Who Decided That?
Data piles up, agents pick tables by their name, and nobody remembers deciding any of it.
In progress · 3 of 10 parts published
Start reading → - Prologue The Reader Changed October 13, 2026 · I closed Trust Erosion saying the fix is a better reader. Then the reader changed: now it's an AI agent, and it doesn't ask questions.
- Part 1 The ROT Weighs October 22, 2026 · Redundant, obsolete and trivial data has mass. The cloud keeps it, and nobody decided to.
- Part 2 The Agent Chooses October 28, 2026 · For an agent, every table is one step away. Only the name is left to choose by. Who decided which tables it sees?
- Part 3 Written for Someone Else November 5, 2026 · Eighty-two measure descriptions, all written by an agent, all approved. Most said nothing. Who were they written for?
- Part 4 Coming soon. Subscribe to get it by email.
The Trust Erosion Series
Every number in this series was correct. None of them earned the room's trust.
Complete · 10 parts
Start reading → - Part 1 False Semantic Consensus July 15, 2026 · Everyone agreed on the number. That was the problem.
- Part 2 The Table Is Not the Concept August 6, 2026 · You can define the concept perfectly and still model it wrong.
- Part 3 Your Glossary Isn't Semantics August 12, 2026 · A definition tells you what a customer is. It doesn't tell you which rows qualify.
- Part 4 When Aggregation Lies August 18, 2026 · The general manager wanted his quantity column. The model kept returning a blank. The thirty minutes that followed were not about the formula.
- Part 5 The Number Without Perimeter August 27, 2026 · +12%. Correct to the decimal. Nobody in the room could defend it.
- Part 6 The Metric as Mirror, Not Window September 1, 2026 · The number said the ER was excellent. Sitting in that waiting room, it told me nothing I needed to know.
- Part 7 Goodhart's Law in Practice September 9, 2026 · He won awards for the number. The number was the problem.
- Part 8 When Obedience Is the Disaster September 14, 2026 · Nobody gamed the number. They obeyed it perfectly. That was the disaster.
- Part 9 When Code Betrays the Model September 22, 2026 · The definition was right. The code said something else -- and nobody was reading the code.
- Part 10 The Discipline of Distrust September 29, 2026 · Every number in this series lied. The fix was never a better number -- it was a better reader.
- Epilogue The Reader Changed October 13, 2026 · I closed Trust Erosion saying the fix is a better reader. Then the reader changed: now it's an AI agent, and it doesn't ask questions.
YAML Metadata-Driven Ingestion
A Bronze layer that disappears behind a YAML commit, and what production taught us about it.
Complete · 6 parts
Start reading → - Part 1 Bronze Should Be Boring April 16, 2026 · Designing a metadata-driven ingestion layer that disappears behind a YAML commit
- Part 2 Metadata All the Way Down April 18, 2026 · Governing your ingestion metadata with more metadata -- and knowing which metadata is which
- Part 3 The Scheduler's Contract April 23, 2026 · How scheduling becomes infrastructure -- data freshness as a promise, not a hope
- Part 4 Battle Scars April 25, 2026 · What happens when a YAML-driven platform meets production reality
- Part 5 Runbooks as Infrastructure May 2, 2026 · The 3am test, and why the instruction for a human is engineering, not documentation
- Part 6 We Act on Metadata May 4, 2026 · How a boring ingestion layer becomes the foundation for ops, quality, and governance -- without a separate catalog project