Correlation-Cluster identifies hidden risk concentration: trades the user thought were diversified but are actually correlated by cluster. The classic retail trap is opening four positions across what feels like four different markets — say EURUSD, EURJPY, EURGBP and EURAUD — and believing the risk is 4x diversified when in reality the risk is 4x concentrated in EUR direction.
The detector classifies each closed trade into a cluster from the symbol alone (no broker-side correlation matrix needed) and walks the trade history with a sweep-line algorithm to compute the peak simultaneous open count per cluster: the largest number of positions ever open at the same time within that cluster.
A peak of 3 or more is the warning signal — even a single moment of holding 3 same-cluster positions means the direction-of-risk during that window was 3x the user's declared per-trade risk.
Cluster-classified trades over the period. The sweep-line on EUR pairs reveals a moment where three different EUR-base pairs were open within the same short window.
Severity bands on the peak count: - Watch — one moment of cluster concentration. Could be coincidence; flag it in the journal and check whether the moment was deliberate (sector thesis) or drift. - Concentrated — the user is holding multiple times the declared risk in one direction at the peak moment. Almost always a hot-hand bias: the trader liked one setup and re-applied it across 'similar' instruments without sizing the bundle as a single bet. - Stacked — catastrophe-shaped risk. One unexpected news event in that cluster drains a multiple of the per-trade budget at once. Cap the cluster at the per-trade budget.
The Onyx-Engine assigns the dominant cluster to one of these bands; the cutoff thresholds are TradeOnyx-internal calibration.
Three causes of cluster stacking: 1. Hot-hand bias — one setup worked, the trader re-applied it across the next-most-similar instruments without recognising the correlation. 2. Sector momentum chase — tech is moving, so AAPL + MSFT + NVDA seem like three independent ways to ride the move. They aren't. 3. Currency event chase — ECB is dovish, so trader shorts EUR via three pairs simultaneously to 'spread the entry'. Spread = leverage in this case.
Tier: Pro Plus. Heavyweight risk-management read; sits alongside Risk-of-Ruin Monte Carlo as the deepest correlation/concentration analysis the engine does.
How to read the card:
1. Hero (left) — peak simultaneous count with the severity verdict. The single number is the worst moment; the verdict tells you how seriously to take it. 2. Three MicroStats — Dominant cluster (e.g. 'EUR'), Trades in cluster, Of total (% of all eligible trades that fell into this cluster). 3. Hint line — actionable: 'position-size by cluster, not by symbol' anchored to the dominant cluster name.
The Correlation-Cluster → Position-Sizing pairing. This card and the Position-Sizing-Consistency card both surface the same root question: are your positions sized according to your risk model? Position-Sizing catches per-trade size drift; Correlation-Cluster catches per-CLUSTER drift. A trader passing Position-Sizing but failing Correlation-Cluster has disciplined per-trade sizing but is letting cluster bundles slip past the budget.
Re-look frequency: weekly. Cluster-stacking patterns stabilise over ~10-20 trades; daily noise isn't useful.
Tier: Pro Plus.