Exploration Graph.

Thirty years in, still scratching the surface.

I've been exploring the San Rafael Swell for thirty years. Long enough to learn the one rule that actually works: pick one new place each trip. Anything more is a gamble — a long drive that doesn't pan out, a day you don't get back.

The guidebooks don't help you choose. Each one is good on its own — none of them talk to each other. And after thirty years, I've learned the harder thing: I haven't even scratched the surface. There's far more to explore than any single book can point me to.

Every labelled point is a place this case study names. Behind them, roughly 4,900 more known places fill the same county — and this is one county.

The answer was never in one book.

Years of research paid off — just not the way I expected. I'd read across every genre I could find: geology books, ghost-town histories, rockhounding guides, off-road atlases, old outlaw stories. Each one knew something the others didn't. The places worth the trip were always where the books overlapped — a wash with an old mine around the bend, outlaw history in the canyon above it, and agate on the ground if you knew to look.

For a couple of years I kept circling the same idea: the combination is the thing no book can give you, but a computer could. AI finally made it possible — stitch the open data together, let it find the overlaps for me. I wouldn't have started without the momentum from Open Brain — once I'd built one system that captured an idea and ran with it, this one stopped feeling out of reach. I captured the idea, then went looking for the data.

Where the recommendations come from.

Everything in the graph comes from public, authoritative data — the kind of sources geologists and land managers actually use. No scraped blogs, no guesswork. If the system says a canyon has a uranium mine and Morrison Formation outcrops, it's because the USGS and the Utah Geological Survey say so.

For a single Utah county, that's roughly 5,000 places, 3,300 active mining claims, and a century of newspaper text — and the same model runs anywhere in the West.

The value is in the combination.

A pile of datasets isn't a discovery engine. The value isn't in any one layer — it's in how they combine. Each source becomes a layer the graph can reason across: what a place means, where it sits, what it's mentioned alongside, what was written about it a hundred years ago, and who manages the ground. Join any two, and you surface something a single search would miss.

Three ways to ask.

The layers pay off as three queries — and every answer is grounded in real data, not a guess.

Designed to withhold.

I built it for myself first.

It runs on my phone, out in the field — I can ask it what's worth the detour, and it answers from the same data I spent years reading by hand.