TSLN: Time-Series Lean Notation

A Novel Data Serialization Format for Token-Efficient Analysis with Large Language Models

Abstract

The integration of Large Language Models (LLMs) with real-time time-series data presents a significant economic challenge: token-based pricing models make data transmission costs prohibitive at scale. We introduce TSLN (TimeSeries Lean Notation), a specialized data serialization format achieving 68-73% token reduction compared to JSON through innovative encoding strategies. TSLN employs schema-first architecture, relative timestamps, differential encoding, and adaptive repeat markers optimized for temporal patterns. Comprehensive benchmarks across cryptocurrency, stock market, and IoT datasets demonstrate consistent compression ratios of 70%+ while maintaining lossless fidelity. For applications processing 1 million records daily, TSLN enables annual savings exceeding $18,000 in LLM API costs. This work contributes a production-ready format, empirical validation methodology, and economic impact analysis, establishing a foundation for tokenefficient AI-powered time-series analytics.