Data Explorer. Pup
Ask the public /data files a question in SQL, in your own browser, and turn the answer into a chart that says where it came from. Every result cites the file and its as_of. Nothing you type leaves this page.
Pure query, no forecasts. Observed, modelled, market and analyst numbers stay labelled as what they are. Not betting advice.
The SQL engine could not load
DuckDB-Wasm did not start in this browser (). The files themselves are still public, dated and readable: open them directly, preview the rows below, or read the example chart.
SQL
Result
Chart this
Method — what this page does and does not calculate
This page loads a curated set of public files from /data/ into DuckDB-Wasm running in your browser tab. Each table is the row array already published inside one file. The only reshaping: a season-keyed array gets a season column (the key is the published value), nested fields become parent_child columns, and lists are joined with commas. Nothing is imputed: null stays null and is never treated as zero. Two fields are not loaded: pool.silva (a source analyst's rank string) and cfb_market.books (the per-book list).
Every result and chart names the file(s) it read and each file's as_of. The kind label — observed, modelled descriptive, model forecast, market, model + market backtest, analyst estimate — comes from the file, not from the query. A query can mix kinds; the citation shows each one, so read them before comparing.
No model runs here. There are no new forecasts, no blends, no hidden calls. Future outcomes in these files are model probabilities and should be read as ranges with their stated assumptions, not as point predictions. Not available in the public data: an independent closing-market line per 2026 NFL game, so "nfelo vs the closing market" cannot be answered here (see the starter list). There is no ADP column in pool.
The SQL engine is DuckDB-Wasm 1.32.0, served from this site (/assets/duckdb-wasm/, MIT licence alongside): no third-party CDN. The engine is about 34 MB uncompressed on a first visit and is cached by your browser after that. If it cannot load, the page falls back to plain file links, row previews and a static example chart. Share links keep the query in the URL fragment, which browsers never send to a server.