The Vertex JournalPractice
The medallion architecture is a slide
Bronze, silver and gold describe a diagram, not a design. On the vocabulary that conquered data engineering without ever being specified.
Reviews
32 entries · complete archive
Tooling
The lakehouse wants the SIEM's storage bill
Six security-log connectors shipped in one day. Databricks is bidding for the most expensive gigabytes in enterprise logging, and the connectors are the easy half.
Platforms
Arrow won by disappearing
The interchange layer no vendor advertises and every engine ships. A note on the standard that succeeded by becoming boring.
Economics
The warehouse pricing models, reviewed as designs
Per-credit, per-slot, per-TB and serverless per-second are not four prices for one thing. They are four incentive systems, and each one trains your team differently.
Tooling
Zero-ETL is ETL you stopped operating, not ETL that stopped existing
AWS defines zero-ETL as integrations that minimize the need to build pipelines. Minimize is the honest word in the sentence. A short review of what the prefix actually deletes, and what it renames.
Platforms
Managed Iceberg, the S3 Tables verdict so far
AWS made the table format a bucket feature. What the managed layer actually buys, what it costs against running your own, and who should wait.
Platforms
DuckDB, two years into 1.x: the re-verdict
We called the single-node renaissance early. Time to check the review against what actually happened, which is the part of criticism most outlets skip.
Economics
A BI seat costs $12 or $115 now, when the vendor still sells one
Power BI holds at $14, Tableau runs $75 to $115, Metabase undercuts both, and Qlik and Domo stopped pricing people entirely. The seat, reviewed.
Platforms
Iceberg v3 outran the engines that read it
Snowflake, Databricks and Dremio shipped Iceberg v3 GA this spring. Trino is partial, Athena and BigQuery still read v2. The interop math, reviewed.
Practice
The four dashboards every data team builds, reviewed as a genre
The executive mosaic, the wall of red, the vanity grid and the one that works. A taxonomy, with verdicts, of the artifacts BI budgets actually produce.
Platforms
Open Semantic Interchange became Apache Ossie: fifty logos, one YAML file, verdict pending
The semantic-layer interchange standard has a real spec, a real Apache incubator and a real gap between signing and shipping. We read the YAML so you can read the roadmaps.
Tooling
DuckDB can write to an Iceberg catalog now, in an alpha
dbt Core's v2 alpha gives DuckDB read-write Horizon and Unity catalogs over Iceberg REST. Useful, checkable, and not yet something to put on a migration plan.
Tooling
dbt Core stayed Apache 2.0, and that was the only question
Fivetran and dbt Labs closed their merger the same day dbt Core v2 shipped. Only one of those announcements changed what a data team can rely on.
Practice
Wait is a verdict, not an absence of one
In defense of the least popular conclusion a review can reach, and why the six-month delay keeps outperforming the launch-week migration.
Practice
The benchmark you didn't run is the one that matters
Vendor benchmarks measure the vendor's best day. A short argument for running your own worst one, with our usual apologies to marketing.
Practice
Governance that ships beats governance that reviews
Access control moved into catalogs and policy moved into code. The governance programs that work in 2026 are the ones nobody has to schedule a meeting with.
Platforms
The table format standoff is over and nobody surrendered
Iceberg, Delta and the rest converged on features and moved the war to catalogs. Where the actual lock-in lives in 2026, and what to standardize on.
Tooling
Orchestrators in 2026, the gap is closing from both sides
Airflow 3.2 shipped asset partitioning, Dagster went pay-as-you-go, and the asset-versus-task argument is ending in convergence. A review.
Economics
The real cost of switching BI tools
The license comparison is the cheapest part of the spreadsheet. Where BI migrations actually spend their budget, and the line items procurement never sees.
Practice
Data contracts work exactly where people said they would
Three years after the discourse peak, contracts settled into a useful, unglamorous role: schema enforcement at team boundaries. A review of what stuck.
Platforms
Postgres is enough until it very suddenly isn't
pg_duckdb hit 1.0 and made the boring answer better. When your warehouse should be your Postgres, and the three signals that the honeymoon is ending.
Economics
Warehouse cost levers, ranked by yield
Most warehouse bills respond to four interventions. The ranking, the yields we have seen, and the lever everyone pulls first that matters least.
Tooling
ELT won. The bill for winning has arrived.
Transformation moved into the warehouse and brought its costs along. What the Fusion-era tooling race says about where the pain actually is.
Practice
Schema evolution strategies, from polite to catastrophic
Schemas change; the only choice is how much warning anyone gets. The four strategies in the wild, reviewed from safest to most exciting.
Platforms
The single-node renaissance is a correction, not a fad
DuckDB made one machine respectable again. A review of what the distributed decade got wrong about the size of actual data.
Tooling
Transformation tooling in 2026, ranked by what actually hurts
The SQL transformation layer has one incumbent, one architectural challenger and a lot of noise. A rundown organized around the pain each tool removes.
Platforms
Lakehouse or warehouse is a staffing question
The formats converged, the engines cross-read each other, and the real decision is about the team you have, not the architecture you admire.
Practice
Change data capture, the primer we keep re-explaining
CDC turns the database log into a stream of what changed. How it works, where it beats polling, and the operational bill nobody quotes up front.
Tooling
Reverse ETL, or the pipeline that apologizes for the last one
Syncing warehouse tables back into SaaS tools is sometimes necessary and often a symptom. How to tell which one you have.
Practice
The minimum viable data quality suite is four checks
Before buying an observability platform, run the four checks that catch most incidents. A review of the floor, not the ceiling.
Practice
Real-time streaming is a bill most teams never needed
Real-time pipelines cost real money and most dashboards are read weekly. A defense of batch, with the arithmetic that vendors skip.
Tooling
Semantic layers, a reality check
Define the metric once, serve it everywhere. The promise is right, the adoption math is harder, and the graveyard has a pattern worth reading.