Release v1.0.4 · ZON-Format/ZON

What's Changed

Full Changelog: v1.0.3...v1.0.4

Colon-less Syntax: Objects and arrays in nested positions now use key{...} and key[...] syntax, removing redundant colons.
Smart Flattening: Top-level nested objects are automatically flattened to dot notation (e.g., config.db{...}).
Control Character Escaping: All control characters (ASCII 0-31) are now properly escaped to prevent binary file creation.
Runtime Schema Validation: New zon builder and validate() function for LLM guardrails.
Algorithmic Benchmark Generation: Replaced LLM-based question generation with deterministic algorithm for consistent benchmarks.
Expanded Dataset: Added "products" and "feed" data to unified dataset for real-world e-commerce scenarios.
Tricky Questions: Introduced edge cases (non-existent fields, logic traps, case sensitivity) to stress-test LLM reasoning.
Robust Benchmark Runner: Added exponential backoff and rate limiting to handle Azure OpenAI S0 tier constraints.

Benchmark Formats: Refined tested formats to ZON, TOON, JSON, JSON (Minified), and CSV for focused analysis.
Documentation: Updated README and API references with the latest benchmark results (GPT-5 Nano) and accurate token counts.
Token Efficiency: Recalculated efficiency scores based on the expanded dataset, confirming ZON's leadership (1430.6 score).

Token Efficiency: Achieved up to 23.8% reduction vs JSON (GPT-4o) thanks to syntax optimizations.
Readability: Cleaner, block-like structure for nested data.

Critical Data Integrity: Fixed roundtrip failures for strings containing newlines, empty strings, and escaped characters.
Decoder Logic: Fixed _split_by_delimiter to correctly handle nested arrays and objects within table cells (e.g., [10, 20]).
Encoder Logic: Added mandatory quoting for empty strings and strings with newlines to prevent data loss.
Rate Limiting: Resolved 429 errors during benchmarking with robust retry logic.