How this business actually works.
Technical writing on the mechanics behind each thing BrokerOS claims to do. No vendor framing, no invented statistics, just the mechanism, explained properly, by people who had to build against it.
Markout: why your liquidity provider looks fine until you measure it
Fill rate and average spread say almost nothing about execution quality. The number that matters is where the market went after you traded, and most brokers have never calculated it.
TakeawayA provider can win on spread and still cost you more per million than the one you rejected.
Read it →Approval-rate decay: how to read a payment cascade properly
A group approval rate is an average of averages, and it hides the only thing you need to see. Decomposition by issuer country, card scheme, and attempt index is where the money is.
TakeawayA single corridor can fall 19 points while the group number moves less than one.
Read it →Rebate tier drift, and the reconciliation nobody runs
Introducer agreements are renegotiated continuously and calculated from a spreadsheet that is updated occasionally. The gap between those two facts is a recoverable liability.
TakeawayVolume attribution, not the tier table, is where most rebate errors actually originate.
Read it →Funded, never traded: the shape of the activation window
The interval between a cleared deposit and a first order is the most predictive series a retail broker owns, and the one most likely to be reported as a single monthly percentage.
TakeawayActivation is not a rate. It is a decay curve, and the useful decisions live in its first 72 hours.
Read it →A-book, B-book, and the real cost of classifying flow wrongly
Internalisation is a risk decision priced in basis points and settled in months. Treating it as a static per-client flag is how a profitable book quietly stops being one.
TakeawayToxicity is a property of an order in a market state, not a permanent property of a client.
Read it →Why the settlement break is never where you look for it
Provider statements, your ledger and the trading platform disagree for structural reasons: cut-off times, conversion timing, and fee netting. Most reconciliation effort is spent rediscovering this.
TakeawayFour break classes explain the overwhelming majority of differences. Only one of them is fraud.
Read it →Slippage is two distributions reported as one number
Positive and negative slippage have different causes and different beneficiaries. Averaging them produces a figure near zero that is true, useless, and hides a systematic transfer.
TakeawaySymmetry is the test, not magnitude. Compare venues on the ratio of tails, never the mean.
Read it →Document expiry is a deposit problem wearing a compliance badge
Onboarding attrition is optimised because it is visible in a funnel. Mid-life document expiry is neither measured nor owned, and it freezes funded accounts that were already trading.
TakeawayChasing thirty days before expiry roughly doubles completion. The message does not need to change.
Read it →Withdrawal speed is a retention metric, not a payments one
The slow mode in your withdrawal distribution is queue time, not rails. Clients rarely complain about it; they quietly withdraw more next time and then stop trading.
TakeawayThe distribution is bimodal, so the reported mean describes no withdrawal that actually happened.
Read it →Swap is a spread business, and most brokers run it at a loss
The client swap table is reviewed when someone remembers. The wholesale tom-next rate moves every night. Everything expensive about overnight financing follows from that one mismatch.
TakeawayWatch margin per lot per symbol, signed, with long and short kept apart. Never the revenue line.
Read it →Stop-out is a sequence, and the order decides the loss
A close-out is a loop that liquidates one position, recalculates, and decides whether to liquidate another. Which position it picks first is a configuration default almost nobody has changed.
TakeawayClosing one leg of a hedge removes the margin relief, so a rescue can push the account further down.
Read it →Chargebacks: the ratio decides your cost, not the disputed amount
Scheme monitoring works on a monthly ratio whose numerator and denominator describe different cohorts. In a falling month the ratio rises while disputes fall, and nothing about your book has changed.
TakeawayThe reason code fixes which evidence is admissible. Evidence that answers a different code is not weak, it is irrelevant.
Read it →CPA looks cheaper than revenue share until you model the payback
The qualification bar is not an accounting detail. It is a specification handed to a counterparty who is paid to satisfy it, and it will be satisfied exactly and no further.
TakeawayPlot first deposits against the bar. The spike at the threshold is how much of the book was engineered.
Read it →Revenue per million is four businesses reported as one number
Markup, commission, swap and risk result have different drivers, different volatilities and different signs. Dividing their sum by notional volume produces a figure that cannot tell you which one moved.
TakeawayThree of the four terms are fees. The fourth is a trading position. Never put them in one headline.
Read it →Net exposure per symbol is the wrong number for almost every question
A book flat on every line can carry a large directional position. Long EURUSD is long euro and short dollar, and the dollar leg of every pair on the board appears on no symbol screen anywhere.
TakeawayThe dollar leg is the largest position on most retail books and the only one never displayed.
Read it →The client money reconciliation is a daily solvency test
It compares what you owe clients against what you hold for them, and a shortfall is funded from the firm the same day. The requirement is floored per client, and that floor is the part that gets missed.
TakeawayNetting negative equity across clients hides a shortfall on exactly the morning it is real.
Read it →
We built BrokerOS because of all of this.
Every article above describes a problem that needs several systems joined together before it can even be measured. That joining is what Atlas does.