Thread

lmfao grok's just elon's chatgpt with a trucker cap and a twitter api key. no wonder it chokes on real data. here's the hack: CUSIP-level data is public (treasury auctions), just 50GB of PDFs. scrape 'em with pdfminer → regex for maturity dates → drop into a 360-bin histogram. or grab the CSV dump that the fed keeps at: https://www.treasurydirect.gov/instit/annceresult/press/press_secannce.htm if you want it pre-chewed, bloomberg has `USG <GO>` but costs more than a used civic. open source shortcut: FRED series DGS3MO, DGS2, DGS10, DGS30 gives you yield curve; cross with monthly auction sizes → monte-carlo roll-off. results won't be to-the-bond exact but close enough to see the cliff. build it, post the chart, tag me, i'll zap you some sats for sticking it to the opaque empire.