Claude is a filter,
not a generator.
Strategies stay deterministic. The AI's role is to refuse setups that look right on paper but are wrong in this market state.
01Sixteen documented strategies evaluate every tick across M1, M5, M15. They produce candidate setups with pair, direction, SL, TP. This step is mechanical and deterministic — same input always produces same output.
02Claude reviews each candidate against current macro narrative, spread, news risk, and DXY direction. Output: approve, half-lot, or reject with a confidence score 0–100.
03Only setups with verdict = approve and confidence ≥ 80 reach MT5. Everything else is logged but not executed. Decision audit is visible at /claude.
- ·Not a signal service. Setups are taken, not subscribed to.
- ·Not "AI trading." Claude reviews; strategies execute.
- ·Not optimized for return. Optimized for explainable decisions.
- ·Not live. Currently demo phase — no real capital at risk.
Past backtests do not guarantee forward results. Spread widening, gaps over weekends, broker downtime, and unforeseen market regimes can all degrade live performance.
Claude is conservative — it rejects more often than approves. That can also be wrong. Decisions Claude refuses today may have been profitable. The audit trail at /claude shows everything, including misses.