feat(studio): LLM-judge evaluators

api/traces.api.md

@@ -7,6 +7,24 @@
 // @public
 export const DEFAULT_TRACE_DIR = ".lousho/traces";
 
+// @public
+export interface EvaluatorTrace {
+    agent_id: string | null;
+    duration_ms: number | null;
+    end: number | null;
+    error_message: string | null;
+    input: string | null;
+    name: string | null;
+    observations: ObservationRow[];
+    output: string | null;
+    start: number;
+    status: string | null;
+    trace_id: string;
+}
+
+// @public
+export function evaluatorTraceFrom(traceId: string, observations: readonly ObservationRow[]): EvaluatorTrace | undefined;
+
 // @public
 export function fanOutExporters(exporters: readonly (TraceExporter | undefined | null)[]): TraceExporter | undefined;
 
@@ -27,6 +45,32 @@ export interface ListTracesOptions {
     limit?: number;
 }
 
+// @public
+export interface ObservationRow {
+    agent_id: string | null;
+    attrs: Record<string, unknown> | null;
+    cost_partial: boolean | null;
+    cost_usd: number | null;
+    end: number | null;
+    input: string | null;
+    kind: string;
+    model: string | null;
+    name: string | null;
+    obs_id: string;
+    output: string | null;
+    parent_obs_id: string | null;
+    session_id: string | null;
+    start: number;
+    status: string | null;
+    trace_id: string;
+    truncated: boolean;
+    usage_cached_tokens: number | null;
+    usage_estimated: boolean | null;
+    usage_input_tokens: number | null;
+    usage_output_tokens: number | null;
+    usage_reasoning_tokens: number | null;
+}
+
 // @public
 export function observeExporter(options?: ObserveExporterOptions): ObserveReporter;
 

apps/agent-forge/e2e/revamp.spec.ts

@@ -125,6 +125,8 @@ const FILTERS: Record<ScreenId, string> = {
   scores: '?eval=e2e-eval&agent=orders-bot&source=eval&range=7d&open=e2e-eval',
   // The only URL state Datasets keeps is the server-side page (`offset`).
   datasets: '?offset=50',
+  // Queues pages server-side the same way (`?offset=`).
+  queues: '?offset=50',
   evals: '',
   prompts: '?diff=1,2',
   hub: '?agent=triage-bot&scope=fleet',
@@ -195,6 +197,9 @@ const CONTROLS: Record<string, (main: Locator, page: Page) => Promise<void>> = {
   '/evals/datasets/:name': async (main) => {
     await expectPreviousPage(main);
   },
+  '/evals/queues': async (main) => {
+    await expectPreviousPage(main);
+  },
   '/prompts/:name': async (main) => {
     await expect(main.getByRole('combobox', { name: 'Compare version A' })).toHaveValue('1');
     await expect(main.getByRole('combobox', { name: 'Compare version B' })).toHaveValue('2');
@@ -218,6 +223,7 @@ const NO_CONTROLS = new Set([
   '/threads/:sessionId', // a thread's turns; the list's filters only ride along in the URL
   '/evals/:section', // evaluators / annotation placeholder screen
   '/evals/:section/:id', // evaluators / annotation placeholder screen
+  '/evals/queues/:id', // one queue's review pane; no URL-backed controls beyond the id
   '/prompts', // the registry list has no filters
   '/settings', // no URL state
 ]);

apps/agent-forge/playwright.config.ts

@@ -54,6 +54,8 @@ export default defineConfig({
       BASE_DIR: baseDir,
       LOUSHO_STUDIO_TOKEN: E2E_STUDIO_TOKEN,
       LOUSHO_OBSERVE_URL: E2E_OBSERVE_URL,
+      // No daemon here: keep the LLM-judge loop (server/judge.ts) from polling it.
+      LOUSHO_STUDIO_JUDGE: '0',
     },
     url: `http://127.0.0.1:${port}/health`,
     reuseExistingServer: false,