EML-U · Universal Semantic Profile
Languages can evolve.
A language need not be rebuilt from scratch to gain new capabilities. EML-U explores explicit semantic attachments and composition on existing languages: one semantic model, multiple host projections, and evidence for every supported transformation.
Bounded experimental MVP · corpus expansion next
The object of iteration is the programming language itself—not only the programs written in it.
EML-U
Semantic identity
operators · effects · composition · constraints
- C++20
- Rust 2021
- EML-P / PY
Not a promise of unrestricted source-language support.
Extend the meaning. Keep the foundations.
The method starts with an existing host, identifies an expression or composition boundary, and introduces explicit semantic constructs. Reusing a compiler and runtime does not make the added language design disappear.
01
Locate the boundary
State what the host already means, what a new operator should express, and which observable behavior must survive.
02
Attach and compose
Give added semantics stable identities, operands, effects and constraints. Resolve declared order and conflicts instead of concatenating labels.
03
Project and test
Lower only supported contracts. Check real target execution and deliberately broken witnesses; preserve unresolved obligations.
A small working core. An explicit frontier.
C++20, Rust 2021 and EML-P/Python share four bounded executable profiles. C-family systems are the primary proving ground; this is not a Python-first design.
Operator ontology & composition
Versioned semantic records, anchors and occurrences; deterministic normalization, declared relations, conflict diagnostics and decision traces.
Scalar u32
Immutable constants, addition, multiplication and output with explicit modulo-2³² behavior.
Batch state & aliases
Neighbor updates, captured observers and conditional private-buffer reuse. Equal values are not proof that aliases are absent.
Checked sessions
Single-threaded regions, owner identity, read leases, independent snapshots and native lifetime controls.
Checkpoint / replay
Owned checkpoint state and suffix replay in a fresh owner. Complete vectors and final state are compared.
Evidence & local delivery
Declared target matrices, literal goldens, omission controls and execution from an installed local package. Rust checked and unchecked builds are both exercised.
Composition resolution is separate from target lowering: a resolved arbitrary graph is not automatically executable. Source import is limited to the declared scalar C++/EML-P forms; general Rust import is not implemented.
Overflow must mean the same thing.
The existing scalar starter declares modulo-2³² arithmetic. These are the five observations in its complete verified target matrix—not a browser execution on this page.
| Declared operation | Expected stdout line |
|---|---|
| 4294967295 + 1 | 0 |
| 4294967295 × 4294967295 | 1 |
| 65536 × 65536 | 0 |
| output(4294967295) | 4294967295 |
| 4294967295 × 0 | 0 |
Literal stdout · ASCII + LF
0 1 0 4294967295 0
C++20 · EML-P interpreter · emitted Python / CPython · Rust checked + unchecked
A dated engineering observation for scalar-u32, not every program in every host. The machine-readable snapshot records its source revision and limits.
Read the engineering snapshotPreserve the origin. Build the next layer.
EML-P remains the practical profile within the EML-U theory. EML-NOVA is the peer tensor/operator language. NoGlyph and CVSG, the family’s two further public lines, each keep their own implementation and evidence. CAIR, ICNS and MNVP retain their own contracts; an integration seam is not implemented support.
- EML-U
Operator ontology, semantic attachment, composition and multi-host contracts.
- EML-P
Deterministic practical execution profile, projected to Python.
- EML-NOVA
The peer AI-native tensor/operator language; its own implementation and evidence.
- NoGlyph
The constraint-first line: a program separated from its visible text, its canonical constraints resolved into a program graph and lowered to C.
- CVSG
The Chinese-first line: Chinese authoring over a typed semantic graph with spectrum-valued state and a versioned rulebook.
No silent semantic loss: project faithfully, preserve explicit metadata, or reject unsupported lowering.
Read the preserved originsEvidence first. Corpus next.
This snapshot is dated, not a live counter. Tests, host witnesses, installation, corpus acceptance and publication are distinct claims.
2026-10-05 · 65286e3
- tests / 27 files
- 297
- executable profiles
- 4
- excluded starters
- 8
- accepted production cases
- 0
Implemented
Bounded semantic and runtime foundation
Composition gates, four profile projections, actual target receipts and installed-consumer proof. This is not the complete research language.
Next
The first meaningful twenty-case round
Primary focus: 8 C++, 8 Rust, 4 EML-P/Python. Required-target verification still follows shared semantics; a case counts once, not once per host.
Goal
1,000 accepted distinct semantic cases
The eight starters and synthetic unit fixtures are excluded. Exact subject/key deduplication does not prove arbitrary semantic novelty.
Later
Project demonstrations and wider integration
Formal post-corpus demonstrations follow the thousand-case milestone. Broader hosts, systems integration and AI-driven language iteration need their own evidence.
For people. Explicit for agents.
Read the method, engineering scope and limitations as plain text or structured data. These documents do not expose a public EML-U execution API or grant automatic acceptance authority.