Why SwipeTA is not Duolingo for trading

People describe SwipeTA as Duolingo for trading. Testers reached for the comparison unprompted, more than once, and it is a good instinct: it is a chart-reading practice app built out of short repetitions, immediate feedback, streaks and a daily goal, and that really is the shape of the thing.

But chart practice has a problem language learning does not, and it is the one thing that matters most about building this: the answer key is noisy.

Where the analogy is genuinely right#

Not grudgingly: most of it holds.

Both are built on short repetitions rather than long study sessions. Both give you the outcome immediately instead of at the end of a chapter. Both assume that the way you get good at something is a large number of small encounters, not a small number of large ones. Both use streaks and daily goals to make the small encounters happen at all, because the enemy of any practice tool is not difficulty, it is Tuesday.

And both are aimed at fast recognition rather than explicit knowledge. Fluency is not conjugation tables recalled quickly; it is having met something often enough that the wrong version registers as wrong before you could explain why. That is the sense in which we are interested in chart reading as a skill — repeated exposure may build faster recognition before a reader can fully articulate why a setup looks familiar. It is an analogy we find useful, not a claim that the two are the same mechanism.

If the analogy stopped there we would use it ourselves.

Where it breaks: the answer key#

In a language app, the correct answer exists before the question is asked. Mesa is the Spanish for table. It was the Spanish for table yesterday, it will be tomorrow, and when you get it right you have demonstrated, with certainty, that you knew a thing. One correct answer is one unit of evidence about you.

Our answers do not work like that. The chart is frozen at a decision point and you call which way price went over the next fifteen minutes. That outcome was not a fact about the chart in front of you — it had not happened yet, and much of what decided it was not visible at the decision point and never would have been.

So a correct call does not mean you read the chart correctly. It means you called it and the market happened to agree. Those come apart constantly, in both directions: you can be right for a reason that was not operating, and wrong having read the situation better than the person who was right.

This is the whole difference, and everything below is downstream of it.

Can a random trader look skilled?#

That sounds like a philosophical caveat. It is a measurable one, and the size of it surprises most people.

Consider a player with no skill whatsoever — someone answering at random. Our question bank is a deliberate 50/50 split between up and down, so that player's true accuracy is exactly a coin.

In a twenty-answer session, that player scores 60% or better one session in four. They score 65% or better — thirteen out of twenty — 13% of the time, about one session in eight.

Streaks are worse, because streaks are what people actually remember.

Across a hundred answers, a player who knows nothing hits a five-in-a-row streak 81% of the time. Put that next to a language app: there, five in a row means you knew five things. Here it is the single most likely thing to happen to somebody with no skill at all.

How many answers separate skill from luck?#

Going the other way is worse. Establishing that a genuinely skilled player is skilled takes far more than a session: about 194 answers for a true 60% player, 783 for a 55% player, and 4,904 for a 52% one — which would still be a real edge.

That is the same arithmetic the app uses to decide when to unlock its own statistics, and we charted it in revenge trading, FOMO and the answer that comes right after a wrong one. Pointed at this question it says something blunter: a session cannot tell you whether you are good at this, and neither can a week of them.

So almost every design decision is about the noise#

Once you accept that a single answer carries almost no information, a lot of the app stops looking like game design and starts looking like error correction.

The bank is balanced by construction. The selection is split exactly evenly, and the competitive bank that comes out of it is 10,136 up against 10,164 down. US equities drift upwards, so a bank that merely sampled reality would hand a player who always swipes up a small free edge that has nothing to do with reading. Duolingo does not have to think about this: there is no direction to be biased toward in a vocabulary test.

Beginner chart practice deliberately includes fakeouts. Roughly a third of the beginner pool is questions where the obvious read — fade the last candle — turns out to be wrong. A curriculum made only of readable charts would teach a rule — and that rule then fails in the competitive modes, which reads to the player as the training having lied to them. A language course never faces this, because Spanish does not contain a subset of words designed to punish the pattern you just learned.

The horizon is written when the question is generated, never chosen afterwards. With several horizons stored per question, picking the flattering one would let us manufacture any accuracy we wanted — one of four places look-ahead bias gets in, and the least visible. That specific hazard does not exist for a vocabulary app either — you cannot retroactively decide which meaning of a word you were asking about.

And the statistics panels stay locked. The app will not tell you that you lean long until it has seen enough answers for that to be a real finding rather than a plausible-looking accident. The number it is waiting for is 85 decided answers, which is where the arithmetic lands, not where the product wanted it to land.

None of those are features a language app needs. All of them exist because our answer key is a sample from a noisy process, and theirs is a dictionary.

Is SwipeTA a trading simulator?#

Not in the sense that phrase usually carries, and the distinction is worth stating plainly because it is the one people get wrong most often.

A trading simulator — paper trading — reproduces a broker. There is an order ticket, a position, a size, a stop, a portfolio to manage, and the practice is mostly about operating all of that under pressure. There is no order ticket here, no instrument portfolio and no broker connection.

But it is not sizing-free either, and it would be misleading to imply otherwise. The competitive modes are bankroll games: you start a run with 1,000 points and, for each of the next fifteen or twenty questions, choose what fraction of the stack to commit and a multiplier — and you can bust. How far your balance falls from its peak is part of what the run is scored on.

What is genuinely absent is the exit. Once a call is committed it settles at its fixed horizon whatever happens in between: there is no stop, no target, and no cutting it early. That is a real difference from paper trading rather than a missing feature — what gets isolated here is the read and the size, not the management of an open position.

It is a simulation in the broad sense — historical market situations, no real money, no broker — but the task it isolates is reading, not execution. That narrowness is the point, and it is also the trade: everything paper trading practises about managing a position, this does not touch. We compared the two, along with backtesting, in chart replay vs paper trading vs backtesting.

The asymmetry we cannot design our way out of#

There is one more difference, and it is the one we like least.

Whether a language app works is, in the end, an answerable question. You can teach someone for six months and then test their Spanish, and the test is not controversial.

Whether practising chart reading transfers to real markets is not settled, and we have no evidence that it does. We said the same thing in what the research says about trader intuition: the honest position is that the transfer question is open. SwipeTA is a training game and a simulation — it holds no money, connects to no broker, and does not provide investment advice.

So the comparison flatters us in the one place it should not. "Duolingo for trading" implies the same contract: do the exercises, acquire the skill, use it. We can offer the first part with a straight face. The second is measurable only in the app's own terms, and needs far more answers than anybody expects. The third we do not claim at all.

What is left is smaller and, we think, still worth building: a place where the calls you made are written down before the outcome is known, so that the story you tell yourself afterwards has something to argue with. The rest of the parameters are on our methodology page.