# EML-U: language evolution through semantic composition

Snapshot: 2026-10-05. Status: bounded experimental MVP, not a completed research language.

## Purpose

EML-U is the EML family's operator-ontology and composite-semantic language. It attaches explicit semantics to existing program structures, resolves declared composition and conflicts, and projects supported contracts into existing hosts without silent semantic loss.

The design method concerns evolution of the programming language itself, not only better programs or more tools. It reuses host compilers and runtimes while introducing explicit semantic constructs, identities, effects and constraints. The research direction is for AI to learn and iterate this language-design method. Autonomous recursive language iteration is not demonstrated by the current MVP.

The host remains authoritative for its native syntax and semantics. Added meanings belong to explicit overlays. Changing a language contract is a versioned design decision, not permission to alter test expectations until a proposal passes.

## Current engineering scope

- Deterministic validation, operator/occurrence identity, anchors, graph normalization, declared composition, conflicts and decision traces.
- Four bounded execution profiles: scalar-u32, batch-u32, session-u32 and replay-u32.
- C++20, EML-P/Python and Rust 2021 projections. Rust checked and unchecked builds are both exercised.
- Source import: declared scalar C++ and EML-P forms only. General Rust source import is unsupported.
- Composition resolution is distinct from lowering; arbitrary resolved graphs are not automatically executable.
- Real host verification, literal goldens, missing-artifact controls, and installed local-package consumer observations.

At the dated snapshot: 297 tests across 27 files, four executable profiles, eight excluded pipeline starters and zero accepted production cases. The goal is 1,000 accepted distinct semantic cases before formal post-corpus project demonstrations. One manually requested round contains twenty new cases with primary focus 8 C++ / 8 Rust / 4 EML-P-Python. A semantic case counts once, not once per target.

These are dated engineering observations, not live counters. [The structured snapshot](./engineering-snapshot.json) is a summary, not a raw execution receipt or an authenticity signature. There is no public EML-U execution API on this page.

## Explicit limits

No general C/C++/Rust parser, floating-point profile, SIMD/parallel/GPU execution, foreign ABI, game-engine adapter, disk/process restore or finalized spatial/glyph notation is provided by this MVP. Performance is not inferred from source compression. Retained C++ batch measurements do not transfer to Rust, session, replay or arbitrary programs. No autonomous AI acceptance or universal correctness is claimed.

## Family and boundaries

EML-P is the practical profile within the EML-U theory. EML-NOVA is the peer AI-native tensor/operator language. CAIR, ICNS and MNVP keep independent contracts; an integration seam is not implemented support.

- [EML-U page](/eml-u/)
- [Preserved origins](/origins/)
- [EML-P](/eml-p/)
- [EML-NOVA](/nova/)
- [CAIR](/related/cair/)
- [ICNS](/related/icns/)
- [MNVP](/related/mnvp/)

## 繁體中文摘要

EML-U 研究與實作的對象，是程式語言本身的演化：不必從零重造語言與生態，也能透過明確的語意附加、算子與組合規則，在既有宿主上增加語言能力。讓 AI 學會並迭代這套語言設計方法，是研究方向；目前尚未證明自主遞迴的語言設計閉環。

有限 MVP 已涵蓋純量、批次、session 與 checkpoint/replay 四種 profile，以及 C++20、EML-P/Python、Rust 2021 的實際見證。這不表示支援一般宿主原始碼、任意圖的執行、最終空間符號語法，或全面加速。組合、降階、執行、效能、接受與發布，必須分別看待。

2026-10-05 快照：297 個測試／27 檔案；八個起始案例不計入正式語料，正式語料為零。下一階段每輪二十個新案例，主焦點 C++ 8、Rust 8、EML-P/Python 4；每個語意案例只計一次。千例里程碑之後，再進行正式的語料後專案示範。