\n| Hybrid \/ C-book<\/td>\n | Some flow warehoused, some hedged out to LPs<\/td>\n | Mix of both, tuned per client<\/td>\n | Depends entirely on disclosure<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n Here is the nuance almost no marketing page admits: B-book is not automatically bad, and A-book is not automatically clean.<\/strong> A well-run B-book desk can give tight spreads and instant fills to ordinary traders, because statistically most retail accounts lose and the broker has no reason to interfere. An A-book broker can still route you to a slow or hostile liquidity provider and hide behind “we just pass it on.”<\/p>\nThe real transparency question is not “which book am I in?” but “does the broker switch my book based on whether I am profitable, and do they tell me?” That silent switch is where honest execution ends.<\/p>\n \n\n\n\nWhen routing is fair<\/h4>\nYour book assignment is disclosed or at least consistent. Fills track the visible market. Winning and losing accounts get the same spreads, the same speed, and the same slippage behavior.<\/p>\n<\/td>\n | \nWhen routing turns hostile<\/h4>\nYou are quietly moved to a worse path the moment you become profitable. Spreads widen only on your account, slippage becomes one-directional, and rejects appear only on trades that would have won.<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n <\/p>\n Toxic flow: why brokers treat some traders differently<\/h2>\nFrom the broker’s side, not all order flow is equal. “Toxic flow” is the industry term for order flow that is consistently profitable against whoever is on the other side, usually because the trader has an informational or speed advantage the broker cannot hedge cheaply.<\/p>\n Flow a risk desk tends to flag as toxic includes:<\/p>\n \n\n\n| Flow type<\/th>\n | Why it is hard for the broker<\/th>\n<\/tr>\n<\/thead>\n | \n\n| Latency arbitrage<\/td>\n | Trader acts on a faster feed and enters before the broker’s quote catches up<\/td>\n<\/tr>\n | \n| News trading<\/td>\n | Large directional moves in milliseconds around scheduled releases<\/td>\n<\/tr>\n | \n| Fast scalping<\/td>\n | High trade count, tiny targets, systematically picks off stale quotes<\/td>\n<\/tr>\n | \n| Multi-account correlation<\/td>\n | Many accounts trading the same signal at the same instant<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n This matters for transparency because toxic-flow handling is exactly where undisclosed execution changes get applied. A broker that cannot profitably warehouse your flow, and will not or cannot hedge it to a liquidity provider, has three options: reject it, delay it, or degrade its price. All three are invisible unless you are measuring.<\/p>\n The honest position:<\/strong> a broker is entitled to protect itself from flow it cannot hedge. The transparency failure is doing it silently while advertising “no dealing desk” and “zero requotes.” Protection is legitimate. Concealment is the problem.<\/div>\n <\/p>\n Last look: the hidden hold window<\/h2>\nLast look is a mechanism where, after you send an order, the party filling it gets a short window to accept or reject the trade after already seeing your request. It originated in institutional FX and is now common in the retail bridge layer too.<\/p>\n The window is usually small, from roughly one millisecond to a few tens of milliseconds. In that window the price can move. What happens next is the whole story:<\/p>\n \n\n\n| Price moves during last-look window<\/th>\n | Symmetric (fair) last look<\/th>\n | Asymmetric (hostile) last look<\/th>\n<\/tr>\n<\/thead>\n | \n\n| In your favor<\/td>\n | Filled, sometimes with price improvement<\/td>\n | Rejected or re-priced away from you<\/td>\n<\/tr>\n | \n| Against you<\/td>\n | Rejected under the same threshold<\/td>\n | Filled at the worse price<\/td>\n<\/tr>\n | \n| No move<\/td>\n | Filled at the quoted price<\/td>\n | Filled at the quoted price<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n Symmetric last look applies the same tolerance in both directions and is defensible as a genuine risk check. Asymmetric last look is heads-the-broker-wins, tails-you-lose.<\/strong> You only ever get filled when the move helped the other side, and rejected when it helped you. Over thousands of trades this is a large, silent tax that looks like “bad luck” on any single fill.<\/p>\nYou can detect asymmetry without any inside access. If you log the requested price, the fill price or rejection, and the mid-market price at the moment of each request, the distribution tells the truth: fair last look rejects roughly symmetrically, hostile last look rejects almost entirely on trades that were about to move in your favor.<\/p>\n <\/p>\n Slippage forensics: symmetric vs asymmetric<\/h2>\nSlippage is the difference between the price you requested and the price you got. Slippage by itself is normal. Markets move. The question that separates a fair broker from a hostile one is the shape of your slippage distribution<\/strong>, not its size.<\/p>\nDefine per-trade slippage in pips as:<\/p>\n # Positive = filled better than requested, Negative = filled worse<\/span>\r\nslippage_pips = (fill_price - requested_price) \/ pip_size * direction\r\n\r\n# direction = +1 for buy, -1 for sell, so the sign is consistent<\/span>\r\n# Then look at the DISTRIBUTION across many trades, not one trade<\/span>\r\nsymmetry_ratio = count(slippage > 0) \/ count(slippage < 0)<\/pre>\n\n\n\n\nFair execution signature<\/h4>\nPositive and negative slippage both occur. The symmetry ratio sits near 1.0. Price improvement is real and shows up in your logs, not just in the brochure. Slippage size scales with volatility, the same for winning and losing trades.<\/p>\n<\/td>\n | \nManipulated execution signature<\/h4>\nSlippage is almost always negative. Symmetry ratio far below 1.0. Price improvement essentially never appears. Slippage spikes specifically on your profitable trades and around news, but not on losers.<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n This is why keeping your own tick-by-tick records matters more than any review site. A broker’s aggregate “average slippage” number can look fine while your distribution is entirely one-sided. The average hides the asymmetry. The distribution exposes it.<\/p>\n <\/p>\n Requotes, spread widening, and server-side execution plugins<\/h2>\nBeyond last look and slippage, three more tools change your fills at the server level, before your order ever reaches a market.<\/p>\n Requotes<\/h3>\nInstead of filling or rejecting, the server offers you a new, worse price and asks you to confirm. On fast strategies the requote delay alone is enough to destroy the trade. A broker advertising “no requotes” that issues them only on your winning trades is running selective execution.<\/p>\n Selective spread widening<\/h3>\nSpreads legitimately widen in thin liquidity and around news. The hostile version widens the spread on specific accounts or specific instruments only when a profitable pattern is detected, then narrows again. Because it is account-specific, you cannot see it by comparing to a public chart. You can only see it by logging your own quoted spread over time.<\/p>\n Server-side execution plugins (virtual dealer plugins)<\/h3>\nThese are server modules that sit between your order and the fill and apply configurable delays, artificial slippage, requote probability, or outright rejection to targeted accounts. They exist specifically to neutralize flow the desk considers toxic. The trader sees nothing except worse results. The only defense is measurement: a plugin that adds, say, 30 to 300 milliseconds of artificial delay to your account leaves a clear fingerprint in your execution-latency log that a clean account does not have.<\/p>\n Pattern to watch for:<\/strong> execution that is perfect during your demo and your first weeks live, then degrades right as your account turns consistently profitable. That timing is the signature of an account being reclassified and moved behind an execution filter.<\/div>\n <\/p>\n Seven tests: is your broker B-booking you against your interest?<\/h2>\nYou cannot read a broker’s routing table, but you can run controlled tests and let the data answer. None of these require special software beyond a trade log with timestamps and prices.<\/p>\n 1<\/span><\/p>\nLog slippage symmetry.<\/h4>\n Record requested vs fill price on at least 200 trades. A healthy symmetry ratio is near 1.0. A ratio far below that, with almost no positive slippage, is the strongest single signal.<\/p>\n<\/div>\n 2<\/span><\/p>\nCompare demo vs live fills.<\/h4>\n Run the identical strategy on demo and live in parallel. Demo is almost always pure B-book with cosmetic fills. A large, persistent gap between demo and live execution quality tells you how your live flow is really handled.<\/p>\n<\/div>\n 3<\/span><\/p>\nTrack rejection timing.<\/h4>\n Note the mid price at every reject. If rejections cluster on trades that would have moved in your favor, you are seeing asymmetric last look, not random risk checks.<\/p>\n<\/div>\n 4<\/span><\/p>\nMeasure execution latency over time.<\/h4>\n Log the milliseconds from send to fill. A sudden step-up in average latency, especially right after your account became profitable, points to an added server-side delay.<\/p>\n<\/div>\n 5<\/span><\/p>\nWatch your spread, not the public spread.<\/h4>\n Record your own quoted spread per instrument. Account-specific widening that does not match the broker’s public quotes is a targeting signal.<\/p>\n<\/div>\n 6<\/span><\/p>\nTest around scheduled news.<\/h4>\n Toxic-flow handling is most aggressive around releases. If fills, spreads, and rejects sharply degrade only in your account during news windows, the desk is treating your flow as informed.<\/p>\n<\/div>\n 7<\/span><\/p>\nScale size and watch what changes.<\/h4>\n Execution that is clean at 0.1 lots but suddenly slips, requotes, or rejects at larger size shows a threshold in the risk logic. Fair execution degrades smoothly with liquidity, not in a step at the point you become material.<\/p>\n<\/div>\n <\/p>\n Measuring execution quality objectively<\/h2>\nEverything above becomes actionable once you stop arguing about intentions and start scoring numbers. These are the core execution-quality metrics any serious trader should track per broker.<\/p>\n \n\n\n| Metric<\/th>\n | What it measures<\/th>\n | What good looks like<\/th>\n<\/tr>\n<\/thead>\n | \n\n| Fill rate<\/td>\n | Share of orders filled vs sent<\/td>\n | High and stable across market conditions<\/td>\n<\/tr>\n | \n| Rejection rate<\/td>\n | Share of orders rejected<\/td>\n | Low, and not correlated with your winners<\/td>\n<\/tr>\n | \n| Slippage symmetry<\/td>\n | Balance of positive vs negative slippage<\/td>\n | Ratio near 1.0, price improvement present<\/td>\n<\/tr>\n | \n| Execution latency<\/td>\n | Milliseconds from send to fill<\/td>\n | Low and stable, no step-changes over time<\/td>\n<\/tr>\n | \n| Requote frequency<\/td>\n | How often you are re-priced<\/td>\n | Near zero, and not selective on winners<\/td>\n<\/tr>\n | \n| Effective spread<\/td>\n | Spread you actually pay, per instrument<\/td>\n | Matches public quotes, no account-specific markup<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n The reason to collect all six rather than one is that a hostile broker can keep any single number looking respectable. It is the combination, scored consistently over time and across market states, that reveals the execution path. This is precisely the logic behind the BJF Broker Execution Quality Index (BEQI), an open methodology for turning your own trade and tick logs into a single comparable execution score, so two brokers can be ranked on evidence instead of reputation.<\/p>\n Principle:<\/strong> reputation is a lagging, gameable signal. Your own execution log is a leading, un-gameable one. Any broker can buy reviews. None can edit the tick-by-tick record sitting on your machine.<\/div>\n <\/p>\n Why this is existential for arbitrage, latency, and news traders<\/h2>\nFor a discretionary swing trader holding for days, a few tenths of a pip of asymmetric slippage is an annoyance. For an arbitrage, latency, or news strategy it is the entire profit and loss.<\/p>\n These strategies win by capturing a very small, very short-lived edge many times. The average profit per trade is tiny by design. That means the strategy is only as good as its worst fills, and a hostile execution path attacks exactly the trades that were supposed to win. As covered in our work on the execution-time gap, an edge that looks strong in a backtest can go negative in production purely because of added execution latency and asymmetric fills, with nothing wrong in the strategy logic at all.<\/p>\n The practical consequences:<\/p>\n \n\n\n| Strategy<\/th>\n | What a hostile execution path does to it<\/th>\n<\/tr>\n<\/thead>\n | \n\n| Latency arbitrage<\/td>\n | Added server delay erases the millisecond edge the whole strategy depends on<\/td>\n<\/tr>\n | \n| News trading<\/td>\n | Selective spread widening and rejects around releases remove the exact fills that pay<\/td>\n<\/tr>\n | \n| Fast scalping<\/td>\n | Asymmetric slippage turns a positive expectancy into a negative one trade by trade<\/td>\n<\/tr>\n | \n| Pair \/ stat arbitrage<\/td>\n | One-legged fills and requotes break the hedge and leave you with naked risk<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n This is why serious execution-sensitive traders do two things: they measure their broker relentlessly, and they model realistic execution in testing rather than assuming perfect fills. Backtesting on real tick data with modeled execution latency and variable spread, the way the SharpTrader Optimizer does, is the only way to know whether an edge survives contact with a real, imperfect execution path before you risk capital on it.<\/p>\n <\/p>\n Frequently asked questions<\/h2>\n\n Is B-book execution always bad for me?<\/h3>\nNo. A B-book broker can offer tight spreads and instant fills, and for most retail traders there is no interference at all. The risk is undisclosed switching, where your execution quietly degrades once you become profitable. The model is not the problem, concealment is.<\/p>\n What is the difference between last look and slippage?<\/h3>\nSlippage is being filled at a different price than requested. Last look is a short window before the fill in which the other side can accept, reject, or re-price your order after seeing it. Asymmetric last look is a common source of one-sided slippage and selective rejections.<\/p>\n Can I prove my broker is manipulating my fills?<\/h3>\nYou can build strong evidence. Log requested price, fill price or rejection, mid-market price, and execution latency for every trade. If slippage is systematically negative, rejects cluster on your winners, and latency stepped up when you became profitable, the data speaks clearly even without access to the broker’s servers.<\/p>\n Does “no dealing desk” or “ECN” mean my execution is fair?<\/h3>\nNot on its own. Those labels describe a routing model, not the quality of the liquidity behind it or whether your specific account is being filtered. Measure fill rate, slippage symmetry, and latency yourself rather than trusting the label.<\/p>\n Why does my strategy work on demo but fail live?<\/h3>\nDemo servers usually simulate near-perfect fills. Live flow, especially flow a desk views as toxic, can be routed behind delays, asymmetric last look, or execution filters. A large, persistent demo-versus-live gap is itself a measurement of how your live flow is handled.<\/p>\n What is a virtual dealer plugin?<\/h3>\nIt is a generic term for a server-side module that applies configurable delay, artificial slippage, requote probability, or rejection to targeted accounts. It is used to neutralize flow a broker cannot hedge profitably. Its fingerprint shows up as added, account-specific execution latency in your logs.<\/p>\n<\/p><\/div>\n <\/p>\n \n Subscribe to BJF trading research<\/h3>\nNew articles, research papers, and product releases, delivered when we publish them. Broker execution, latency arbitrage, and news trading, with the data behind them.<\/p>\n <\/div> | | | | | | |