Downstream · draft, not for sending as-is
Hey [name],
The Supreme Court ended Chevron deference in 2024, which made regulatory survival a genuinely probabilistic question, and two years later I still can't find anyone pricing it.
The bear case first, because you'd raise it anyway. Every input is public, and a frontier model pointed at public data does maybe 70% of this. The bet is that the other 30% is where the value sits: point-in-time discipline so a backtest can't read tomorrow's newspaper, joins verified by hand before anything ships, base rates fit per mechanism type on thirty years of records, and a public track record that includes the misses, kept before anyone was watching. And to be plain about stage: pre-launch, pre-revenue, no customers. What runs today is the record layer, a daily hash-sealed archive, a Federal Register digest, a presidential corpus, a House floor radar, and a free MCP endpoint. Forecasting does not exist yet, deliberately, because a forecast without a checkable record behind it is content.
The working deck is at https://seedownstream.com/deck/. Underwrite it the way you'd underwrite anything: tell me which number you don't believe, which comp doesn't hold, where the base rates are thinner than I think. If it still looks like a company after that, let's talk.
Hey [name], The Supreme Court ended Chevron deference in 2024, which made regulatory survival a genuinely probabilistic question, and two years later I still can't find anyone pricing it. - Statutory time is full of dated, tradable moments. Comment periods close, rules take effect, sunsets fire. One housing provision alone sets a January 7, 2027 notification deadline with penalties up to $1M per violation. Finance prices exactly one species of these dates, the FDA approval calendar. I surveyed the data vendors in early August 2026 and nobody sells the government's calendar; "regulatory calendar" in the market today means your compliance deadlines, in a law-firm PDF. - The right comps are reference layers, not policy software. In credit, in catastrophe risk, in indices, someone built the record, published the methodology, and the scorekeeper ended up worth an order of magnitude more than the tools sold beside it. This year priced the split cleanly: the best-funded policy-software company of the last decade was delisted, a standalone archive of legislation over time sold for about six million dollars, and proprietary benchmark datasets are commanding premium multiples. The record alone is nearly worthless. The record plus the joins, the scores, and the track record is a reference institution. - Commerce funds the free tier. Funds and corporate policy teams pay for the early read, because the lag between passage and consequence is worth money to anyone with revenue on the line. The map stays free for congressional staff, journalists, and researchers, who need it most and will never have budget. We say that out loud instead of hiding it. The bear case first, because you'd raise it anyway. Every input is public, and a frontier model pointed at public data does maybe 70% of this. The bet is that the other 30% is where the value sits: point-in-time discipline so a backtest can't read tomorrow's newspaper, joins verified by hand before anything ships, base rates fit per mechanism type on thirty years of records, and a public track record that includes the misses, kept before anyone was watching. And to be plain about stage: pre-launch, pre-revenue, no customers. What runs today is the record layer, a daily hash-sealed archive, a Federal Register digest, a presidential corpus, a House floor radar, and a free MCP endpoint. Forecasting does not exist yet, deliberately, because a forecast without a checkable record behind it is content. The working deck is at https://seedownstream.com/deck/. Underwrite it the way you'd underwrite anything: tell me which number you don't believe, which comp doesn't hold, where the base rates are thinner than I think. If it still looks like a company after that, let's talk.