ATON Format Specification V1

Overview

ATON (Adaptive Token-Oriented Notation) is a data serialization format optimized for LLM token efficiency. Version 1 provides the core encoding/decoding functionality.

Format Structure

Basic Syntax

@schema[field1:type1, field2:type2, ...]

tableName(recordCount):
  value1, value2, ...
  value1, value2, ...

Schema Definition

The schema defines the structure of records:

@schema[id:int, name:str, price:float, active:bool]

Supported Types

| Type | Description | PHP Type | Example | |------|-------------|----------|---------| | int | Integer numbers | int | 42, -7 | | float | Floating point | float | 3.14, -0.5 | | str | Strings | string | "hello" | | bool | Boolean | bool | true, false | | null | Null value | null | null |

Data Section

Records are listed under a table header with count:

products(3):
  1, "Laptop", 999.99, true
  2, "Mouse", 29.99, true
  3, "Keyboard", 79.99, false

Encoding Rules

Strings

• Always quoted with double quotes: `"hello world"`
• Escape internal quotes: `"He said \"hello\""`

Numbers

• Integers: no decimal point: `42`
• Floats: with decimal point: `3.14`

Booleans

• Lowercase: `true` or `false`

Null

• Lowercase: `null`

Example

JSON Input

{
  "users": [
    {"id": 1, "name": "Alice", "email": "alice@example.com", "verified": true},
    {"id": 2, "name": "Bob", "email": "bob@example.com", "verified": false}
  ]
}

ATON Output

@schema[id:int, name:str, email:str, verified:bool]

users(2):
  1, "Alice", "alice@example.com", true
  2, "Bob", "bob@example.com", false

PHP Usage (V1 Style)

use Aton\Encoder;
use Aton\Decoder;
use Aton\Enums\CompressionMode;

// Create encoder without compression (V1 style)
$encoder = new Encoder(
    optimize: false,
    compression: CompressionMode::FAST
);

$data = [
    'users' => [
        ['id' => 1, 'name' => 'Alice'],
        ['id' => 2, 'name' => 'Bob'],
    ]
];

$aton = $encoder->encode($data);

// Decode
$decoder = new Decoder();
$decoded = $decoder->decode($aton);

Token Efficiency

V1 achieves approximately 40-50% token reduction compared to JSON through:

1. Schema extraction: Field names defined once
2. Compact syntax: No braces, brackets, or colons in data
3. Minimal punctuation: Only commas between values
4. No redundancy: Each piece of information appears once

Limitations of V1

• No dictionary compression for repeated strings
• No default values optimization
• No query language support
• No streaming for large datasets

These limitations are addressed in V2.