**Symptom:** The signup service returned 500 errors during a marketing-driven traffic spike, blocking new signups for approximately the time it took the on-call engineer to scale up the service. **Trigger:** The 5x traffic spike from the 9am marketing campaign launch. **Contributing factors:** 1. The signup service runs with auto-scaling disabled (or with thresholds set for normal load). Auto-scaling would have responded to the spike before it produced errors. 2. The team that ran the marketing campaign did not coordinate with the engineering team in advance. Engineering would have pre-scaled if notified. 3. No load testing was performed at the projected campaign volume. The team didn't know whether the current capacity would hold. 4. The first signal that something was wrong was the on-call engineer being paged by a 500-rate alert, not a leading indicator like queue depth or CPU utilization. By the time the alert fired, customers were already affected. 5. The fallback experience for customers was a 500 page, not a graceful "we're at capacity, please try again in 5 minutes" page that would have preserved the relationship. **The "why didn't we catch this earlier" branch:** The marketing campaign was scheduled in a shared calendar but engineering doesn't routinely check that calendar. A simple cross-functional notification policy ("campaigns expected to drive >2x normal traffic notify engineering 48 hours in advance") would have caught this. The monitoring picked up the problem only after errors started occurring. Leading indicators (CPU, request queue, database connection saturation) were available but not wired to alerts that page on-call before errors hit users. **The "why does this keep happening" branch:** The user noted this is the postmortem for yesterday's incident specifically. Worth checking: has the signup service had similar incidents before? If yes, the contributing factors above are probably stable patterns rather than one-time gaps. If no, this is the moment to fix them before they recur. **What to fix vs. what to accept:** 1. **Enable auto-scaling** — fix. The single biggest lever here and it's well-understood operationally. 2. **Cross-functional notification policy** — fix. Cheap; mostly a process change. 3. **Load testing for campaign launches** — fix as a process, not as automation. Before any campaign expected to >2x normal traffic, run a quick load test against staging. 4. **Leading-indicator alerts** — fix. Wire CPU and request queue depth to alerts that page before error rates do. 5. **Graceful degradation page** — accept for now. Worth doing eventually; not a top priority unless the team expects this scenario to recur often. **Lessons that aren't "we should be more careful":** - Add a campaign-volume notification to the team-handbook process for marketing. - Configure auto-scaling on the signup service with sane thresholds (and test that it actually scales by triggering it once in a controlled environment). - Add a leading-indicator alert on CPU > 70% sustained for 5 minutes on the signup service. - Hold a 30-minute joint review with marketing and engineering quarterly to compare upcoming campaigns to current capacity.