Top Last 100 Articles Here <— Click

Martyn Rhisiart Jones

Aranjuez, 20th December 2023

The Medallion Architecture, or: How We Learned to Stop Thinking and Love the Bronze Table

Somewhere, right now, as you read this, a data engineer is renaming a table. Again. They’re calling it Bronze, because that’s what the priests say it must be called. Bronze. Not “raw”, not “stuff we barely understand”, not “this arrived at 3AM and no one knows why”. Bronze. They do it solemnly, reverently, like a monk copying out the Book of Leviticus but with YAML. And when they are done, they will feel a sense of calm, because order has been restored to the universe. Temporarily.

This is the Medallion Architecture. It began life as a perfectly sensible idea. It was like seatbelts, indoor plumbing and chocolate teapots. It has since metastasised into a full-blown enterprise religion, a crazy sect without rhyme or reason. Consultants preach it. Vendors engrave it on stone tablets. Chief Data Officers nod sagely, as if Moses himself just came down from the mountain clutching a Delta Lake whitepaper.

Bronze, Silver, Gold. Raw, refined, ready. A journey toward truth. A spiritual pilgrimage for CSV files.

In practice, Bronze is a skip (a dumpster, a roll-off container, or a debris box). A flaming skip. It is a landfill of unloved, undocumented and unbelievable data. Automated pipelines tip data in, and nobody dares turn them off because something or somebody might be using them. Silver is where the work is supposed to happen. This includes deduplication, schema enforcement, and business logic. However, it’s usually done in a hurry. Often, this work happens on a Friday by someone who has already handed in their notice. And Gold, the shining city on the hill? Maybe nobody uses it at all. Perhaps everyone uses it and hates it. Or possibly, it contains twenty-one slightly different definitions of “revenue” because nobody could agree, and time was running out.

And the money. Oh my God, the money. Every layer needs pipelines. Pipelines need monitoring. Monitoring needs dashboards. Dashboards need meetings. Storage multiplies like rabbits on a government grant. Compute bills grow until Finance starts asking questions like, “What exactly is a Silver?” or “WTAF are you doing?” and nobody can answer without a diagram and a deep sigh.

Add one column to a source system. The entire structure collapses like a Jenga tower. It is as if it were built by people who hate each other. Bronze breaks. Silver panics. Gold sulks. Tickets are raised. Change boards convene. Three weeks later, the business asks why the numbers don’t match Excel. The data scientists have been sensibly querying Bronze directly this whole time. They quietly look at their shoes.

Because here’s the thing: the Medallion Architecture isn’t about data. It’s about morality. It’s about being seen to do data properly. It provides a bureaucratic comfort blanket for organisations. They won’t fix ingestion and won’t enforce contracts. They also won’t delete anything and won’t admit that half their data is utter rubbish. Medallion doesn’t solve bad data culture; it gives it a laminated badge and a governance committee.

It’s also, purely coincidentally, absolute catnip for vendors. Funny, that. The more layers you build, the more compute you burn. The more compute you burn, the more invoices arrive. The loudest voices claim this is “industry best practice.” These same voices are selling the very machines that keep the hamster wheel spinning.

Defenders will say it gives lineage. Yes, until the lineage diagram looks like the London Underground drawn by someone having a breakdown. They’ll say it enforces quality. Yes, until Gold becomes another swamp with better lighting. They’ll say it scales. Yes, right up until your CFO develops a nervous twitch.

The alternative isn’t chaos. Its simplicity is so aggressive that it makes people uncomfortable. Fix the data when it comes in. Enforce contracts. Test it continuously. Have fewer tables. Delete things. Say no. These are not revolutionary ideas. They are just unfashionable, because you can’t sell them as a three-layer diagram with a gradient.

The real tragedy is that organisations keep adding more layers. They are convinced that this time—this new Silver, this improved Gold, will make trust magically appear. Years pass. Millions vanish. The business is still waiting.

Because the problem was never that there weren’t enough layers.

The problem was believing that more layers could replace thinking.

And that, I’m afraid, is not architecture. It’s theatre.

Thanks for reading, and a Merry Christmas.