Dealing-Desk Intervention
Broker-side software or manual control that alters how client orders are handled between receipt and fill — added delay, requote thresholds, asymmetric slippage tolerance — on platforms where the broker is the counterparty to the trade.
On a dealing-desk (B-book) platform the broker does not route the client’s order to an external market; it takes the other side. Internalisation is not inherently abusive — it is how much of retail FX and CFD trading is served at all, and a broker that hedges its net book can be economically neutral to client outcomes. But the structure creates a standing conflict: on the unhedged portion of the book, the client’s loss is the broker’s gain, and the broker also operates the software that decides how the client’s orders are executed.
Dealing-desk intervention is the exercise of that position. The tooling is documented rather than hypothetical: server-side plugins for the dominant retail platforms — sold openly under names like “virtual dealer” — allowed brokers to configure, per client group, execution delays, requote thresholds, and slippage rules. Regulatory actions against US retail FX brokers in the early 2010s documented the abusive configuration specifically: slippage applied asymmetrically, so that price improvement during the delay was kept by the broker while price deterioration was passed to the client. The delay itself is the enabling primitive — a manufactured last look window in which the market’s drift can be observed before the fill is priced.
The operative fact for traders is that intervention is invisible in any single trade and statistical in aggregate. A delayed fill looks like latency; an asymmetric slippage rule looks like bad luck — once. Across hundreds of executions, the signatures separate cleanly from noise: execution-delay distributions with structure that infrastructure cannot explain, slippage asymmetry conditioned on interim drift, requote clustering around the moments discretion is most valuable, and feed behaviour like censoring around the same windows. None of this requires access to the broker’s server. The broker’s own execution records, timestamped and compared against an independent reference feed, carry the fingerprint — or its absence.