What Trading Coaches Should Track for Their Students (Screenshots Are Not Data)

Why Screenshot Reviews Do Not Scale

Start with the charitable reading, because it is also the correct one: students are not hiding their bad trades. They are showing you the trades they remember. Those are two different biases and only the second one is universal.

Memory of a trade tracks the size of its outcome and the strength of the feeling attached to it. The 8R runner gets remembered. The full stop-out that hurt gets remembered. What does not get remembered is the body of the distribution: the 0.3R scratch, the trade closed early out of boredom, the entry taken twenty minutes outside the session the curriculum specifies, the position that was a quarter larger than planned for no reason the student could reconstruct afterwards. Those trades are forgettable precisely because nothing dramatic happened, and they are also where the drift lives.

So a screenshot review gives you a sample conditioned on outcome magnitude. You see the tails and never the middle. Every conclusion you draw from it is drawn from the wrong distribution, and the error does not shrink as the student sends you more screenshots - it just gets better documented.

Three further problems ride along with it:

• No denominator. A screenshot of a winning breakout is uninterpretable without knowing how many breakouts the student took. Five winners out of six is a student with a setup. Five out of thirty is a student with a lottery ticket. The screenshot is identical in both cases.
• Not comparable. Dollar outcomes across different symbols, account sizes and position sizes cannot be averaged into anything meaningful. Until each result is normalised by the risk taken on that specific trade - an R-multiple - you cannot compare two of a student's own trades, let alone two students.
• It does not scale arithmetically. Twelve students averaging twenty trades a month is 240 trades. At three minutes of genuine attention each, that is twelve hours of review before you have thought about a single one of them.

None of this is an argument against looking at charts with a student. It is an argument against charts being the only instrument you have. The chart review is where you teach; it is a terrible place to find out what needs teaching.

The Four Numbers That Tell You Where a Student Actually Is

You do not need a dashboard of forty metrics. You need four numbers that answer four different questions, read in the same order every week.

Expectancy is the mean of the R series, and it is first because it is the only one of the four that speaks to whether the student has anything at all. It is also the one most likely to be built on five trades, so never read it without the trade count sitting beside it.

Risk adherence is the one coaches skip, and it is the one that matters most. It gets its own section below, because there is a piece of arithmetic in it that surprises people.

Tag coverage is a process measurement disguised as a data-quality measurement. If a third of a student's trades carry no setup label from your curriculum, that is not laziness about journaling - that is a third of their trading happening outside the thing you are teaching them. The untagged pile is usually the most interesting object in the account.

Rolling SQN combines mean and spread into one figure, and it comes with a warning label. The formula carries a square-root-of-sample-size factor, which means the same unchanged system scores higher the longer it is traded. A student whose SQN rose from 0.7 to 1.9 over a quarter may have improved, or may simply have logged more trades. Compare a student only against their own history at a similar trade count, and never against another student's figure computed over a different number of trades.

Diagnosing Process Failure vs Edge Failure

This is the single most valuable thing a coach can do, and almost nothing on a standard dashboard helps with it. A student is down over sixty trades. There are two completely different diagnoses and they look identical in the equity curve:

• Process failure. The strategy is fine. The student is not trading it - skipping signals, entering early, moving stops, sizing by feel.
• Edge failure. The student is following the rules faithfully and the rules do not work, or have stopped working in the current regime.

The prescriptions are opposites. Coach discipline at a student whose edge is broken and you will spend six months making them very disciplined at losing money. Rewrite the strategy for a student whose problem was execution and you have removed the only thing that was working.

The data cut that separates them is simple and almost nobody runs it: partition the trades by rule compliance, not by outcome, then compute expectancy on each subset separately.

That third row is uncomfortable and it is the reason to run the cut honestly. A student who is quietly outperforming the thing you taught them is telling you something about your own rules, and the only way to hear it is to have run the partition before you knew which way it would come out.

Two constraints make or break this. First, compliance has to be recorded when the trade is logged, not reconstructed at review time. Nobody can honestly answer "did I follow my rules?" about a trade from three weeks ago while looking at whether it worked. The answer will be contaminated by the outcome every time. This is what a pre-trade entry checklist is for: in SignalDeck the checklist items attach to the individual trade, and a per-user preference can hard-gate the Log Trade button until every item is ticked, so the record is made at the only moment it can be made truthfully. The post-mortem fields on a closed trade - setup quality, stop management, verdict, training gaps, and a free-text mentor feedback field - carry the rest.

Second, both subsets will be small. Splitting sixty trades into forty-two compliant and eighteen non-compliant leaves you with two noisy averages, and the difference between them is noisier still. This cut is directional evidence for a conversation, not a verdict. It earns its place because the alternative - guessing - has no error bars at all.

Risk Adherence Is the Leading Indicator

Students do not usually blow up because their strategy stopped working. They blow up because their position size stopped being a calculation and started being a reaction. That failure arrives months before the edge fails, and catching it early is most of the job.

Here is the part that surprises people, and it is the same arithmetic that makes R-multiples so useful everywhere else: the R series cannot see position-size drift at all.

An R-multiple is the outcome divided by the risk taken on that trade. A student who plans to risk 1% and takes a full stop-out logs −1R. A student who sizes up to 3% on the same setup, in a fit of impatience, and takes a full stop-out also logs −1R. The normalisation that makes trades comparable is the same normalisation that erases the single most dangerous thing the student did.

Four identical-looking losses in the R column. Nearly double the planned damage in the column that decides whether the account survives. If your review reads only the R series, this student looks like someone having a normal bad run, and you will find out otherwise when the drawdown limit is breached.

So track planned risk as a percentage of equity as its own series, and read two things from it. The first is its dispersion: a student running fixed-R position sizing should produce an almost flat line, and any visible spread is the finding. The second is its correlation with the previous trade's outcome, which is where the diagnosis lives. Sizing up after a loss is revenge. Sizing up after a win is overconfidence. Sizing down after a loss is fear, and it is the one that quietly caps a student's upside while looking like prudence. All three are invisible anywhere except this column.

One related leak worth checking in the same pass: whether the stop that defines R is the one the student set before entry. R is anchored to the original stop, so a student who widens a stop mid-trade has increased their real risk without the R series recording it. A journal that keeps the original stop as the anchor will show that as a trade whose realised loss exceeds −1R, which is a clean, searchable signature for a habit that is otherwise very hard to catch.

Building a Shared Setup Taxonomy

Everything above assumes a student's labels mean what you think they mean. Across a cohort, that assumption fails immediately unless someone designs against it.

Leave the setup field open as free text and within a few weeks it contains four spellings of the same pattern, a couple of single-letter entries made to get past the form, and at least one misspelling that will silently split a setup's statistics in two forever. Every per-setup number you compute after that is wrong in a way that does not announce itself.

A taxonomy that survives a cohort has four properties:

• Small. Five to nine setups. A student choosing between twenty-five labels is not classifying, they are guessing, and the sample per setup never reaches a readable size anyway.
• Mutually exclusive. If two labels can both apply to one trade, students will split inconsistently and per-setup expectancy becomes uninterpretable.
• Named after the trigger, not the quality. "Bull Flag Breakout" is a taxonomy. "A+ Setup" is a grade, it is assigned after the outcome is known, and it will correlate beautifully with results while telling you precisely nothing.
• Defined by a written entry condition the student can check against the chart before entering. If two of your students would label the same chart differently, the definition is not finished.

It helps to separate the two kinds of label. One strategy per trade - the thing being scored, drawn from a closed list you control - and as many tags as are useful for the conditions around it: session, news proximity, emotional state, whether it was a re-entry. SignalDeck models them exactly this way, one strategy and many tags per trade, because they answer different questions: the strategy label is what per-setup expectancy is computed on, and the tags are what you slice it by.

Fix the taxonomy before the cohort starts. Retagging three hundred historical trades is the task that never gets done, and a taxonomy changed mid-programme gives you two incomparable halves of a dataset instead of one.

Reviewing a Cohort Rather Than an Individual

Once every student's trades carry the same setup labels and the same R normalisation, the most valuable view stops being per-student and becomes per-setup across the whole cohort.

Suppose eleven of your twelve students have negative expectancy on the same breakout setup. Reviewed one student at a time, that is eleven separate conversations about discipline, and eleven students who conclude they are the problem. Reviewed across the cohort, it is one finding about your teaching - or about the regime the market has been in since you taught it.

The shape of the distribution tells you which:

• Uniformly bad across students points at the instruction or the regime. Individual skill does not usually fail in lockstep.
• Wide and mixed points at execution difficulty: some students have it and others do not, which means the setup is teachable but the teaching is not landing evenly. That is a different intervention, aimed at different people.

SignalDeck's team view was built for this shape of question. A team gets a shared trade feed, a leaderboard, per-member monthly R contribution, shared post-mortems, per-member journaling streaks and a team-level streak. The journaling streak is worth more attention than it sounds: the student who stops logging is the one you most need to see, and their absence from the data is itself the earliest signal you will get.

Verified Track Records for Graduating Students

The last thing a good programme produces is a student who can prove something about themselves to somebody else - a prop firm, an allocator, a first client. This is where the screenshot problem comes back in its most expensive form.

A track record is credible in proportion to what its publisher could not choose. A screenshot of a P&L figure is a claim with no such property: it is selected, un-auditable, and trivially the best day of the year. A profile computed from a complete trade log is a stronger claim, because the numbers that matter are derived rather than asserted - the trade count, the win rate, the equity curve, and above all the maximum drawdown, which is the number nobody would ever volunteer.

A student on SignalDeck can enable a public profile hosted on sgnldk.com and share the link. The owner controls three things and only three: whether the profile is on at all, which strategies appear on it, and whether dollar P&L is hidden while the R-multiples stay visible. That last option is genuinely useful - account size is nobody's business and R-multiples are the more informative number anyway.

The honesty constraint is the second of those three, and it has to be said plainly: because the owner can filter the profile down to selected strategies, a public profile is a verified claim about the strategies shown, not an audit of the whole account. That is a real limitation of using one as proof, and a post about cherry-picking would be a poor place to pretend otherwise. Two practical consequences follow. If you are reading someone's profile, check the trade count and the period before you treat it as evidence. If you are a student publishing one, publish the account unfiltered - a profile that shows the losing strategy alongside the winning one is worth several times more than one that does not, for exactly the reason this whole post exists.

Two administrative notes. Public profiles are a paid-tier feature, and the page returns a 404 if the subscription lapses - so a link on a CV can go dead. And the profile belongs to the student, not to the programme, which is the correct arrangement: it is portable evidence they keep, which makes it worth more to them and, indirectly, a better advertisement for you than any testimonial you could write.

A graduating student with two hundred logged trades, a visible drawdown, a stable tag distribution and an honest record of the trades that did not work is a better result than one with a screenshot of an 8R day. It is also considerably harder to fake, which is the entire point.

None of the above requires a student to journal more than they already do. It requires the journal to be complete rather than curated, the labels to be shared rather than personal, and the planned risk to be written down before the trade rather than inferred after it. Those three properties are what turn a pile of student screenshots into something you can actually coach from.