EdgeMax 3.0

Inspiration

I've spent ~6 years trading using ICT/Smart Money Concepts, and ran a trading academy under the MurshidFX brand teaching this methodology. The problem: SMC analysis is deeply manual — reading market structure, liquidity, order flow, and multi-timeframe context by eye, in real time, across dozens of markets. I wanted to encode that expertise into a system that never gets tired, never misses a timeframe, and applies the logic with total consistency. EdgeMax is that system.

What it does

EdgeMax is an institutional-grade algorithmic trading platform that analyzes 40+ markets across a 12-frame timeframe ladder (from Monthly down to M1), using two core trading models — a Market Maker Model and a Universal Model — built on ICT/Smart Money Concepts. It ingests live market data, builds structured market intelligence (market structure, liquidity, PD arrays, order flow, time profiles, directional bias), and synthesizes that intelligence into evidence-based trade signals through a dedicated decision pipeline.

How we built it

EdgeMax's entire codebase is written in Rust for performance and correctness. My role has been architect: I explain the ICT/SMC logic — market structure shifts, liquidity sweeps, fair value gaps, order flow — in detail to AI coding agents, primarily Codex, which implement it as working code across an extended series of autonomous sessions. I review, test, and audit every piece against the actual codebase rather than trusting a summary of what was built. EdgeMax 3.0 itself is a from-scratch rebuild: I discarded an earlier live production version (V2 — ~822 files, 3,167+ tests, running at edgemax.space) after identifying deep architectural issues, and have spent about 9 months rebuilding it properly as V3.

Challenges we ran into

The biggest challenge wasn't code — it was catching AI-introduced assumptions before they became load-bearing. During development, an AI assistant introduced errors that damaged parts of the architecture, and separately, I found that some "governance sign-off" documentation had been self-certified by an agent rather than reviewed by an actual human. That forced a shift in process: every claim of "done" or "approved" now gets independently re-verified against the real code before I trust it. I also had to track down a genuine production bug this way — a PD Arrays detection function that was structurally unreachable due to a wiring issue upstream, caught only through direct code audit, not by trusting test-pass claims.

Accomplishments that we're proud of

• Diagnosed and fixed a server-side throughput bottleneck (a synchronous fsync inside a global mutex) that was capping the system at ~185 records/second, and confirmed superlinear concurrency scaling afterward
• Completed a full audit and remediation pass across the Market Intelligence layer, resolving structural contradictions and reconciling dozens of module-level issues
• Rewrote a core "end-to-end" test that had actually been hand-constructing intermediate objects rather than exercising the real pipeline, so it now genuinely proves the full evidence-to-signal chain
• Verified the actual module architecture against the code (14 top-level domain modules) rather than accepting outdated design docs

What we learned

Rigor has to be applied to the AI-assisted development process itself, not just the trading logic. Consistency checks, honest test counts, and independent verification against source code turned out to matter as much as the ICT/SMC methodology itself — an agent that reports success isn't the same as a human confirming it.

What's next for EdgeMax 3.0

Closing out the remaining operational verification gates, running a 24–72 hour feeder soak test, wiring live alerts and secret rotation, and getting the real opportunity classifier and trade lifecycle live-wired — the last steps before EdgeMax is ready for funded trading.

Built With

• 5
• algorithmic
• architecture
• async
• codex
• concepts
• confirm)
• hexagonal
• metatrader
• money
• mql5
• openai
• postgresql
• redis
• rust
• smart
• systems
• tokio
• trading
• using
• websocket
• you're
• zeromq