The tasks THIS evaluator can pick right now — the dashboard list.

Reuses the exact eligibility the router uses to serve the next case (store.eligible_tasks_for_evaluator already drops tasks the evaluator has labeled or that are at capacity), with the same hard/real filters, so the dashboard count matches what /tasks/next would hand out. Multimodal is only a serving PREFERENCE (V3 falls back to text), so it is NOT a hard filter here: the list shows everything the evaluator may work on. Cards are lightweight metadata only — the case content still loads through the existing task flow.