Posts

All the articles I've posted.

• Your Cache Hit Rate Is Not Your Token Savings

  Published:Sep 17, 2026 |  at 06:00 AM

  Measure uncached input, cache writes, cache reads and output separately. Request hit rate cannot tell you what a long-context workload costs.

• Kafka Rebalancing Changed. Did Your Consumer Configuration?

  Published:Sep 17, 2026 |  at 06:00 AM

  Kafka 4.0 introduced the GA consumer protocol. Separate broker capability, client opt-in and server-owned settings before migrating a group.

• Your RAG Works Until You Add a Tenant Filter

  Published:Sep 17, 2026 |  at 06:00 AM

  Measure nearest-neighbor recall inside the eligible corpus. A global candidate budget can starve filtered retrieval even when authorized answers exist.