ABOUT THIS EPISODE
Ben Jones turned his book Avoiding Data Pitfalls into a Claude connector, an MCP server with 78 rules that lets Claude check its own data work. The connector never sees your data. It gives the session the rules and review instructions, and Claude does the checking.
In this solo episode, Ben builds a dashboard with Claude's new dashboards artifact from U.S. Department of Energy vehicle registration data, runs it through the connector, and walks through what came back:
• A 2025 hybrid jump that looks like a classification change, not real growth
• A summary comparing against a baseline year from before a known break
• Big percent changes on tiny, rounded counts
• One color meaning three different things in adjacent charts
• Side-by-side charts with axes that don't line up
• Registration estimates described as vehicles on the road
• A missing color legend
…and the pitfalls the dashboard avoided.
He closes with two other ways to use the same rules: the free datapitfalls CLI on npm, and the Pitfall Detector in the Lab on dataliteracy.com.
Chapters
(0:00) Welcome, and why Avoiding Data Pitfalls needed a new form
(1:28) Data Pitfalls: a Claude connector so Claude can check its own work
(2:19) The data: U.S. Department of Energy vehicle registrations by fuel type
(2:42) Building a dashboard with Claude's dashboards artifact
(3:09) Adding the Data Pitfalls connector (and why it never sees your data)
(4:07) A tour of the dashboard Claude built
(5:06) Running the dashboard through all 78 rules
(5:34) Pitfall 1: the 2025 hybrid jump and measurement system drift
(6:55) Pitfall 2: a baseline year from before the break
(7:18) Pitfall 3: big percent changes on tiny, rounded counts
(9:08) Pitfall 4: one color, three meanings
(10:30) Pitfall 5: side-by-side charts with mismatched axes
(11:28) Pitfall 6: registrations aren't vehicles (the data-reality gap)
(12:22) Pitfall 7: a missing legend, and the pitfalls it avoided
(12:47) Applying all seven fixes
(13:11) From good to great: why fine-tuning decides adoption
(14:17) The datapitfalls CLI on npm
(14:47) The Lab on dataliteracy.com and the Pitfall Detector
(15:31) Demo: scanning a campaign bar chart
(16:22) Wrapping up, and guests coming soon
Links
Episode page and full transcript: https://poweredbydata.com/podcast/pitfall-proof-your-claude-dashboards
Data Pitfalls connector for Claude: https://claude.ai/directory/data-pitfalls
Pitfall Detector in the Lab: https://dataliteracy.com/lab/pitfall-detector
Data Pitfalls Detector: https://poweredbydata.com/tools/data-pitfalls
datapitfalls on npm: https://www.npmjs.com/package/datapitfalls
Avoiding Data Pitfalls: https://www.avoidingdatapitfalls.com/
Note: this is a screen-share episode, so the video version on Spotify and YouTube shows the demos.
