Your Trade Review Process Has Two Jobs. Only One Fits in 30 Seconds.

Every journal tells you to review your trades. Almost none say which review, because the word covers two jobs that have nothing in common.

The first is capture: writing down what the trade was while you still remember the entry accurately. It happens once per trade, it takes seconds, and the data it produces expires.

The second is analysis: asking what forty closed trades have in common. It happens once a week, it takes a quarter of an hour, and it is impossible before the trades exist.

Collapse the two and you get the trade review process most traders abandon by week three. Ten minutes per trade, written as prose, spent deciding whether the trade was a good one. That process asks a question a single trade cannot answer, using data that was already contaminated by the time you sat down to write it.

This post separates them, and spends its weight on the first, because the first is the one with a deadline.

Why One Trade Cannot Tell You Whether the Trade Was Right

Take the setup used in how many trades you need before going live: a 1 to 2 payoff that wins 40% of the time. Expectancy is 0.4 times 2R minus 0.6 times 1R, so 0.20R per trade. The standard deviation of those individual outcomes is about 1.47R.

Those two numbers settle the argument. On one trade, the spread of possible results is 1.47R against an edge of 0.20R. The noise is 7.35 times the signal. The result you are staring at is almost entirely variance with a trace of edge in it.

Put the same thing in sample terms. The two standard error bar closes below the edge at n = (2 times 1.47 / 0.20) squared, which is about 216 trades. Your one closed trade is roughly one two-hundred-and-sixteenth of the evidence needed to say the decision behind it was sound.

So "was that a good trade?" is not a hard question. It is an unanswerable one at n = 1. Every minute the post-trade review spends on it is spent on the single thing that moment cannot deliver.

What One Trade Can Settle

One thing. Whether you followed your own process.

That is a factual question, not a statistical one, and it is decidable in four taps while the trade is fresh. It also has a shape worth drawing, because adherence and outcome together are what decide whether there is anything to do at all:

Read that and notice what it does to the standard review habit. Three of the four squares contain no corrective action. The square most traders spend their review time on, the loss where every rule was followed, is explicitly one of them. A loss taken correctly is the running cost of having an edge. Treating it as a mistake is how a trader arrives at a fifth rule by Friday and a seventh by the end of the month, each one added to prevent a loss that was never preventable.

The square that gets skipped is the top right.

The Square Nobody Reviews

Work it out on 100 trades. Say the win rate is 40% and a rule gets broken on 20 of them. Assume, as the most conservative case available, that breaking the rule changes nothing about the outcome. Then about 8 of those 20 rule breaks still pay out.

Eight trades that rewarded you for ignoring your own plan, against twelve that charged you for it. On the ledger that is fine: the twelve outweigh the eight. On the learning it is not, because a reward that arrives only sometimes is the kind that makes a behaviour stick, which is a point about learning schedules rather than about trading, and a review process that opens only the losing trades never sees a single one of those eight.

This is the entire case for filling the adherence fields on winners. Not for completeness, and not for a tidier dashboard. The winner that broke a rule is the most expensive trade in the month and the only way to catch it is to have recorded, on a day you were pleased with yourself, that the rule was broken.

The 30 Seconds, Ordered by What Expires First

Order the capture by perishability rather than by importance. These are the four things, and they fit in half a minute because none of them requires you to think about whether the trade was any good.

1. The rule fields. Right now you know whether you waited for the candle to close, whether the session was the one you trade, whether the size was the size you planned. In an hour you will know the outcome far better than you know the entry, and your account of the entry will have quietly rearranged itself to fit the result. This is not a character flaw and it is not fixable by trying harder. Labels applied after the outcome is known are partly labels on the outcome, which is why they have to be picked while the answer is still unknown or immediately after, and treated as locked once written.

2. How the trade ended. Stop, target, or your own hand. One tap, and it is a count rather than an estimate, which makes it the most diagnostic field in the file. Fifteen losers of which eleven were manual exits before the stop is a completely different problem from fifteen clean stop-outs, and no win rate, profit factor or equity curve tells those two apart. If you import from MetaTrader, the platform's own report already carries it as the Close Reason column, so you get it without typing anything.

3. The screenshot. Two seconds, and it is the only record of what the chart looked like before you knew the answer. Reconstructed later from a flat price series, every entry looks either obvious or insane depending on what happened next.

4. One line, and only if something broke. What you expected, and what told you that you were wrong. Skip it entirely on a clean trade: the rule fields already carry that information in a form you can filter on, and a line written for the sake of the routine is a line nobody reads.

What does not belong in the 30 seconds: a paragraph about how you felt, a verdict on whether the trade was good, and above all a change to the rules. Rules get changed on a sample, in daylight, not by a trader holding a fresh loss.

What the 30 Seconds Cannot Do

Everything that needs more than one row. Which setup actually pays, whether Tuesdays are real or noise, what the cost drag is doing to expectancy, whether the drawdown you are in is normal for your win rate. None of that is visible from inside one trade, and attempting it per trade produces a conclusion a week, each one overturning the last.

That is the weekly pass, and it has its own order of operations: check coverage before comparing anything, because fields filled on 90% of winners and 60% of losers will report a difference that is pure selection. The three questions that make a weekly review worth fifteen minutes run coverage first for exactly that reason, and tracking performance across weeks and months covers which numbers are stable enough to read at that cadence.

The division is clean. Capture is per trade and perishable. Analysis is per sample and patient. Trying to do the second during the first is what makes a review process collapse.

The Checkbox Is Only as Honest as You Are

Here is the limit that applies to every version of this, whether a human does the judging or software does.

"Rule met" means you ticked a box. Nothing verifies it against the chart. A trader who waited for most of the candle to close, or took size that was close enough, will tick met, and from that point forward the entire adherence number is measuring self-report rather than behaviour.

Two defences, both cheap. First, tick at entry wherever the field allows it, so the box is answered before the outcome exists to argue with. Second, cross-check against measurements no memory touches: position size, minutes since the previous trade, trade number of the day, how many conditions were present. If the trades you tagged as clean are indistinguishable from the rest on every measurable dimension and differ only in result, you have logged your P&L twice. Turning adherence into a number is worth doing, and it is worth knowing it measures what you recorded.

What a Machine Can Decide Here, and What It Cannot

The post-trade moment is a good fit for automation, because the only decidable question in it is mechanical.

A machine can do three things well. It can state which square of the grid the trade fell in, since that is a lookup against the fields you filled. It can quote your recent numbers back with their scope attached, which is arithmetic and should be computed rather than written. And it can tell you the second time a pattern appears rather than the fortieth, because it remembers the first.

It cannot do four things, and none of them are a model problem.

It cannot tell you your rule set is wrong. Adherence is measured against the rules you wrote, so a trader with perfect discipline on a bad rule scores 100% while losing money.

It cannot see the trade you talked yourself out of. Hesitation leaves no row, so the most expensive habit a trader has is invisible to every journal ever built.

It cannot separate an edge from a run, for the reason in the arithmetic above. Nothing can, before the sample arrives.

And it reads your checkbox as the truth. An automated verdict inherits the honesty of the field it is given, which puts the useful question back where it started: did you tick it before or after you knew?

What an AI trading coach can and cannot tell you works through the four grades of statement behind that split, from arithmetic you can verify in one click down to prediction that sits outside the data entirely.

How This Runs in TradingSFX

The product is built around the split in this post, so it is worth saying plainly which parts are free and which are not.

Quick Mode exists for the capture pass. Toggle it in the trade form and a trade is win or loss plus an optional dollar amount, with no entry, exit, stop or price fields required. With no amount entered it falls back to R, so a winner logs at your planned reward multiple and a loser at -1R, which keeps the equity curve in R honest even on a day you never opened the platform again. Available on every plan, including the free one.

Rule fields with a required answer. Any confluence can carry the answer that counts as met, which is what turns the grid above from a mental exercise into a column. Adherence is then pooled division across every rule check, so it matches the dashboard rather than being written by a model. Free accounts get one custom field, paid plans more.

The verdict arrives on its own. Save a trade on a paid plan and a short judgement appears without being asked for, one to three sentences, following the grid: rules met and a win confirms the kind of edge, rules broken and a win warns against normalising it, rules broken and a loss names the rule, rules met and a loss is classified as variance rather than a mistake. Any number it quotes from your recent history is tagged with its scope, so "your last 20 trades" never gets passed off as a lifetime figure. Paid plans include a generous monthly allowance for it.

Three details that follow from the thesis rather than from a feature list. The verdict is skipped entirely on a trade whose entry date is more than two weeks old, because a judgement on a trade you have already stopped thinking about is noise with a cost. Replay and backtest trades get the fields but no per-trade verdict, since a session produces dozens of closes and the useful read there is the batch. And once the coach has a profile of your trading, a repeat offence reads as the pattern it is instead of a fresh observation.

The thumbs are there for a reason. Every verdict takes a rating, and a thumbs down with a reason is how a wrong one gets fixed instead of tolerated.

Comparing how different journals handle the automated half, rather than how to run the review, is the AI journal roundup instead.

The 30-Second Checklist

Tape this somewhere. It is the whole per-trade routine.

1. Tick the rule fields. Winners included, especially winners.
2. Stop, target, or your hand?
3. Screenshot.
4. One line, only if something broke.
5. Close the form. Do not decide anything.

Ten trades a week is five minutes. The version that takes ten minutes a trade is better on paper and gone by the third week, and a review routine that only runs on good weeks produces a sample made entirely of good weeks.

Bottom Line

The review that decides whether you were right needs 216 trades. The review that records what happened needs 30 seconds and cannot be done later. Almost all trade review advice is aimed at the first and performed at the time of the second, which is why so much of it feels like work and produces so little.

Separate them. Spend half a minute per trade capturing facts that expire, spend a quarter of an hour a week on questions that need a sample, and stop asking a single closed trade to tell you something it does not contain.

Start with the one field you do not have yet: how the trade ended. Log it on every trade for a month, then count how many of your losers you closed by hand before the stop. It is one number, it takes a month to earn, and it says something no win rate on your dashboard can.

Logging the result takes seconds on the free plan, 10 trades a month, no card. The automated verdict and the per-rule analysis sit on Pro at $19.99 a month.

Published October 9, 2026. Every figure in this post is arithmetic worked openly from the stated illustrative sample: no survey, study effect size or user count is cited anywhere, and no competitor feature, limit or price is claimed. Product behaviour was read from the code on 9 October 2026 rather than from memory, including the verdict's adherence and outcome decision rules, its 20-trade context scope and scope-tagging requirement, the two-week entry-date cutoff after which it is skipped, the exclusion of replay trades, and Quick Mode's fallback to R when no amount is entered. Plan facts: Basic is free forever at 10 trades a month, Pro $19.99/mo or $215.99/yr, Premium $29.99/mo or $323.99/yr.