| | AI is speeding up analytics work, but the fundamentals still decide whether anyone trusts the output. dbt’s 2026 State of Analytics Engineering Report looks at AI-assisted coding, governance gaps, infrastructure costs, and the growing pressure to deliver reliable insights faster. Learn more.
*Message from this week's sponsor, dbt Labs |
| | |
| 🔗 Note: Data Elixir is taking next week off for the holiday weekend and will be back in your inbox in two weeks. | Also, I’m still hearing that some Outlook configurations are blocking links in Data Elixir. If that includes you, this issue is also available on the web: dataelixir.com/archive/issue-579.html | Have a great couple of weeks! | -Lon |
| | |
| | | Everyone agrees that data quality matters but nobody agrees on what it means. Abraham Thomas offers a four-rung ladder for thinking about quality, from unit-level accuracy to business outcomes. The useful part is how it explains why data teams so often talk past each other when they argue about “bad data.”
Pivotal | Abraham Thomas |
| | |
| | This is a fun, surprisingly clarifying experiment: can gzip act like a language model? Sort of. By priming it with a corpus and using compressed length as a scoring function, Nathan Barry shows how prediction is hiding inside compression. The generated text is messy, but the idea lands beautifully.
Nathan Barry |
| | |
| | A practical pytest guide for Python users who have moved beyond one-off scripts and want their code to be easier to trust, refactor, and maintain. It covers fixtures, parametrized tests, mocks and stubs, generated test data, dependency injection, and snapshot testing, with examples that are useful for making data-adjacent Python projects more reliable.
Andros Fenollosa |
| | |
| | Recovered Factory turns live data into social video by treating the video as a web page. A hidden 9:16 page renders the map, charts, counters, and captions; Playwright captures it frame by frame; ffmpeg stitches the result into video; and GitHub Actions regenerate drafts when the data changes. It’s a clever pattern for turning data products into shareable, reproducible media without manual exports or stale numbers.
Recovered Factory | David Eads |
| | |
| | The best data visualizations often depend on choices readers never see. Dominic Royé uses a temperature anomaly chart to explore a subtle but important problem: what should the baseline be? The post compares different ways to model “normal” temperatures and shows how an overly smooth curve can hide meaningful seasonal structure. It’s a good example of reproducible analysis, baseline construction, and defensible chart-making.
Dominic Royé |
| | |
| | Ready to hone your skills? Fidelity can provide the platforms and dedicated time needed for mastering new technologies. Enjoy the curiosity of a startup with the stability of a private firm that cares about helping you grow. Discover your tech future at Fidelity.
// sponsored by Fidelity Careers |
| | |
| | | | Good discussion of F3 and the bigger question of what could realistically come after Parquet. The thread gets past the launch-page pitch and into the hard parts: compatibility, random access, ML workloads, embedded Wasm decoders, security, and the enormous advantage of being supported everywhere.
Hacker News - 655 upvotes |
| | |
| | | For students and recent grads who missed out on internships, Hack Your Summer offers a practical alternative: spend four weeks building a public-facing project with mentor support, peer accountability, and demo sessions. It’s free and virtual, but unpaid; the point is to leave with evidence of what you can do.
Hack Your Summer |
| | |
| | | For almost 2,000 years, the Herculaneum scrolls were a cruel kind of archive: preserved by the same eruption that buried Pompeii, but too fragile to open. Now one has been read end-to-end without touching it. The Vesuvius Challenge used X-ray scans, geometry, and machine learning to virtually unwrap the scroll and recover its text.
Vesuvius Challenge |
| | |
| | | This week's issue sponsored by: | | |
|
