FLOX C++ trading systems framework with MCP

v0.6.4 Feature

This release adds 4 notable features for engineering teams evaluating rollout.

Published 8d MCP Developer Tools

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

algorithmic-trading algotrading backtesting codon c++ javascript

+3 more

python quant trading

Summary

AI summary

VenueStack now drives env, backtest, paper, and live trading uniformly.

Changes in this release

Type	Severity	Summary	CVE
Feature
Feature	Medium	Unified VenueStack drives env, backtest, paper, and live trading workflows. Unified VenueStack drives env, backtest, paper, and live trading workflows. Source: llm_adapter@2026-05-26 Confidence: high	—
Feature	Medium	Action surface now supports signed-quantity limit orders with TIF. Action surface now supports signed-quantity limit orders with TIF. Source: llm_adapter@2026-05-26 Confidence: high	—
Feature	Medium	Observation surface includes policy's open orders and penalizes rejected fills. Observation surface includes policy's open orders and penalizes rejected fills. Source: llm_adapter@2026-05-26 Confidence: high	—
Feature	Medium	Portfolio-mode multi-symbol observation and action spaces are now available. Portfolio-mode multi-symbol observation and action spaces are now available. Source: llm_adapter@2026-05-26 Confidence: high	—
Feature	Medium	Walk‑forward validation for RL uses a fresh stack per fold. Walk‑forward validation for RL uses a fresh stack per fold. Source: llm_adapter@2026-05-26 Confidence: high	—
Feature	Medium	Env→paper alpha‑decay CI gate fails if PnL decay exceeds tolerance. Env→paper alpha‑decay CI gate fails if PnL decay exceeds tolerance. Source: llm_adapter@2026-05-26 Confidence: high	—

Full changelog

This release makes RL strategies actually deployable. Before v0.6.4, training a policy in the gym env and putting it on live were two different worlds — different code, different physics, and no way to know how much the env was lying to you. Now the same VenueStack drives env, backtest, paper, and live, so a policy you train is the policy you ship (#328, #329, #332).

Action and observation surfaces match what live trading actually needs: signed-quantity limit orders with TIF (#330), the policy seeing its own open orders and getting penalized for rejected fills (#331), portfolio-mode multi-symbol observation and action spaces (#335). Walk-forward validation works on RL the same way it works on classical strategies, with a fresh stack per fold (#333).

The env→paper alpha-decay CI gate (#334) is the trust anchor: synthetic generators with known edge train in the env, then replay through paper. If PnL decay exceeds tolerance, CI fails — silent divergence between training and live can't sneak in.

Full Changelog: https://github.com/FLOX-Foundation/flox/compare/v0.6.2...v0.6.4

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track FLOX C++ trading systems framework with MCP

Get notified when new releases ship.

About FLOX C++ trading systems framework with MCP

All releases →