AI Engine

Strategy intelligence that shows its work

Ten stages, one principle: every conclusion is traceable to evidence — and when evidence is thin, the engine says so.

The Golden Rule. A conclusion is only “validated” when ≥2 independent pieces of evidence support it. Otherwise the engine returns UNKNOWN / Not enough evidence — it never guesses a number.
P1

Behavior Analysis

Real behavioral fingerprint from trade deals — timing, direction, holding, lot structure, SL/TP, grid/martingale.

P2

Evidence

Every signal is tied to a source; external evidence (.set/.mq5) is parsed and added, never fabricated.

P3

Knowledge Graph

Nodes + edges (entry/exit/indicator/filter/money-management/blueprint), each with confidence and evidence count.

P4

Comparison

Similarity between two graphs (Dice × confidence closeness) — the single comparison primitive.

P5

Ranking

Scores and ranks many strategies from their graphs, with explainable penalties.

P6

Classification

Detects strategy type (Scalping, Grid, Martingale, Trend…) with multi-label confidence and a quality grade.

P7

Digital Twin Replay

Validates the blueprint against real history; tick-level replay is honestly UNAVAILABLE without a tick DB.

P8

Similarity

Aggregates comparisons across many accounts — nearest-neighbours and strategy families.

P9

Verification

Gate-based decision: VERIFIED / PRELIMINARY / INSUFFICIENT EVIDENCE — not a single score.

P10

AI Recommendation

Reasoning layer that composes all engines: weaknesses, evidence gaps, why-not-verified, risks and next actions.

🧾

Explainable confidence

Confidence is additive and traceable — you can see which evidence contributed how much.

⚖

Conflict detection

Contradicting evidence is surfaced, not hidden — conflicts lower verification instead of being averaged away.