# Cursor missing-history proof

Environment for every capture: real web client, Chromium Browser panel, 1280 x 800 CSS viewport, light theme, November 12, 2026 UTC. Each version runs its own development server and fresh database with T3CODE_DEV_SEED=0; no production database is read or copied. Same live Cursor account on both sides, supplied through CURSOR_API_KEY in the server environment. No mocked API responses or cached-history fixtures.

Base: ed54f30644c2e51bc03d5786dc1f0707b66dea88
Head: f9ba0da9cf544492a1f797cdd2e4eb39d50c671c

1. Install each checkout with `vp i`, then start `T3CODE_DEV_SEED=0 vp run dev`.
2. Enable Cursor with `python3 seed.py CHECKOUT` after settings.json exists. This is the same settings edit applied during verification.
3. Make a real Cursor request in a scratch directory: `cursor-agent -p --trust --mode ask --model auto 'Reply with just OK. Do not use any tools.'`. It returned OK at approximately 07:09 UTC.
4. Pair the Browser panel to each server using its private startup link. Open `/usage`, select Tokens, choose Cursor alone from All providers, and select 30 days.
5. Allow the live account history fetch to complete. The head's first scan showed a transient read error; refreshing fetched the history successfully. Both independently fetched caches ultimately contained 103,901 records, newest November 3 at 20:30:33.799 UTC. Both 30-day views showed 1.51B tokens across 4,474 sessions.
6. Select Day in Breakdown to avoid exposing model identifiers. Capture the settled 30-day view. For each published still, two screenshots at least 500 ms apart were byte-identical.
7. Record the same interaction: from the 30-day view, select Past 24h and allow it to settle. Capture the settled zero-history view.

Observed: base silently shows no activity for the last 24 hours and no notice on either window. Head preserves the same history and displays “Cursor has not reported your recent usage history.” on both windows.

Direct live diagnostics confirmed 200 empty objects for 24-hour and eight-day history ranges, while the current billing cycle reports approximately 2.98% used. No diagnostic payloads, credentials, account identifiers, or raw event rows are included in these uploads.

This proves the notice fallback, not recovery of the missing events. Replay scenarios cannot reproduce Cursor-specific live endpoint behavior; manual verification was used as directed.

AI-assisted; verified with the tests and tools listed above.