Section 10: Six Ways to Break a Price Watch - Interviewer Deep-Dive Challenges

Executive Summary: The three deep dives are done, and the interviewer turns adversary. This section runs six escalations against the design: a 10× event day, a seller's $0.01 misprice, a scripted flood of fake prices, a 30-minute FCM outage, a product with 3 million watchers, and a fetch budget cut in half overnight. Each gets three beats: what it really tests, the answer with numbers, and the senior line.

The interviewer caps the marker. "Okay. Let me try to break it." Answer each challenge with a number, a mechanism and the condition that would change your mind.

Challenge One: "It's Prime Day. Everything is 10×."

What they're really asking: do you know which 10× matters? Inbound traffic is the distraction; the fetch budget and the burst are the constraints.

The answer: the Watch API peaks at about 3,500 lookups/s, around 440/s per node across 8 nodes, which is why Section 6 shipped without a cache. The fetch budget doesn't grow, so you apply Section 7's reallocation: hot tightens to 5 minutes, cold stretches to 12 hours, and the schedule uses 94% of the budget, leaving ~320 checks/s for verification. Change events climb to a budgeted 500/s, about 8/s per shard over 64 shards. The burst is what needs preparation. The 500K-watcher ASIN crossing every target means 600K pushes, and the 60-second SLO needs 10,000 pushes/s. Around 40 pods × 250/s is the floor; pre-scale ~20% above it (≈ 48 pods, ~12K pushes/s) the night before, because the guard and the pipeline already spend a few of the 60 seconds.

The senior line: "Prime Day is a config change plus a capacity plan, not a re-architecture: tighten hot, stretch cold, pre-warm senders."