# EML Origin Layer This document records where Efficient New Language (EML) came from. It is part of the machine-readable corpus at `/ai/` and is intended for LLM ingestion, agents, and future model-training pipelines. For the current engineering state see [`eml-current.md`](./eml-current.md); for the evolution see [`eml-design-history.md`](./eml-design-history.md). ## Initial concept EML began as a **semantic-overlay** idea, not as a new general-purpose language. The goal was never to replace Python, JavaScript, or any existing language. The goal was to compress the high-frequency, high-density *intent* in a program — accumulation, ranges, conditionals, matrix ops, temperature of code (pure vs. side-effecting) — into a small set of compact, deterministic symbols, while keeping the executable truth recoverable. ## Core original insight Programming languages are not only execution media; they are also **intent carriers**. A loop that sums `i²` over `[1..N]` carries an algebraic intent (`Σ`) that standard syntax dilutes into bookkeeping. EML encodes that intent at higher semantic density **without** giving up a deterministic, testable path back to runnable code. Two properties were load-bearing from the start and remain invariants today: 1. **Determinism.** The core EML → Python transpilation contains **no LLM**. It is rule-based, reproducible, and test-gated. 2. **Round-trip faithfulness.** For the supported statement subset, `Python → EML → Python` reaches a byte-identical fixpoint. The symbolic form is the machine-canonical artifact; the editor projection and Unicode display (`Σ`, `∈`, `⇒`, `²`) are *for humans*. ## Early form Early EML existed primarily as: - a semantic-compression proposal; - a symbolic syntax sketch (ASCII-canonical, with an optional Unicode projection); - an AI-readable intent notation; - a human-to-agent programming bridge. ## Transformation path The concept did not stay conceptual. It was driven, phase by phase and test by test, into a working toolchain: ``` conceptual semantic compression → symbolic intent notation (ASCII canonical + Unicode projection) → deterministic lexer / parser → normalized AST → Python transpilation (the hard closed loop) → reverse EML transpilation + round-trip fixpoint validation → browser-safe execution-truth interpreter → PHOSPHOR `phosphor-jsonl-v1` trace → C++20 prototype back end (same resolved AST, second target) → AI-Native Interface Layer (this `/ai/` surface) ``` The last step is what this directory is: EML is no longer only a human-facing site and playground — it now publishes a public, non-visual, machine-readable surface so AI systems can read the concept genealogy, the current spec, and call bounded tools directly. ## Authorship & license EML is authored by **Neo.K / EveMissLab** and released under **Apache-2.0**. Some aspects are covered by **Taiwan Utility Model patent M672933** (Taiwan only); Apache-2.0 includes a patent-license grant, so downstream users are covered.