Launching today
RLX Backtester
Fast Rust backtesting engine for quantitative traders
4 followers
Fast Rust backtesting engine for quantitative traders
4 followers
RLX Backtester is a native macOS app built for AI-powered quantitative research. Connect Claude, Cursor, or any MCP-compatible agent to design, backtest, optimize, and stress-test trading strategies locally. Unlike traditional backtesting tools, RLX combines a high-performance Rust engine, reinforcement learning workflows, and an AI-native interface that lets autonomous agents iterate on strategies in real time. Free tier included, with Pro for unlimited research.
How does the RLX engine handle walk-forward validation versus simple in-sample fitting, especially when the reinforcement learning loop starts overfitting to historical regimes?
@gkegjn9
"Great question. The RLX engine handles this through a combination of strict walk-forward cross-validation (WFCV) and deterministic policy freezing:
### 1. Walk-Forward Validation vs. In-Sample Fitting
• Sliding/Anchored Windows: Instead of a single train/test split, the engine runs walk-forward analysis ( walk_forward tool) across multiple rolling windows. The model
(whether rules-based or a neural policy) is optimized/trained strictly on an In-Sample (IS) window and then executed on an adjacent, completely unseen Out-of-Sample
(OOS) window.
• Walk-Forward Efficiency (WFE): We evaluate robustness using the WFE metric:
Average OOS Return
WFE = ──────────────────
Average IS Return
If WFE is close to or above 1.0, the edge is robust. If it decays significantly (e.g., below 0.5), the system flags the configuration as overfitted to local
volatility/trend regimes.
### 2. Preventing Reinforcement Learning Regime Overfitting
When using the RL engine (e.g., DQN or Actor-Critic):
• Deterministic Policy Evaluation: During training, the RL agent uses stochastic actions (via ε-greedy or entropy-regularized policy distributions) to explore. However,
during validation and walk-forward OOS testing, the policy is frozen and evaluated deterministically using the argmax of policy logits. All exploration noise is
disabled.
• Non-Stationary State Space Normalization: To prevent the neural network from memorizing nominal price levels (which leads to regime overfitting), the RLX engine
processes inputs through statistical normalization. Features are converted into stationarity-adjusted metrics such as return z-scores ( ret_z_24 ), drift scores, and
Shannon price entropy. This forces the model to generalize patterns of market structure rather than specific historical price zones.
• Benchmark Comparison: The engine benchmarks OOS performance directly against passive Buy & Hold and randomized baseline agents to ensure that the RL model is capturing
actual alpha rather than riding beta or overfitting to a specific market direction."
Love the MCP integration angle, this feels like the right direction for quant tools right now. One thing I'd really want to see is a built-in walk-forward analysis mode that splits data into rolling in-sample and out-of-sample windows automatically, instead of just a single train/test split. Would make the RL workflow feel a lot more trustworthy out of the box.
@dnetzemengh7s
"Thanks for the feedback! I completely agree that automated rolling walk-forward analysis is critical for making RL workflows trustworthy.
We actually already have this built directly into the RLXBT engine. The platform includes a native walk_forward execution mode that automatically splits the loaded
dataset into rolling (or anchored) in-sample (IS) and out-of-sample (OOS) windows, calculating Walk-Forward Efficiency (WFE) across multiple testing windows.
However, the absolute game-changer for me—and my favorite feature of the platform—is the Idea Map (Hypothesis Canvas). Here is how it solves the exact research and trust
issues you mentioned:
### 1. Visualizing Strategy Lineage (No More Loose CSVs)
Instead of looking at isolated backtest reports, the Idea Map acts as a visual git-tree for our quant research. It tracks the parent-child mutations of a strategy. For
example, if we start with a basic breakout model (Parent) and add a Shannon Entropy exit filter to avoid fakeouts (Child), the map visually connects them.
### 2. Anchoring Robustness to the Hypothesis
Every time we run an automated walk-forward cross-validation or a Monte Carlo bootstrap simulation, the engine binds the results (WFE, risk of ruin, out-of-sample window
performance) directly to that hypothesis card as unalterable Evidence.
### 3. Curing Selection Bias
This evolutionary tracking prevents the classic quant trap of 'selection bias' (where a researcher runs 50 backtests and only shows the luckiest one). The Canvas forces
you to look at the WFE decay across iterations, making the entire RL development loop auditable and trustworthy out of the box.
When connected to an AI agent via the Model Context Protocol (MCP), the agent reads this Idea Map context instantly. It knows exactly which parameters previously failed
out-of-sample and mutates the rules systematically rather than repeating past mistakes.
It completely changes the research workflow from chaotic parameter-sweeping to structured, evolutionary science!"