Sandbox #1: The 25-Show DMB Dataset
Sandbox #1 — Turn data you forgot you had into something you can see.
I have been to twenty-five Dave Matthews Band concerts.
I did not know that number off the top of my head. I had a vague sense of a lot, maybe twenty?, but I couldn’t have told you how many shows, in what years, at which venues, alongside which gap years. The data existed. A site called DMBAlmanac tracks shows for anyone willing to log them, and somewhere around 2009 I started logging mine. Then I forgot I was doing it. Then sixteen years went by.
This week I exported the data. Dropped it into Claude. Built three different things from the same dataset in about ninety minutes of a spreadsheet, a research-style analysis, and a one-page handout I could hand to my Dave fans who care to know these things.
The lesson isn’t about Dave Matthews. The lesson is that one piece of source material can become three artifacts in the time it used to take to build one. That’s the move worth practicing this week.
I’ll walk you through all three. Each one has a prompt you can copy. Each one ends in something you can use, share, or throw away.
Experiment 1: Turn screenshots into a structured spreadsheet
DMBAlmanac doesn’t let you export. The data lives on the page with venue names, dates, song counts, setlist details, but there’s no download button. So I did what most people don’t know is possible: I took nine screenshots of my member page and asked Claude to turn them into a spreadsheet.
The prompt:
I’m attaching screenshots from my profile page on DMBAlmanac.
This is every Dave Matthews Band concert I’ve attended.
Please build me an Excel spreadsheet with multiple sheets:
- Every show in chronological order (date, venue, city, state, length)
- Top-line stats (shows attended, songs seen, venues, states)
- Shows per year, with notes on gap years
- Venues ranked by frequency
- States ranked by frequency
- Most-seen and least-seen songs
- Longest single-song performances
- Band members seen (current, past, guests)
- Setlist highlights (opener, closer, encore by show)
Make it clean enough to share. Add a README sheet explaining what each sheet contains.
The result was an eleven-sheet workbook with a README, every show logged, every venue ranked, every band member counted. I did not type a single row of data. Claude read the screenshots, extracted the structured information, and built the file.
Try this with your own life:
Any data that lives somewhere you can screenshot but not export. Your Strava history. Your Goodreads list. Your district’s public board meeting archive. Your Substack stats page. Take five screenshots, paste them in, ask for a spreadsheet. Watch what happens.
Try this with your work:
The data that exists in a vendor dashboard but won’t export cleanly. Curriculum mapping that lives in someone’s PowerPoint. A roster page that prints ugly. Anything visual that should be tabular but isn’t.
Experiment 2: Ask the data what it knows about you
Once I had the spreadsheet, I ran the original Sandbox prompt of the one that started this whole series:
Give me a summary of key insights and patterns from the data attached.
Also generate visual charts that visualize these key patterns and trends.
That’s the whole prompt. The interesting part is what you ask next.
After Claude gave me the standard summary of total shows, venues, song counts, I kept pushing. These are the four follow-ups that actually made the dataset come alive:
1. Plot shows per year as a bar chart. What story do the gap years tell?
2. Map every city I have seen a show in.
3. Which songs were always played at shows I attended, and which were rare?
4. What does this 17-year dataset say about me as a fan? Be honest.
The fourth one is the move I want to name out loud. It is not a data question. It is a self-portrait question dressed up as a data question. That kind of follow-up is what separates running a prompt from thinking with a model.
What came back was statistical. Peak years were 2009, 2010, and 2013 with four shows each. The 2002 to 2007 gap was real. Ten of twenty-five shows were at the same Indiana venue.
What came back was also honest enough to make me want to figure out more things to explore and see the data and my life in new perspectives.
You are not a Dave Matthews Band fan. You are a person who once arranged your summers around two-night runs in the Midwest. The fan was a phase. The arranging-your-summer-around-something part — that's the part worth asking about.
I don’t know if it’s true the way the model means it. But I know I’ve never quite said it about myself, and now I have a chart that sort of suggests it. I disagree that I am not a fan as I do listen to the live shows each Friday night at home and excited to see them live again. The beauty of AI is not that we have to agree with all the things, but how does the model interpretation of data cause new ways of thinking?
So I did pushback to explain the decision thinking that made the model say I am not a fan and after a long explanation this is essentially how it updated the statement above
Twenty-five shows in seventeen years, and seventy-two percent of them were the second night of a run. I'm not a fan who goes to a show. I'm a fan who goes on a trip.
Experiment 3: Reshape the same content into something shareable
Here’s where most people stop. They get the chart and the summary, screenshot it, and move on. The real unlock is asking the model to reshape the same content into different formats depending on what you actually need it for.
I ran two more prompts on the same dataset. Each one took less than thirty seconds.
Prompt A — A one-page handout I could text to my brother:
Build me a one-page printable handout that summarizes my
DMB concert history. Make it look like a clean editorial
infographic, not a corporate report. Include the top stats,
a year-by-year chart, the most-seen songs, and my top venues.
Generate it as an artifact I can download as a PDF.
Prompt B — A short presentation for the road trip to Alpine Valley:
Build me a five-slide presentation called “The Numbers”
based on this dataset. One slide per major theme:
total shows, peak years, top venues, longest jams,
and a closing slide with one honest observation about
what this data says about me as a fan. Make it visual,
sparse on text. Generate it as a PowerPoint I can download.
Same dataset. Three formats. The work was not in building each one. The work was in deciding what each one was for. That’s the actual skill. Not prompt-writing, not even AI literacy. It’s knowing what artifact fits which moment.
Try this with your work:
Take one piece of source material you’ve been sitting on like a long report, a research summary, a year-end data pull. Ask the model to give you the same content as a one-pager, a five-slide deck, and an email summary. Compare them. Notice which one you actually wanted to begin with.
Try this with your life:
Take a project of yours and ask for three versions of the same story. A printable card for the fridge. A short slide deck. A two-paragraph summary. The exercise teaches you something the AI can’t about what you actually wanted to communicate, separate from the tool that made it.
What to send me
If you try any of these of the screenshot-to-spreadsheet move, the data-as-mirror move, or the reshape-the-content move, send me back what you built and one sentence about why you made it.
I’ll feature the best ones in a future Sandbox issue. The series gets better when it includes what you tried.
See you next week.
— A-A-Ron