Back-of-envelope estimation

Explain to a new engineer who freezes when asked "how big will this be?":

"Walk me through sizing a 2M-events/sec feed for a year — raw throughput, stored bytes, and a rough yearly cost — and show me which two assumptions the whole number depends on."

Rubric (self-check — aim for 5/6)

• Converts events/sec + bytes/event into bytes/sec (throughput).
• Scales to a year with seconds/year ≈ 3.15 × 10^7.
• Applies compression (down) and replication (up) to get stored bytes. Says that these are the dominant assumptions.
• Separates storage cost ($/GB-month) from query cost ($/TB-scanned).
• States that pruning (FND-06) reduces TB_scanned and so reduces the bill.
• Frames the output as an order of magnitude with explicit assumptions, not a precise number.