Guides: the problem, why it happens, the fix
Each guide starts from a symptom the way you would type it into a search box, explains the cause in plain words, and walks through the fix with the code or settings involved. Built from public data and documented behaviour, never from a client's system.
Databricks
The pipeline failed because a source added a column
Why a new or renamed column breaks a load at 3 am, and how schema evolution, the rescued data column and expectations make it a notice instead of an outage.
About 8 minutes Databricks, costThe Databricks bill doubled and nothing changed
The five settings that quietly multiply a bill, the system table query that shows where the money went, and the order to fix them in.
About 9 minutes Databricks, governanceUnity Catalog grants that actually work
Why "permission denied" follows a grant, the three-level model, groups and service principals, row filters, and the audit query that shows who read what.
About 8 minutes Analysis, DuckDBDuckDB for the analysis that does not need a cluster
Reading Parquet straight over HTTP on a laptop, the 27-second query over 4.3 million trips, where it stops fitting, and the handover to a lakehouse.
About 7 minutesPower BI
The nightly refresh takes hours
Setting up incremental refresh so it actually works: the two parameters, query folding, the policy, the first refresh, and the two traps that make people give up on it.
About 9 minutes Power BI, semantic modelTwo reports show different revenue
Why every report carrying its own definitions ends in a meeting about whose number is right, and how one shared semantic model with written measures ends the meeting.
About 7 minutes Power BI, securityRow-level security without killing performance
A mapping table, one dynamic role, a filter on the dimension, and the three ways RLS makes a fast report slow.
About 8 minutesAI assistants and agents
The assistant answers questions it should refuse
A safety classifier in front of the model: what it checks, how it fails safely when it is slow, the keyword fallback, and what to log so you can prove it worked.
About 10 minutes AI, evalsEvals before you ship an agent, in a week
How to write the test set, which cases must be in it, how to grade answers, what threshold to set, and how to run it on every change so a prompt edit cannot break a behaviour quietly.
About 9 minutes AI, agentsTool permissions and approval steps for agents
A tool registry with scopes, reads separated from writes, an approval gate for anything that changes a system, and the audit trail that proves it.
About 9 minutesOperations and checklists
A runbook someone can use at 3 am
The one-page template per alert, a worked example for a failed pipeline, and the drill that proves the page works before it is needed.
About 7 minutes Databricks and Fabric, costTen settings that keep a Databricks or Fabric bill down
Settings and habits, not rewrites, in the order that matters, each marked for the platform it applies to.
About 6 minutesHave the symptom but not the time?
Send the three lines and we reply within a working day with which guide applies, or with a quote if it needs hands.