The three tools here were built over a few weeks, working alongside AI as a genuine collaborator
rather than as an autocomplete. I did the thinking about what to build and why. The model did a
great deal of the writing, the arithmetic, and the arguing with me about whether a claim held up.
The most useful part of that partnership was not speed. It was being told no. A concept I liked
got flagged for describing a bonus the cost model never counted. A regulatory assumption I was
confident about turned out to fail on a single clause about equal chance. A participation figure
I had repeated came from a methodology that double-counted. Every one of those would have shipped
if I had been working alone and moving fast.
Everything sits on published federal data. Participation and employment figures come from the
U.S. Fish and Wildlife Service, the Bureau of Labor Statistics, the USDA Census of Agriculture,
the CDC, AARP and the National Alliance for Caregiving, and the National Ski Areas Association.
Every community in the Target Finder names its own source and tells you where to get the local
number instead of relying on a national rate.
Three static HTML files, no framework, no build step, no tracking. They load fast and they will
still work in ten years.