Two pieces, built together: NYCH, a symbolic language that compresses natural language into deterministic emoji anchors; and MG8, a bounded execution/container family that runs, states, gates and traces whatever NYCH hands it. Both are early, working, and open.
NYCH maps words onto a small, versioned table of fixed semantic-anchor emoji. Anything not yet in the table falls back to a mechanical consonant skeleton — the word with its vowels and doubled letters dropped — so meaning is never guessed at, only either matched or openly left uncompressed. Names, drug names, and scientific terms are explicitly passed through untouched; NYCH never symbolizes a protected term.
A reproducible internal run took a 37-character instruction through the full expansion pipeline into roughly 4,200 characters of unpacked context for the receiving model. That ratio is real and rerunnable. What it's actually worth to an LLM's answer quality — compared to just sending the plain instruction — has not been independently measured. That comparison is the first thing this needs from people who aren't the one who built it.
Click a word.
Each identity stays separate on purpose — state, gate, position, and trace never collapse into one blob. A `.mg8pk` package can bundle several `.mg8` units the same way.
Every .g8son gate operates over the same
primitive: a state representation SR = (s, D, C)
— a state s, the domain
D it's read in, and the constraints
C active on it. Chain
.mg8.g8son + g8son + g8son … across one run
and, formally, that chain is a trajectory — a sequence of transforms, each one carrying a state
representation to the next.
Trajectory-Constrained Transform Algebra (TCTA — a separate, proprietary framework of the author's) defines an admissible space: the set of trajectories a given set of constraints actually permits. The overarching intent for MG8 is to map a gate chain against that admissible space, so a run is constrained toward a declared goal or target transform — not just whatever sequence of gates happened to fire. That coupling is the architectural target, not working code yet — same honesty bar as everything else on this page: it's the direction, not a claim of what's built.
This doesn't depend on NYCH. NYCH's emoji/skeleton layer is one choice of representation for the
s in
SR = (s, D, C) — a
compressed one. Swap it out and the same gate-chain-as-trajectory structure still holds, across
whatever application runs it. Concretely: MG8 works standalone, with no NYCH symbolization at all,
as deterministic natural-language gates in front of an LLM — plain-language conditions an output
has to satisfy to pass, explicit bounds, and a fail-closed UNRESOLVED
exit instead of a silent infinite retry. NYCH is one thing you can run through that pipeline, not a
requirement of it.
Everything above is a working repo you can clone and run today, not a pitch deck. The most useful thing anyone can do with it right now:
nych today — check it stays that way before and after your change.NYCH and MG8 are developed alongside paying work, not instead of it. If this is useful to you, sponsorship is what turns nights-and-weekends progress into something faster.