Your starter still runs the multi-step mock pipeline inside one HTTP request. Use the live URL from deploy.
1. Complete a run
- Open your live URL.
- Enter
NVDA→ Research. - Keep the tab open until the memo appears.
You should see a live tracker for several seconds (facts, then signals / catalysts / risks, then memo), then a memo.
Optional check: DevTools → Network → /api/research stays pending, then returns JSON.
That pending request is the whole run. The tracker mirrors the mock steps, but it is still UI over one HTTP request.
2. Abandon a run
Research is delayed about 8 seconds (RESEARCH_DELAY_MS), so you have time to leave.
- Start research again with
AAPL. - While the tracker is still moving, close the tab.
- Open the same URL again.
The page looks fresh: empty form, and nothing stored to resume. There is no memo from that abandoned run.
3. What that means
The web service may still finish the old request.
Your browser never got a receipt, so the reopened page has nothing to ask for.
That missing receipt is what you fix next.
What you learned
- A full run works while the request stays open
- Leaving mid-run leaves no receipt
- Next: clone the fork and wrap the research steps as Workflow tasks