Rebuild summaries whenever the add-models box is ticked

.github/workflows/add-models-to-runs.yml

@@ -9,7 +9,7 @@ on:
         description: 'Comma-separated config IDs whose latest run gets the models'
         required: true
       rebuild_summaries:
-        description: 'Rebuild homepage, leaderboards and model summaries at the end'
+        description: 'Rebuild homepage, leaderboards and model summaries at the end (even if nothing changed)'
         type: boolean
         default: true
       rejudge:

src/app/api/internal/add-models-to-runs/__tests__/route.test.ts

@@ -47,7 +47,7 @@ describe('POST /api/internal/add-models-to-runs', () => {
     expect(vi.mocked(addModelsToLatestRun).mock.calls[0][3]).toEqual({ rejudge: true, retryFailed: true });
   });
 
-  it('keeps going after a blueprint throws, and skips the rebuild when nothing was added', async () => {
+  it('keeps going after a blueprint throws, and rebuilds when asked even if nothing was added', async () => {
     vi.mocked(addModelsToLatestRun)
       .mockRejectedValueOnce(new Error('boom'))
       .mockResolvedValueOnce({ configId: 'b', status: 'skipped' });
@@ -57,6 +57,16 @@ describe('POST /api/internal/add-models-to-runs', () => {
     await flush();
 
     expect(addModelsToLatestRun).toHaveBeenCalledTimes(2);
+    expect(actionBackfillSummary).toHaveBeenCalledTimes(1);
+  });
+
+  it('does not rebuild unless asked', async () => {
+    vi.mocked(addModelsToLatestRun).mockImplementation(async (configId) => ({ configId, status: 'added' }));
+
+    const res = await POST(request({ models: [APERTUS], configIds: ['a'] }));
+    expect(res.status).toBe(202);
+    await flush();
+
     expect(actionBackfillSummary).not.toHaveBeenCalled();
   });
 

src/app/api/internal/add-models-to-runs/route.ts

@@ -17,7 +17,8 @@ const MAX_CONFIGS = 300;
  *
  * Returns 202 immediately and works through the blueprints one at a time in
  * the background, logging under "add-models:". With rebuildSummaries, it
- * rebuilds the homepage, leaderboards and model summaries once at the end.
+ * rebuilds the homepage, leaderboards and model summaries once at the end,
+ * even when no blueprint changed (so the box doubles as a plain rebuild).
  * With rejudge, listed models a run already has are judged again from their
  * saved responses instead of being skipped. With retryFailed, they are asked
  * again on the prompts where their answer failed.
@@ -91,7 +92,7 @@ async function runJob(
   logger.info(`[AddModels] Done: ${count('added')} added, ${count('skipped')} skipped, ${count('failed')} failed.`);
   for (const r of results.filter(r => r.status === 'failed')) logger.warn(`[AddModels] Failed: ${r.configId}: ${r.reason}`);
 
-  if (rebuildSummaries && count('added') > 0) {
+  if (rebuildSummaries) {
     try {
       logger.info('[AddModels] Rebuilding homepage, leaderboards and model summaries...');
       await actionBackfillSummary({ verbose: false, dryRun: false });