The Silent Schema BreakQuality · medium · ~13 minAt 09:00 the CFO's office pings #data-eng: "Yesterday's revenue is showing ₹4.2 lakh. It should be around ₹4.2 crore."No pipeline failed. No schema-registry violation fired. Every job is green, row counts are normal, and the consumer services deserialized every event without error.Something upstream changed *meaning* without changing *shape* — find it, quantify it, and design the guardrail that would have caught it before the CFO did.Evidence

Metrics

Logs

Config

Checkpoints

Every system is green and yet revenue is wrong. What actually happened?

Why did the schema registry — with compatibility checking ON — not catch this?

By what constant factor is reported revenue now understated? (whole number)

Design the guardrail: what would a real data contract for `payment_events` contain, and how would you roll enforcement out so it catches the NEXT unit change before the CFO does?