For a full year, I measured the Data-Freshness SLO of Kafka streaming services using Throughput and Offset Lag. I built dashboards, tuned thresholds per topic and consumer group — and it still didn't work.

Then I asked myself: why am I measuring *how many* messages are waiting, instead of *how long* they've been waiting?

In this session, I'm sharing with the audience:
- Why Offset Lag and Throughput are misleading (with real examples)
- Why estimating Time Lag from OffsetLag/Throughput is useless
- What actual Time Lag is and how it solves the problem
- The before/after of our SLO dashboard

If you run Kafka services and you're trying to define an SLO that actually means something to the business — this talk is for you.