Most pre-market routines prepare the market. Few warm up the reader.

Search for a pre-market routine and the advice converges fast. Check the index futures. Read the overnight news and the macro calendar. Build a watchlist of three to five names rather than twenty you barely know. Mark your levels. Set your alerts. Then settle yourself — breathing, a walk, coffee, ten minutes of something calm.

It is good advice. We are not about to argue with any of it.

But look at what those two halves have in common. One half gathers information. The other half adjusts your state. Neither of them is a repetition of the thing the day is about to ask you to do, which is look at a chart and commit to a direction.

A warm-up usually contains a repetition#

In many skilled activities it does. A pianist plays scales before a concert — the movement itself, slowly, at no stakes. A goalkeeper faces shots before facing shots that count. The briefing and the rehearsal are both there, and they are different things.

The warm-up is a low-stakes performance of the skill; the briefing is information about the conditions. Both are useful. Only one of them is practice.

Many published pre-market routines are heavy on the first two categories — gathering information and settling state — and light on the third. What is often missing is a low-stakes repetition of the chart-reading decision itself, before the one that counts.

That gap is what our warm-up mode is for: twenty real historical charts before the open, long, short or pass, about two minutes. The first read of your day happens where being wrong is free.

Then the arithmetic told us what it must not do#

Here is where this stops being a nice idea and starts being a design problem.

The obvious feature to build on top of twenty calls is a score. Everyone wants to know whether they are sharp today. Every product instinct says: give them a number, colour it, compare it with yesterday, call it a readiness reading.

Twenty calls do not earn us the right to call that number readiness, and one line of arithmetic is why.

The standard deviation of a proportion is sqrt(p(1-p)/n). For a reader whose true rate is exactly 50%, over twenty calls, that is 11.2 percentage points.

In practical terms: an unchanged 50% reader will commonly produce twenty-call sessions around 40% or 60% from sampling noise alone. ⚠️ The band is approximate in a second way as well — twenty calls can only land on multiples of five points, so "39%" describes the spread rather than a score anyone will be shown.

There is a blunter version of the same fact. A reader with no skill at all scores 60% or better on twenty balanced calls in about one session out of four:

So "I got 60% this morning" cannot honestly mean ready. One morning in four says it about someone who is guessing.

It gets worse for the reader who is genuinely good. Take someone whose true rate is 61% — well above chance. Over twenty calls their one-sigma band is 50% to 72%. On a completely ordinary day, that reader has a substantial chance of being shown a score starting with a five. A tool that tells a competent reader they are off form, on a day when nothing was wrong, has done real damage for no informational gain.

⚠️ And Pass makes this worse rather than better. The number of directional calls is chosen by the reader, so someone who skips ten questions has a denominator of ten, where one sigma is 15.8 points. The more selective you are — which is the behaviour we want — the noisier any score becomes.

What it does instead#

Five rules follow from that, and they are constraints rather than preferences.

It observes and describes; it does not judge. No "Sharp", no "Off Rhythm", no state verdict, in any language.

It does not synthesise a score. Correctness, pace and direction are shown separately and are never combined into a single readiness number, because combining them would hide the sample size of each.

Where results are shown, they are printed as counts and nothing else — no verdict, no comparison with last time, no colour coding, no arrows.

Nothing is revealed until the round is over. This one is not about statistics: telling you the answer to card three contaminates your reading of cards four through twenty. What the round is for is twenty calls made without earlier answers influencing later ones — uncontaminated rather than statistically independent, a distinction we have had to learn the hard way.

And there is no leaderboard. A warm-up is a mirror, not a competition. Score it publicly and people start answering for the score rather than answering honestly, which breaks the only thing it was for.

The 81% that killed the clever version#

The tempting compromise is to skip the score and instead surface an observation — "you were slower on the shorts today", that kind of thing. Scan the session, find the most interesting pattern, say it.

We simulated what that costs before building it. Under a Monte Carlo model of a reader choosing directions at random, with a 10% pass rate, a threshold of "at least four instances, and at least 75% of them one way" would have produced at least one observation in about 81% of rounds. Four rounds in five, a reader with no pattern at all would be handed a sentence about their pattern.

The rule set we shipped fires in about 16.5% of simulated rounds under the same model. Most rounds get a neutral fallback, deliberately. ⚠️ Both figures are simulation references under one specific model of behaviour, not predictions of what real readers will see.

And the rules are pre-specified: six of them, fixed, in a fixed priority order, at most one shown per round. Nothing scans the session at runtime for its most striking subset, and nothing generative writes the sentence. That is the same discipline as not choosing your test after seeing the data — an observation selected because it was the most dramatic one available is a comment about noise wearing the costume of insight.

Pass is a decision, not a mistake#

The third button matters more than it looks.

If a chart does not tell you anything, the correct answer is to skip it — that is the judgement being trained. So a skipped question is stored as NULL, not as false, with a database constraint enforcing it. That is not fastidiousness about types. It is what structurally prevents any shared history table in the product from recording a Pass as a wrong answer — the kind of thing that decays into a red badge two years later if it is left as a convention.

What we left out of our own statistics#

The last decision is the one we would most like to have made differently.

Warm-up answers count toward the training journal's volume — questions answered, days in a row, charts read. They are deliberately excluded from the journal's hit rate and from its decision-time median.

The reason is the same arithmetic, plus one more: the warm-up runs on a five-second timer and the other modes do not, so mixing their decision times would produce a median that describes neither. It would have been easy, and slightly flattering, to fold everything into one number. That number would have meant less than either of the two it replaced.

What this does not claim#

Not that a two-minute warm-up prepares you for a trading day — nothing here measures that, and the question of whether practice like this transfers at all is one we have written down but cannot answer.

Not that the standard routine is wrong. Futures, news, levels and a short watchlist are all worth doing, and none of them is replaced by twenty charts.

And not that scoring a short session is impossible in principle — only that at these sample sizes it is dishonest. The threshold where a session score starts carrying information is much further away than it feels.

The claim is narrow: a routine that contains no repetitions of the skill is a briefing, not a warm-up — and a warm-up short enough to do before the bell is, by the same arithmetic, too short to grade you.

SwipeTA is a training game and a simulation: no real money, no broker, and it does not provide investment advice. The parameters behind our measurements are on the methodology page.