Pigskin Lab

Pigskin Lab started as a question I kept wanting to ask about football and could not get answered without a spreadsheet. It became a full stats site: every NFL play since 1999, every team, player and game, and a dashboard maker that builds its answer from a plain sentence.

You type what you want to see — "Brock Purdy week 16 2025", "49ers 2022 to 2025", "Super Bowl 2025" — and it builds a board of stat tiles, graphs and tables for exactly that. Then you keep talking to it: "compare with Bo Nix", "only leave the charts", "now week 2". The board changes each time instead of starting over — a follow-up like "4th quarter" narrows every chart on it at once — and every name and number is checked against the database before anything is drawn, so it cannot make up a player or a statistic. In edit mode boards can be dragged, resized and themed, widgets copied and pasted, and the result shared with a short link or exported as a picture.

Reading a sentence is done entirely by rules I wrote, not by an AI model. I tried the model route and measured it: in a 506-request test the rules built nearly every board themselves, and the model rescued none of the few they missed while costing server memory and money every month. So the model came out. When the site cannot read a request it says so, with an example that works, and every miss becomes a new rule.

Around the dashboard maker is the rest of a real stats site: sortable reference tables, a page for every team and player, a page for every game that opens with its story — who decided it, the turning points, the difference — over win-probability, scoring and drive charts, and a Models page where each prediction is fitted only on seasons before the one it scores — so what you see is what the model would really have said at the time, scored against the betting line. It is free, with no ads and no paywall.