Neural Trader: what are you building? (patterns, plugins, and the MT5 EA reference example)

[ruvnet]

Kicking off a discussion thread for everyone using — or curious about — the ruflo-neural-trader plugin and the broader algorithmic-trading surface in ruflo. Drop your strategies, questions, war stories, and feature requests here.

---

What ruflo-neural-trader actually is

For folks new to it: ruflo-neural-trader is a thin-but-real bridge between Claude Code and the neural-trader npm package. It gives you skill-level commands for the full strategy lifecycle:

Skill	What it does
trader-train	Train LSTM / Transformer / N-BEATS models on OHLCV data
trader-signal	Z-score anomaly detection → ranked signals (spike, drift, flatline, oscillation, pattern-break)
trader-backtest	Walk-forward Rust/NAPI back-testing (8-19× faster than Python)
trader-regime	Classify the current market regime (bull / bear / range)
trader-risk	VaR, CVaR, Kelly criterion sizing, circuit breakers
trader-portfolio	Mean-variance optimisation + rebalancing
liquidity-sweep (new)	Hammer / shooting-star sweep detection at fixed RR

Underneath, every run is wired into ruflo's memory + intelligence stack:

• AgentDB persists every signal and back-test summary with 384-dim ONNX vector embeddings
• HNSW search finds historically similar setups (150×–12,500× faster than linear scan)
• SONA consumes win/loss verdicts via hooks post-task and adapts pattern weights over time
• EWC++ prevents catastrophic forgetting when you add a new symbol or regime

Translation: every back-test you run makes the next one smarter, not just slower.

---

Reference example: an MT5 EA from issue #1720

Issue #1720 asked for an MT5 Expert Advisor that trades hammer / shooting-star liquidity sweeps at fixed 1:1 or 1:2 RR. We built it as a complete worked example — same strategy implemented twice (MQL5 for native MT5 execution, JavaScript for back-testing) with optional ruflo integration on top.

• Tutorial in repo: v3/docs/examples/mt5-hammer-star-sweep/
• Public gist (single-page reference): https://gist.github.com/ruvnet/07f1571246910ac551fc0f91774045e2

Highlights worth lifting for your own strategies:

1. One-source-of-truth pattern — strategy logic lives in bridge/strategy.mjs. The MQL5 EA mirrors the same rules. You back-test in Node, deploy in MT5, and the rules can't drift apart.
2. Optional ruflo integration — add --ruflo --symbol EURUSD to the back-tester and it'll: lazily run memory init, look up prior back-tests for that symbol, store the new result, record a SONA trajectory verdict. Without the flag, it's a pure-JS strategy with zero deps.
3. Regime gating — the live bridge calls neural-trader --signal scan and blocks longs in bear regimes / shorts in bull regimes. Cheap pattern + smart filter beats either alone.
4. 26 tests proving the rules are correct before any money moves. npm test is the first line of the README for a reason.

---

Discussion prompts

If you've got time, pick any of these to chime in on:

1. What strategies are you running through ruflo today? Trend-following, mean-reversion, pairs, options vol, something weirder?
2. What's missing from the current plugin? Asian-session filters? Order-book microstructure? Crypto on-chain signals? Multi-leg options?
3. Brokers / venues — which MT5 brokers, Alpaca, IBKR, crypto exchanges, prop-firm rules are you wiring into?
4. Reproducibility — how are you handling slippage / spread modelling in back-tests vs. live? Any patterns worth standardising?
5. AI-specific — has SONA's pattern learning actually shifted your signal quality over time, or is it noise? Share data if you can.
6. Risk management — the example uses fixed 1:1 / 1:2 RR with 1% per trade. What's your sizing approach when you stop wanting fixed RR?

Drop links to your own gists / repos / write-ups. The goal is to build a shared playbook so people aren't starting from a blank chart every time.

---

Contributing back

If you ship a strategy you're happy with, the cleanest way to share it is:

1. Fork the repo
2. Add a folder under v3/docs/examples/<your-strategy>/
3. Include a back-test, the strategy code, and tests (the MT5 EA example is the template)
4. Open a PR — we'll review and merge

Or if it's smaller scope, just paste the gist link in a reply.

—
Pinging anyone who's been working on the trading side: feel free to @ in folks I missed.