Downstream · draft, not for sending as-is

For the systems-design read

Hey [name],

Lawmaking is the one high-stakes system in the country that still runs open-loop, and I'm building the instrumentation for it.

The unit of analysis is the provision, not the bill. A thousand-page act is a bundle of orders with deadlines, money that may or may not move, and tripwires set years out. "Did the rule this clause ordered appear within 180 days" is checkable against thirty years of public records. "Is this bill good" is not, and we don't touch it.
Consequence enters the record from any branch, at any point on the timeline, and the loops run backward too. A veto returns a bill, a vacatur returns a rule to the agency, an oversight finding returns to Congress. So the map puts the House floor schedule, the Federal Register, and presidential directives on one timeline and watches all three doors. Courts appear as outcomes for now; their own corpus is bigger than the other two combined, so it is deferred on purpose rather than half-done.
The official record turned out to be optional, and we measured it. whitehouse.gov published 30 presidential memoranda in 2026. As of our own captures on August 19, 2026, 21 had printed in the Federal Register, 7 were confirmed absent, and 2 were too recent to call. The Register prints a memorandum only when the President orders it printed, and memoranda carry no number, so nothing in a sequence ever looks missing.

A memorandum carries essentially the same legal force as an executive order, and it can set rule-writing and money in motion with no statute anywhere upstream. That is why the archive runs first and the forecasting comes later. A forecast only means something against a record someone else can check, and today no institution keeps that record. What exists now is the record half: a daily archive sealed with a hash chain, a Federal Register digest every business day, the presidential corpus, a House floor radar with sponsorship joins, and an MCP server so other people's machines read the same record people do. Forecasting does not exist yet, and every consequence claim we've drafted is still held back as unverified. That is the design, not a backlog.

This is a systems pitch, so break it like one. The work-in-progress deck is at https://seedownstream.com/deck/. Tell me where the map of the machine is wrong and which loop I've missed. And if the machine grabs you, let's talk.

Hey [name],

Lawmaking is the one high-stakes system in the country that still runs open-loop, and I'm building the instrumentation for it.

- The unit of analysis is the provision, not the bill. A thousand-page act is a bundle of orders with deadlines, money that may or may not move, and tripwires set years out. "Did the rule this clause ordered appear within 180 days" is checkable against thirty years of public records. "Is this bill good" is not, and we don't touch it.
- Consequence enters the record from any branch, at any point on the timeline, and the loops run backward too. A veto returns a bill, a vacatur returns a rule to the agency, an oversight finding returns to Congress. So the map puts the House floor schedule, the Federal Register, and presidential directives on one timeline and watches all three doors. Courts appear as outcomes for now; their own corpus is bigger than the other two combined, so it is deferred on purpose rather than half-done.
- The official record turned out to be optional, and we measured it. whitehouse.gov published 30 presidential memoranda in 2026. As of our own captures on August 19, 2026, 21 had printed in the Federal Register, 7 were confirmed absent, and 2 were too recent to call. The Register prints a memorandum only when the President orders it printed, and memoranda carry no number, so nothing in a sequence ever looks missing.