How SwipeTA builds and judges its questions

Where the charts come from

Questions are built from historical US equity intraday data at a 5-minute resolution, taken from a licensed historical dataset. Only the regular session is used — 09:30 to 15:59 ET. Pre-market and after-hours bars are excluded, because they trade differently enough that mixing them in would be a different exercise.

Data provided for free by IEX. By accessing or using IEX Historical Data, you agree to the IEX Historical Data Terms of Use.

Where the decision point falls

Each question freezes a session at one bar — the decision point. You see up to 40 bars ending there, which is roughly half a trading day, and never anything after it.

• At least six bars — 30 minutes — of the session must have traded before a question can be asked, so there is always a session to read rather than a single opening candle.
• The bars from the open to the decision point, and the outcome window after it, must be gap-free. A session with a hole in it is skipped rather than patched.
• At most one question is taken per session, keeping that day's most decisive setup. Two questions from the same day would share most of their chart, and the later one's chart would contain the earlier one's answer.

What the question is, and how the answer is judged

The question is always the same: 15 minutes after the decision point, was price higher or lower?

It is judged on the close of the 5-minute candle three bars later, against the close at the decision point. Higher is up, lower is down. An exact tie is recorded as up — at real tick sizes this effectively never happens, but the rule is fixed rather than left open.

No fees, spread, slippage, or leverage enter that judgement. The question is direction and nothing else. Position sizing exists in some modes as a game mechanic, and it changes what a call is worth to your score; it does not change whether the call was right.

Which setups become questions

Not every moment makes a fair question. A candidate is only kept if the move over the answer window is between 1× and 3× the session's recent volatility, measured as a 14-bar average true range at the decision point.

• Below 1×, the outcome is inside the noise — the chart contained no information about it, so the question would be a coin toss dressed up as a skill test.
• Above 3×, the move is a shock rather than a continuation or a reversal, and nothing visible at the decision point was going to tell you it was coming.

Outcome windows are kept disjoint, so no two questions in the bank are decided by overlapping bars and the answers stay independent of one another.

How look-ahead bias is kept out

This is the failure that quietly ruins backtests and chart quizzes: something on the screen was computed with knowledge of what happened next.

Everything the player is shown — the candles, the VWAP line, the session context markers, the tags used to describe the setup — is computed only from bars up to and including the decision point. The bars that decide the answer are held separately and are never an input to anything you can see.

Two things follow from that, and both are deliberate: the session high and low shown to you are the high and low so far, not the day's final ones; and the chart is cropped at the decision point rather than drawn from the full session and masked.

How the bank is kept from being solvable without reading the chart

A question bank can be beaten from either end, and both would make the practice worthless.

• Direction balance. The bank is exactly 50% up and 50% down, and it is filled round-robin across symbols and across times of day. US equities drift upwards over time, so an unbalanced bank would hand a player who always swipes up a free edge that has nothing to do with skill.
• Naive-rule share. One dumb rule — "the last candle reverses" — answers roughly 60% of all candidate questions correctly. That is a property of intraday price action, not a flaw, but it means the proportion of questions where that rule fails has to be controlled. Publication aborts outright if the competitive pool's share drifts more than 5 percentage points from the underlying population.
• Practice is not sanitised. The beginner pool deliberately keeps a share of questions where the naive rule fails. A pool of only readable charts would teach a rule that then fails roughly two times in five in competitive play, which is worse than teaching nothing.

Splits, adjustments and silent corruption

Data vendors adjust history retroactively for splits and dividends. That means regenerating the same question later can legitimately produce different prices — and therefore a different answer — while players' past attempts still point at it.

Every question stores a fingerprint of the exact bars its answer depends on. If a republished question no longer matches, that is detected rather than silently overwriting the answer to a question thousands of people have already played.

Where the answer lives

In the competitive modes, the answer is never on your device. A round is run by a server process that holds the answer for the duration; the app is sent the chart, your call is sent back, and the result is judged server-side. Fields that carry or imply the answer are stripped before anything is published to the systems the app can reach, and the build refuses to proceed if one of them is found where the app could read it.

What a score means — and what it does not

Your accuracy, rating and rank describe how you performed on this bank of historical questions, under a timer, with no money at stake.

They are not a measure of trading ability and we do not claim they predict one. Historical results say nothing about future market behaviour, and nothing in this app involves the costs, execution, position management or emotional load of trading real money. SwipeTA does not provide investment, financial or trading advice.