gex-llm-patterns: test whether LLMs reason about options gamma exposure without dates or tickers

A research repository exploring whether large language models can identify structural constraints in options markets by reasoning rather than memorizing training data. The core method — temporal obfuscation — removes dates, ticker symbols, and context from option chains so models must infer dealer positioning from strike-level numerical structure alone. The application domain is gamma exposure (GEX): the aggregate hedging burden that options dealers carry from their inventory positions. Who this is for Researchers and engineers building LLM-based financial reasoning systems, especially those working on market-structure inference or options-risk modeling. The workflow targets people who want to test whether their models generalize to stripped financial inputs rather than relying on memorized labels. Evidence of readiness The methodology has been peer-reviewed across three venues: IEEE BigData 2025 (single-day detection, arXiv 2512.17923), AIAI 2026 (multi-day regime detection, DOI 10.1007/978-3-032-30801-6_10), and JRFM 2026 (combined methodology). LaTeX source and prompt templates live under docs/papers/. The dataset is SPY options from 2024 with Alpha Vantage Premium feeds. What may need