If a 12M-token context window actually worked with full fidelity, it would be a money-printing machine.Here's a conservative example using only publicly available data.Take every company in the S&P 500.For each one:> Pull the latest earnings call transcript. These calls are public. A typical earnings call will be 45-60 mins with prepared remarks and analyst Q&A, producing a transcript of roughly 8k-12k tokens. For all 500 companies, that's about 5M tokens.> Now add a year of daily price and volume data for ...