Ops Notes · 18 October 2026
The 12 checks in a Production Ops Teardown
What the free 30-minute call looks at. This is the method. It is not a client result, and it does not use anyone’s numbers.
- What is in production. Sources, transforms, models and consumers. If nothing is live yet, the check is the first system to build.
- The last incident. One recent failure, what broke, and how it was found. Not a tour of every diagram.
- Who gets paged. The person who is woken up, and whether they can see the failing table.
- Time to find the broken table. How long it takes today to go from “the number is wrong” to the table that caused it.
- Owners. Which tables have a named owner, and which do not.
- Notice before a person does. Whether a failed or late run raises an alert, or waits until someone asks.
- Rerun. Whether someone other than the author can rerun the pipeline.
- Cost. Where the warehouse or platform bill is growing faster than the work it supports.
- Base layer or the pipelines on it. Whether Databricks, Snowflake, BigQuery or the current platform is the cause, or the jobs on top of it are.
- What the call can see. Schemas, definitions, volumes and a conversation are usually enough. Production access is not the default.
- The few changes that pay off. Two or three items for the next 90 days, with effort and impact. Not a forty-item wishlist.
- Whether we are the right team. If we are not, the call ends there. You keep the notes either way.
For the call, bring your stack, one recent incident, and who gets paged.