(scope).keyboard.press('Escape');
+    await test$1.expect(dialog).toBeHidden({ timeout: 10000 });
+}
+
 /**
  * LTI tab helpers — Playwright bindings for the `AgentLtiTab` component from
  * `@iblai/web-containers`.
@@ -5214,6 +6012,7 @@ Object.defineProperty(exports, "expect", {
 exports.AuthFlowBuilder = AuthFlowBuilder;
 exports.CHAT_PRIVACY_LABELS = CHAT_PRIVACY_LABELS;
 exports.CustomReporter = CustomReporter;
+exports.EVALS_LABELS = EVALS_LABELS;
 exports.LTI_LABELS = LTI_LABELS;
 exports.LTI_TEST_IDS = LTI_TEST_IDS;
 exports.MailsacClient = MailsacClient;
@@ -5221,7 +6020,9 @@ exports.PRIVACY_LABELS = PRIVACY_LABELS;
 exports.SCREENSHARE_LABELS = SCREENSHARE_LABELS;
 exports.TASKS_LABELS = TASKS_LABELS;
 exports.VOICE_LABELS = VOICE_LABELS;
+exports.addManualScore = addManualScore;
 exports.addMemory = addMemory;
+exports.addQaPairsManually = addQaPairsManually;
 exports.archiveFirstMemory = archiveFirstMemory;
 exports.archiveMemoryByContent = archiveMemoryByContent;
 exports.billingAutoRechargeSection = billingAutoRechargeSection;
@@ -5229,13 +6030,17 @@ exports.billingCreditsSection = billingCreditsSection;
 exports.billingPlanSection = billingPlanSection;
 exports.buildReportUrl = buildReportUrl;
 exports.canChatWithEmbedMentor = canChatWithEmbedMentor;
+exports.cancelDeleteEvaluation = cancelDeleteEvaluation;
 exports.cancelDeleteInstance = cancelDeleteInstance;
 exports.cancelDeleteSkill = cancelDeleteSkill;
 exports.cancelDisconnectInstance = cancelDisconnectInstance;
 exports.cancelEnableChatPrivacyMidSession = cancelEnableChatPrivacyMidSession;
 exports.cancelKeyDelete = cancelKeyDelete;
+exports.cancelLlmJudge = cancelLlmJudge;
 exports.cancelNewInstanceDialog = cancelNewInstanceDialog;
+exports.cancelStartEvaluation = cancelStartEvaluation;
 exports.checkAdminStatus = checkAdminStatus;
+exports.checkRunStatus = checkRunStatus;
 exports.clearDateRangeFilter = clearDateRangeFilter;
 exports.clearInstanceSearch = clearInstanceSearch;
 exports.clickBackHome = clickBackHome;
@@ -5246,14 +6051,18 @@ exports.clickBillingUpgrade = clickBillingUpgrade;
 exports.clickChatPrivacyToggle = clickChatPrivacyToggle;
 exports.clickDownloadAgain = clickDownloadAgain;
 exports.clickManualDownloadLink = clickManualDownloadLink;
+exports.closeBenchmarkItemsDialog = closeBenchmarkItemsDialog;
 exports.closeCreditBalanceDropdown = closeCreditBalanceDropdown;
+exports.closeEvaluationDetailDialog = closeEvaluationDetailDialog;
 exports.closeKeyDetail = closeKeyDetail;
+exports.closeManageBenchmarksDialog = closeManageBenchmarksDialog;
 exports.closeWithEsc = closeWithEsc;
 exports.confirmEnableChatPrivacyMidSession = confirmEnableChatPrivacyMidSession;
 exports.confirmKeyDelete = confirmKeyDelete;
 exports.connectToInstance = connectToInstance;
 exports.copyEndpoint = copyEndpoint;
 exports.createAuthSetup = createAuthSetup;
+exports.createBenchmark = createBenchmark;
 exports.createEnvConfig = createEnvConfig;
 exports.createInstance = createInstance;
 exports.createKey = createKey;
@@ -5264,10 +6073,12 @@ exports.createTool = createTool;
 exports.creditBalancePanel = creditBalancePanel;
 exports.creditBalancePlanBadge = creditBalancePlanBadge;
 exports.creditBalanceTrigger = creditBalanceTrigger;
+exports.deleteEvaluation = deleteEvaluation;
 exports.deleteFirstMemory = deleteFirstMemory;
 exports.deleteInstance = deleteInstance;
 exports.deleteKey = deleteKey;
 exports.deleteMemoryByContent = deleteMemoryByContent;
+exports.deleteQaItem = deleteQaItem;
 exports.deleteSkill = deleteSkill;
 exports.deleteTask = deleteTask;
 exports.disableSkill = disableSkill;
@@ -5278,7 +6089,10 @@ exports.editLink = editLink;
 exports.editSkill = editSkill;
 exports.editTool = editTool;
 exports.enableSkill = enableSkill;
+exports.expandReview = expandReview;
+exports.expandTrace = expandTrace;
 exports.expectAllEndpointsVisible = expectAllEndpointsVisible;
+exports.expectBenchmarkListed = expectBenchmarkListed;
 exports.expectBillingAutoRechargeSection = expectBillingAutoRechargeSection;
 exports.expectBillingCreditsSection = expectBillingCreditsSection;
 exports.expectBillingPlanSection = expectBillingPlanSection;
@@ -5299,6 +6113,7 @@ exports.expectCreditBalancePanelForFreePlan = expectCreditBalancePanelForFreePla
 exports.expectCreditBalancePanelForPremiumPlan = expectCreditBalancePanelForPremiumPlan;
 exports.expectCreditBalancePanelForTrialPlan = expectCreditBalancePanelForTrialPlan;
 exports.expectCreditBalanceVisibilityForTenant = expectCreditBalanceVisibilityForTenant;
+exports.expectDetailWaitingForTraces = expectDetailWaitingForTraces;
 exports.expectEndpointUrl = expectEndpointUrl;
 exports.expectEntitySelected = expectEntitySelected;
 exports.expectFailedTasks = expectFailedTasks;
@@ -5316,14 +6131,23 @@ exports.expectLtiHeader = expectLtiHeader;
 exports.expectMentorVoiceTriggerShows = expectMentorVoiceTriggerShows;
 exports.expectNoAccessibilityViolations = expectNoAccessibilityViolations;
 exports.expectNoAccessibilityViolationsOnDialogs = expectNoAccessibilityViolationsOnDialogs;
+exports.expectNoBenchmarksNotice = expectNoBenchmarksNotice;
 exports.expectNoLogsForSelectedTask = expectNoLogsForSelectedTask;
 exports.expectOutputFilterEnabled = expectOutputFilterEnabled;
 exports.expectPrivacyFieldsHidden = expectPrivacyFieldsHidden;
 exports.expectPrivacyFieldsVisible = expectPrivacyFieldsVisible;
 exports.expectPrivateModeSelected = expectPrivateModeSelected;
 exports.expectPrivateModeTabReady = expectPrivateModeTabReady;
+exports.expectQaItemListed = expectQaItemListed;
+exports.expectReviewInList = expectReviewInList;
+exports.expectReviewStatus = expectReviewStatus;
+exports.expectRunInTable = expectRunInTable;
+exports.expectRunNotInTable = expectRunNotInTable;
+exports.expectRunStatus = expectRunStatus;
+exports.expectRunsTableEmpty = expectRunsTableEmpty;
 exports.expectScheduleStartTimeInPastError = expectScheduleStartTimeInPastError;
 exports.expectScreenShareDisabledHint = expectScreenShareDisabledHint;
+exports.expectSelectedBenchmark = expectSelectedBenchmark;
 exports.expectSttSelectDisabled = expectSttSelectDisabled;
 exports.expectTaskInList = expectTaskInList;
 exports.expectTaskNotInList = expectTaskNotInList;
@@ -5335,9 +6159,11 @@ exports.expectToolInList = expectToolInList;
 exports.expectToolNotInList = expectToolNotInList;
 exports.expectToolsEmpty = expectToolsEmpty;
 exports.expectTotalTasks = expectTotalTasks;
+exports.expectTraceScore = expectTraceScore;
 exports.expectTtsSelectDisabled = expectTtsSelectDisabled;
 exports.expectVoiceProviderSelected = expectVoiceProviderSelected;
 exports.expectVoiceVisible = expectVoiceVisible;
+exports.exportRunCsv = exportRunCsv;
 exports.fillLinkName = fillLinkName;
 exports.fillToolForm = fillToolForm;
 exports.filterByAction = filterByAction;
@@ -5347,14 +6173,19 @@ exports.filterByActorAndVerify = filterByActorAndVerify;
 exports.filterByDateRange = filterByDateRange;
 exports.generateBrowserSetupProjects = generateBrowserSetupProjects;
 exports.generateProjectConfig = generateProjectConfig;
+exports.getAddItemsDialog = getAddItemsDialog;
 exports.getAuditLogRowCount = getAuditLogRowCount;
 exports.getAvailableActors = getAvailableActors;
+exports.getBenchmarkCombobox = getBenchmarkCombobox;
+exports.getBenchmarkComboboxDropdown = getBenchmarkComboboxDropdown;
+exports.getBenchmarkItemsDialog = getBenchmarkItemsDialog;
 exports.getBillingAutoRechargeStatus = getBillingAutoRechargeStatus;
 exports.getBillingPlanLabel = getBillingPlanLabel;
 exports.getBrowserKey = getBrowserKey;
 exports.getCallConfigForm = getCallConfigForm;
 exports.getChatPrivacyConfirmDialog = getChatPrivacyConfirmDialog;
 exports.getChatPrivacyToggle = getChatPrivacyToggle;
+exports.getCreateBenchmarkDialog = getCreateBenchmarkDialog;
 exports.getCreateKeyButton = getCreateKeyButton;
 exports.getCreateLinkButton = getCreateLinkButton;
 exports.getCreateToolButton = getCreateToolButton;
@@ -5362,10 +6193,13 @@ exports.getCreditBalancePlanLabel = getCreditBalancePlanLabel;
 exports.getCreditBalanceRemaining = getCreditBalanceRemaining;
 exports.getCurrentModel = getCurrentModel;
 exports.getCurrentTenantShowPaywall = getCurrentTenantShowPaywall;
+exports.getDeleteEvaluationDialog = getDeleteEvaluationDialog;
+exports.getDeleteQaItemDialog = getDeleteQaItemDialog;
 exports.getEndpoint = getEndpoint;
 exports.getEndpointCopyButton = getEndpointCopyButton;
 exports.getEndpointUrl = getEndpointUrl;
 exports.getEntityChip = getEntityChip;
+exports.getEvaluationDetailDialog = getEvaluationDetailDialog;
 exports.getInstanceHealthLabel = getInstanceHealthLabel;
 exports.getInstanceRowCount = getInstanceRowCount;
 exports.getInstanceStatusLabel = getInstanceStatusLabel;
@@ -5387,15 +6221,23 @@ exports.getLinkNameInput = getLinkNameInput;
 exports.getLinkRow = getLinkRow;
 exports.getLinksEmptyState = getLinksEmptyState;
 exports.getLinksSection = getLinksSection;
+exports.getLlmJudgeDialog = getLlmJudgeDialog;
+exports.getLlmPickerDialog = getLlmPickerDialog;
 exports.getLtiSubTab = getLtiSubTab;
+exports.getManageBenchmarksButton = getManageBenchmarksButton;
+exports.getManageBenchmarksDialog = getManageBenchmarksDialog;
 exports.getMemoryCount = getMemoryCount;
 exports.getMentorIdFromUrl = getMentorIdFromUrl;
+exports.getNewEvaluationButton = getNewEvaluationButton;
 exports.getOutputFilterSwitch = getOutputFilterSwitch;
 exports.getPaginationInfo = getPaginationInfo;
 exports.getPrivateModeCard = getPrivateModeCard;
+exports.getReviewRow = getReviewRow;
+exports.getRunRow = getRunRow;
 exports.getScheduleTaskButton = getScheduleTaskButton;
 exports.getSearchInput = getSearchInput;
 exports.getSkillRowCount = getSkillRowCount;
+exports.getStartEvaluationDialog = getStartEvaluationDialog;
 exports.getTaskRow = getTaskRow;
 exports.getTenantChatPrivacyRow = getTenantChatPrivacyRow;
 exports.getTenantChatPrivacySwitch = getTenantChatPrivacySwitch;
@@ -5405,6 +6247,7 @@ exports.getToolModal = getToolModal;
 exports.getToolRow = getToolRow;
 exports.getToolsEmptyState = getToolsEmptyState;
 exports.getToolsSection = getToolsSection;
+exports.getTraceRow = getTraceRow;
 exports.getVoiceProviderCard = getVoiceProviderCard;
 exports.getVoiceRow = getVoiceRow;
 exports.goToFirstPage = goToFirstPage;
@@ -5413,6 +6256,7 @@ exports.goToNextPage = goToNextPage;
 exports.goToPage = goToPage;
 exports.goToPreviousPage = goToPreviousPage;
 exports.inviteUserTest = inviteUserTest;
+exports.isEvaluationTabVisible = isEvaluationTabVisible;
 exports.isFirefox = isFirefox;
 exports.isJSON = isJSON;
 exports.isLtiTabVisible = isLtiTabVisible;
@@ -5434,8 +6278,10 @@ exports.navigateToAuditLog = navigateToAuditLog;
 exports.navigateToAuditLogAndWaitForData = navigateToAuditLogAndWaitForData;
 exports.navigateToDataReports = navigateToDataReports;
 exports.navigateToReportDownload = navigateToReportDownload;
+exports.openAddItemsDialog = openAddItemsDialog;
 exports.openAddMemoryDialog = openAddMemoryDialog;
 exports.openAgentPromptEditModal = openAgentPromptEditModal;
+exports.openBenchmarkItems = openBenchmarkItems;
 exports.openCallConfigVoicePicker = openCallConfigVoicePicker;
 exports.openCreateKeyModal = openCreateKeyModal;
 exports.openCreateLinkModal = openCreateLinkModal;
@@ -5451,12 +6297,18 @@ exports.openKeyActionsMenu = openKeyActionsMenu;
 exports.openKeyDelete = openKeyDelete;
 exports.openKeyDetail = openKeyDetail;
 exports.openLLMProviderPicker = openLLMProviderPicker;
+exports.openManageBenchmarksDialog = openManageBenchmarksDialog;
 exports.openMentorVoicePicker = openMentorVoicePicker;
 exports.openNewInstanceDialog = openNewInstanceDialog;
+exports.openNewReviewForRun = openNewReviewForRun;
+exports.openNewReviewFromDetail = openNewReviewFromDetail;
 exports.openNewSkillDialog = openNewSkillDialog;
+exports.openRunActionsMenu = openRunActionsMenu;
+exports.openRunResults = openRunResults;
 exports.openScheduleTaskDialog = openScheduleTaskDialog;
 exports.openScreenSharePromptEditor = openScreenSharePromptEditor;
 exports.openSkillActionsMenu = openSkillActionsMenu;
+exports.openStartEvaluationDialog = openStartEvaluationDialog;
 exports.parseReportUrlParams = parseReportUrlParams;
 exports.previewCallConfigVoiceInline = previewCallConfigVoiceInline;
 exports.previewMentorVoiceInline = previewMentorVoiceInline;
@@ -5465,8 +6317,10 @@ exports.pushConfiguration = pushConfiguration;
 exports.readEndpointUrl = readEndpointUrl;
 exports.readKeyPublicJwk = readKeyPublicJwk;
 exports.readKeyPublicKey = readKeyPublicKey;
+exports.refreshEvaluationDetail = refreshEvaluationDetail;
 exports.reliableClick = reliableClick;
 exports.reliableFill = reliableFill;
+exports.removeTraceScore = removeTraceScore;
 exports.renameKey = renameKey;
 exports.resetCallConfig = resetCallConfig;
 exports.retry = retry;
@@ -5477,11 +6331,14 @@ exports.saveCallConfig = saveCallConfig;
 exports.saveScreenSharePrompts = saveScreenSharePrompts;
 exports.saveVoiceSettings = saveVoiceSettings;
 exports.scheduleTask = scheduleTask;
+exports.searchBenchmarks = searchBenchmarks;
 exports.searchInstances = searchInstances;
 exports.searchTasks = searchTasks;
 exports.searchVoices = searchVoices;
+exports.selectBenchmark = selectBenchmark;
 exports.selectCallMode = selectCallMode;
 exports.selectDateFromCalendar = selectDateFromCalendar;
+exports.selectJudgeLlm = selectJudgeLlm;
 exports.selectLLMModel = selectLLMModel;
 exports.selectLlmProvider = selectLlmProvider;
 exports.selectPrivacyAction = selectPrivacyAction;
@@ -5519,8 +6376,12 @@ exports.shouldSaveEditedCSVAndTriggerDownload = shouldSaveEditedCSVAndTriggerDow
 exports.shouldShowCombiningReportsDialog = shouldShowCombiningReportsDialog;
 exports.shouldVerifyCSVEditorDialogAccessibility = shouldVerifyCSVEditorDialogAccessibility;
 exports.signUpWithEmailAndPassword = signUpWithEmailAndPassword;
+exports.startEvaluation = startEvaluation;
 exports.submitLinkModal = submitLinkModal;
+exports.submitLlmJudge = submitLlmJudge;
 exports.submitToolModal = submitToolModal;
+exports.switchToAddItemsSubTab = switchToAddItemsSubTab;
+exports.switchToEvaluationTab = switchToEvaluationTab;
 exports.switchToLtiSubTab = switchToLtiSubTab;
 exports.switchToLtiTab = switchToLtiTab;
 exports.switchToMemoryTab = switchToMemoryTab;
@@ -5538,6 +6399,7 @@ exports.toggleAutoPush = toggleAutoPush;
 exports.toggleMemorySwitch = toggleMemorySwitch;
 exports.toggleSkill = toggleSkill;
 exports.toolFields = toolFields;
+exports.uploadQaCsv = uploadQaCsv;
 exports.verifyAgentConfigPromptsVisible = verifyAgentConfigPromptsVisible;
 exports.verifyAuditLogEmptyState = verifyAuditLogEmptyState;
 exports.verifyAuditLogEntryStructure = verifyAuditLogEntryStructure;