The whole argument · five minutes
The rest of the site argues these at full length. This page is the spine — read it and you know what FrogNet claims, what has been shown, and where to push back.
Requests, replies, retries, serialization, discovery, failure handling — a large fraction of distributed application code exists because the programming surface forces communication to be expressed explicitly. On a FrogNet, wide-area distributed programs don't talk to each other. They share memory. REST isn't gone — it's underneath, where assembly went when programmers moved to C.
17 April 2026: Dan changed a value in New York. The value traveled through Seattle. A lamp three feet from him came on. His program didn't know Seattle existed, or the lamp — it wrote memory. Externally hosted; an external witness.

The network monitor does Nagios' job with no agent, no server, no check definitions — two HTTP calls and a name. No endpoint to design, no verb to choose, no status code to map. The last five bugs you chased in a distributed system were probably in code this model doesn't have.
A transcontinental HD call over an 850 Kbps link, degrading through 450 and recovering without redialing. On camera, the ladder walks from 1080 down to 160×120 — 164 Kbps, audio holding, no dropped frames — and climbs back. The wire carries diffs against structure both ends already hold: one measured megabyte of JSON moved as 92 KB.
One port must answer. Here is the host, here is the port: nmap came back with "1 service unrecognized despite returning data," and the scan is published as it ran, shared website ports and all. Not invulnerability — the real threat vectors are listed next to it.
Bounded stale state, no cross-value transactions, an allocator ceiling of ~56 nodes per tunnel with the arithmetic published, and a memory contract whose clauses are marked verified, implied, or undetermined. The six ways its own author got the model wrong in one day have their own page.
A simulator for arbitrary topologies, named oracles with defined failure conditions — take the broker, show the pivot — and seats for the people who want to argue with the spec rather than the marketing. A red run is a finding; a hypothetical is not.
We built the machine. Now prove us wrong.