Morse the message. Then Morse the Morse.
How little does one machine need to say for another machine to act correctly?
We routinely transmit JSON, XML, schemas, field names and repeated descriptions even when communicating systems already share context. Morse/36 explores a narrower question than general-purpose compression: how much representation can disappear when two machines share deterministic semantics?
The current Workbench explores an M36/3 profile with up to twelve 3-character words and an M36/2 profile with up to eighteen 2-character Base36 words. These are experimental packing profiles, not a frozen wire standard. The point is to expose the tradeoff between vocabulary size, composition and density.
Morse/36 is not claiming arbitrary information can magically fit into 36 characters. Shared vocabularies cost something. Registries cost something. High-entropy information still has to travel or be referenced. Try your own payload, inspect every morsing layer, and find where the abstraction breaks.
Word grammar, FTIP canonicalization, registry economics, collisions, context negotiation, failure behavior, security, conformance and reproducible comparisons with compressed JSON, CBOR, Protocol Buffers and other appropriate alternatives. Benchmark 001 exists to measure the idea rather than manufacture a win.
The Workbench currently visualizes experimental deterministic mappings. Its byte reduction excludes registry distribution, negotiation and transport overhead. Nothing here should yet be treated as a stable production wire standard.