xron-format

Extensible Reduced Object Notation — a lossless, drop-in replacement for JSON.stringify/JSON.parse that cuts LLM token usage by removing the structural overhead JSON pays on every request: repeated keys, repeated values, and punctuation a tokeniser has to spend tokens on.

XRON.parse(XRON.stringify(data)) deep-equals data. Nothing is dropped, summarised, or approximated — it's an algorithmic encoder, not an LLM.

Install

npm install xron-format

Usage

import { XRON } from 'xron-format';

const data = [
  { id: 1, name: 'Alice', dept: 'Sales' },
  { id: 2, name: 'Bob', dept: 'Engineering' },
  { id: 3, name: 'Carol', dept: 'Sales' },
];

const xron = XRON.stringify(data);
// @v2
// @S A: id, name, dept
// @D: Sales
// @N3 A
// 1, Alice, $0
// 2, Bob, Engineering
// 3, Carol, $0

const restored = XRON.parse(xron);
// restored deep-equals data

How much smaller is it?

Depends on how much of your payload is structure versus information — XRON removes repeated keys and repeated values, so it can't shrink data that's already unique. Measured with the real o200k_base tokenizer:

| Your data looks like | Token reduction | |---|---:| | Identical repeated rows, long string values | up to 91% | | Wide tables, small vocabulary (30 cols) | ~80% | | Boolean/flag matrices (20 cols) | ~71% | | Typical business records (6 fields) | ~48% | | Unique UUIDs or unique prose | ~21% |

Full methodology and per-shape figures: BENCHMARKS.md.

API

XRON.stringify(value, options?): string

| Option | Type | Default | Description | |---|---|---|---| | level | 1 \| 2 \| 3 \| 'auto' | 2 | Compression level. 'auto' picks whichever level produces the smallest output. | | minCompressSize | number | 0 | Below this JSON byte size, 'auto' returns plain JSON rather than attempting compression. | | tokenizer | 'o200k_base' \| 'cl100k_base' \| 'claude' | 'o200k_base' | BPE tokeniser profile used by Level 3's optimisation pass. |

Guarantees in auto mode: output always round-trips losslessly, and is never larger than JSON.stringify(value).

Throws TypeError on circular references.

XRON.parse(input): any

Parses an XRON string back to the original value.

Verification

Round-trip identity (parse(stringify(x)) === x) is tested against a boundary-focused fuzzing corpus across all four levels, plus three rounds of independently-written adversarial fuzzers. Full bug list, test counts, and methodology: docs/VERIFICATION.md.

License

MIT — see LICENSE.