CVSG · Chinese Variable-Spectrum Semantic Graph Language

Chinese is the surface. The graph is the program.

CVSG is a Chinese-first, rulebook-driven language. Chinese text is how a program is written; the canonical program is a typed semantic graph whose state lives on spectra, not only true and false, and a versioned rulebook decides how each sentence is read.

Research line · O1.0 closed · 281 tests passing

Chinese surface ≠ canonical program. Text is a view of the graph; the graph is canonical.

The idea

A program is a graph with a rulebook.

The canonical program is G = (V, E, X, R): semantic entities, events, states and operators; typed relations between them; typed variable-spectrum state; and the active rulebook. Chinese is a projection of that structure, and an authoring interface over it.

01

Truth is a spectrum

Predicates take values in [0, 1]; crisp true and false are the special case {0, 1}. The frozen logic profile uses AND = min, OR = max and NOT = 1 − p.

02

The rulebook is the contract

Every program carries a versioned rulebook. Under a different rulebook the same sentence can be read differently, and an ambiguous sentence is kept as explicit candidate graphs instead of failing at once.

03

Deterministic worlds

Rules run as world-state transitions over a cycle-start snapshot, with deterministic priority, traces, a linked history commitment, checkpoints and replay. O1.0 adds causes scheduled for later cycles.

04

AI-assisted, not AI-authoritative

AI may propose interpretations. Canonical state changes are explicit and versioned, and a committed state never changes silently because a parser, model or rulebook implementation changed.

Written in Chinese

Order and cause are written, not assumed.

Two short pieces from the language’s own examples. Time and causality are explicit relations in the graph: the order of the lines alone never becomes world time or a cause.

設甲為村民。
設乙為使者。
甲信任乙。
其懼度0.40,確信0.80,趨向0.10。
因(先(甲信任乙),後(狼近,甲懼增)),故(甲逃)。
Two scoped instances, a trust relation, three spectrum-valued states, and a causal clause whose order is explicit.
規則 cvsg.rule:fear-retreat-direction 優先80:若甲懼度至少0.60,則甲趨向減1.20。
規則 cvsg.rule:calm-fear-drop 優先60:若有安撫於甲,則甲懼度減0.25。
Rules written in Chinese compile to the canonical transition-rule model, and render back to normalized Chinese.
Where it stands

Origin-Zero is closed. So is O1.0.

Each stage closes only after local validation, and no later stage rewrites an earlier one.

Implemented

  • Origin-Zero (O0 to O0.9): composition rules, recursive programs, a deterministic rule-transition engine, Chinese rule authoring, spectrum predicates, quantified and graph-pattern rules, graph mutation and events, and a multi-cycle world runtime.
  • O1.0: an internal temporal calendar and scheduled causality, such as 「2週期後發出安撫於甲」 and 「第4週期發出警示於丙」.
  • Checkpoints and resumption that match uninterrupted runs, under a linked history commitment.

Validated

  • O1.0, closed on 2026-10-04: 281 of 281 tests and all eleven stage demos pass (Windows, Python 3.14.5).

Boundaries

  • Semantic domains may be unbounded; every runtime profile is finite.
  • A research line: no public release, package or license yet.
In the family

Five public lines

  1. EML-U

    Language evolution

  2. EML-P

    Practical

  3. EML-NOVA

    Structure-first

  4. NoGlyph

    Constraint-first

  5. CVSG

    Chinese-first

Back to the family
ΣEML 2026

Efficient New Language — a high-density semantic overlay for humans and AI agents.

This site’s demos run on the real EML toolchain, in your browser.

© 2026 EveMissLab (一言諾科技有限公司) / Neo.K. All rights reserved.