RL Alpha Labs — Quantitative Research & Field Notes Survivorship-free Indian equity (NSE/BSE) simulation, statutory & FIFO tax modeling, and pure-JAX XLA reinforcement learning research. https://rlalphalabs.com/ 2026-10-15T15:30:00Z https://rlalphalabs.com/og.png RL Alpha Labs connect@rlalphalabs.com https://rlalphalabs.com/ Indian Equity Trading Costs, CDSL ₹15.34 DP Charges & FIFO Capital-Gains Tax Hurdle https://rlalphalabs.com/notes/trading-costs/ 2026-10-15T12:00:00Z Why retail NSE equity delivery portfolios face a steep nonlinear hurdle from STT, stamp duty, GST, flat ₹15.34 CDSL depository sell charges, and mandatory FIFO s.111A/s.112A capital-gains realizations. The 8 Backtest Traps That Inflate Indian Equity Returns (+0.3%/yr to +29.6%/yr) https://rlalphalabs.com/notes/backtest-traps/ 2026-10-15T12:00:00Z Forensic audit of 8 common Indian market backtest errors: survivorship bias from dropped delisted NSE symbols, same-bar close fills, quarter-end vs. broadcast timestamp leakage, and unadjusted corporate actions. Twelve Trading Ideas We Tested on NSE and Closed (And Why) https://rlalphalabs.com/notes/ideas-we-closed/ 2026-10-15T12:00:00Z Empirical post-mortems across 12 closed Indian equity hypotheses—from high-turnover continuous RL and LLM corporate filing readers to equal-weight 300-stock retail baskets. How We Run AI Research and Coding Agents Under Frozen Pre-Registration https://rlalphalabs.com/notes/how-we-run-ai-agents/ 2026-10-15T12:00:00Z Lab rules for operating multi-model AI research and engineering agents with SHA-256 frozen experimental designs, AST lookahead gates, and human sign-off at every promotion stage. Problem Formulation & Indian Market (NSE/BSE) Microstructure https://rlalphalabs.com/research/problem-statement.html 2026-10-15T12:00:00Z Formal POMDP formulation, 2015–2025 Indian equity regime breakdown, NIFTY 50/100/500 liquidity and bid-ask spread tables, and statutory CNC vs. MIS cost equations. Three-Plane Architecture & Anti-Overfit Research Methodology https://rlalphalabs.com/research/methodology.html 2026-10-15T12:00:00Z Point-in-time Data Plane (8.77M security-session rows), Pure-JAX XLA Model Plane, and One-World Simulation Plane pinned to 1e-9 precision across 3,865 pre-registered trials. Pure-JAX XLA Scaled Training & GPU Throughput Benchmarks https://rlalphalabs.com/experiments/scaled-training.html 2026-10-15T12:00:00Z Hardware benchmarks across vectorized environments (500,000 env-steps/sec), fused jax.lax.scan loops, Patch-GTrXL sector-graph transformers, and 51× replay buffer HBM compression.