What tilt looks like in the data
Ask a trader who has blown an account on tilt and you may hear that they knew the next trade was wrong while placing it. Knowing did not stop it. That detail matters for anyone designing a retention system, because it means the fix cannot rely on the trader noticing. The signal has to be read from the outside, and it has to be read from the data.
The data cooperates. Three signals show as trades are placed, before the P&L settles.
Velocity. A disciplined session has a rhythm: setups are waited for, entries are spaced out. We expect a tilted session to compress: trades that arrived once or twice an hour start arriving in clusters. The operational threshold Discentra's engine is designed around is five or more trades inside fifteen minutes. That is our construct, not a peer-reviewed constant. For a trader whose sessions run at once or twice an hour, sustained velocity at that rate looks more like state than strategy, and a firm can check the line against its own traders.
Reentry latency. The tell our revenge-trade rule is built on is the gap between a loss and the next position. A trader following a plan absorbs a loss, reassesses, and re-enters when the setup returns. A trader chasing a loss may re-enter the same symbol inside a minute, at larger size. Loss, fast reentry in the same symbol, bigger position: that is the revenge trading shape our rule looks for, and it is legible from timestamps, symbols, sizes and P&L alone. The one-minute line, like the velocity threshold, is our operational marker rather than a published constant.
Sizing drift. Position size climbing while win rate falls. Each individual trade can look defensible. The trend across a session cannot. Our own size rule is narrower: it reads each trade against the maximum lot size the firm declares for that symbol.
None of this requires new instrumentation. Every prop firm already holds the timestamps, the sizes, and the sequence. What tilt detection adds is the reading of them while the session is still running. The seven fields that reading needs, and the price fields it refuses, are set out in a separate post. The stress science behind the spiral is one reason we do not rely on the trader to self-interrupt; the telemetry is how the firm knows when that moment has arrived. There may also be a quieter failure mode that runs in the opposite direction, where a stressed trader shrinks instead of escalating; that signature and why thresholds miss it is its own subject.
Why the existing stack reads it too late
The standard objection to all of this is that firms already monitor their traders. They do. The question is the tense.
Dashboards aggregate by day. The risk desk reviews yesterday's sessions, and the review is genuinely useful, but a signature that lives inside a fifteen-minute window has been averaged into the daily numbers by the time anyone reads them. The slower churn metrics, read over weeks, are worth tracking; they are a different instrument, tuned to a different timescale.
Loss limits fire at the threshold. A daily loss limit is mechanical, reliable, and always on, and it ends the sequence only after the sequence has run. The trades that carried the account from healthy to breached happened inside the window the limit does not watch.
Journals see the session at 9pm. Post-session review is where traders learn, and it is the right home for reflection. It is also, by definition, after every decision it analyses.
Each of these tools works at its own job. The gap is not the tooling. It is the tense: everything in the standard stack speaks about the session in the past, and tilt is a present-tense event. This is the detection half of the argument made in behavioural risk management for prop firms: rule enforcement catches the violation, and a behavioural layer is built to catch the trader before it.
Who already claims tilt detection
The phrase is starting to appear on products, which makes precision worth the trouble. Three kinds of product are easy to find today, and it is worth being clear about what each one does.
Trader self-rating tools ask the trader to score their own discipline, session by session. Honest answers when calm, and structurally absent in the moment that matters, because the trader on tilt is not filling in a form about it.
Post-session analysis tools score uploaded trade history and flag revenge trades and overtrading after the fact. The signals are right. The tense is wrong: the analysis lands when the session it describes is already over.
Market sentiment gauges measure a different subject. A gauge of how traders in aggregate are positioned across markets is crowd data, and it says nothing about whether one specific trader is coming apart. Market tilt and trader tilt are different measurements, even where they share a word.
Stated precisely: detection of an individual trader's behavioural state, from live trade events, on the firm's side, wired to an intervention, is the position none of these products holds. That is a narrower claim than "nobody detects tilt," and the narrowness is the point. The behavioural triggers worth monitoring are documented. Discentra is designed to read them while they are happening and place a call.
Detection is necessary, not sufficient
A firm could build everything described above and still retain nobody, because a detected signature with no intervention is a better-documented blowup. Detection finds the moment. Someone still has to call the item.
The constraint is the intervention window: the few minutes between the trigger and the next trade, which we treat operationally as a target of about four minutes. Inside that window, the sequence can still change. After it, the next trade may already be placed, and the detection has produced a timestamped record of a loss nobody reached in time.
That window sets the engineering bar for the whole category. Trigger evaluation has to happen in under a second per trade event. The response has to reach the trader within seconds, not at the next dashboard refresh. And the medium matters: at active tilt we use a voice call rather than a notification, because a notification waits in a queue the trader can ignore and a ringing phone asks for an answer now.
What tilt detection is not
It is not surveillance. Traders opt in, consent to the monitoring and the calls, and can leave the programme. The same telemetry read against the trader would deserve the surveillance label; read for the trader, at the moment it helps, it is the firm noticing before the damage instead of after.
It is not trade-blocking. Nothing described here closes a position, cancels an order, or sizes anything. Enforcement tools own hard stops, and the two layers should stay distinct.
And it is not financial advice. Detection triggers coaching: a voice that names what the data shows and points the trader back to their own plan. It never recommends a trade, predicts a price, or suggests a position size. Coaching, not financial advice.
What to ask a vendor who says "tilt detection"
Four questions separate a real-time behavioural layer from a relabelled analytics report.
Which signals, beyond P&L thresholds? If the answer is drawdown percentages, it is enforcement wearing a new name. Velocity, reentry latency, and sizing drift are behavioural; a loss number is financial.
Detected when? Live trade events processed in under a second, or a batch job overnight? The answer decides whether the system can ever act inside the window.
Then what happens? Detection that ends in a dashboard entry reaches nobody in the session. The follow-through, reaching the trader before the next trade, is the part a firm can test against its own retention numbers.
And is the compliance work done? A system that monitors behaviour and phones traders carries consent, disclosure, and data-protection weight. The compliance package is most of what separates a deployable product from a demo.
Discentra is built as the answer to those four questions: behavioural trigger detection from live trade events, a coaching call placed within seconds, opt-in by design, and architected for the consent, disclosure and data-protection work a deployment carries. The full retention layer is designed around the detection described here. The fields it reads are in your data today. The question is whether anything reads them before the next trade.
