Compute an Error Budget and Decide a Release Policy

A checkout API has an SLO of 99.95% success over a rolling 30-day window, serving roughly 4M requests/day.

1. Compute the total error budget (allowed failed requests) for the 30-day window.
2. 22 days into the window, 38,000 requests have failed. Compute the remaining budget and the current daily burn rate. At that burn rate, will the budget be exhausted before the window closes? Show the arithmetic.
3. Based on your answer to (2), what concrete policy decision should the team make about an upcoming risky release (a major payment provider migration) scheduled for day 25, and why?