What does a bar hide?
A bar records its open, high, low and close and discards the path between them, so for any trade whose stop and target were both inside the bar's range the bar cannot say which was reached first.
That ambiguity is the entire intrabar problem. A long trade entered at the open of a bar with a stop below and a target above has a known outcome if the bar reached only one of them. If the bar reached both, the trade won if the high came first and lost if the low did, and the bar contains no information about the order. The tester has to assume, and the assumption is invisible in the result.
How often this happens depends on the ratio of the stop and target distances to the bar's typical range. On a daily chart with a stop of one ATR, the entry bar rarely contains both levels. On a five-minute chart with a stop of a few pips, it happens on a large fraction of trades, and the assumption decides the strategy's apparent edge.
Finer data does not remove the ambiguity; it shrinks it. A minute bar inside an hourly bar has the same problem at a sixtieth of the scale, and a tick record removes it only down to the resolution at which quotes were captured.
When is minute data enough?
Minute data is enough when the strategy's stop and target distances are large relative to a minute's range, which for most rules on hourly and higher timeframes means the minute record settles the order of events on nearly every ambiguous trade.
The arithmetic is about ratios. If a strategy's stop is one hourly ATR and the entry bar's range spans both stop and target, the sixty minute bars inside that hour will usually show the levels being reached in separate minutes, and the order is established. The residual ambiguity is the handful of trades where both levels fall inside a single minute, and on a one-ATR stop those are rare.
Minute data also covers the case that matters for session rules, opening ranges and time-based exits, where the question is not only which level came first but at what time within the hour an event occurred. A bar-based test cannot place an event inside the bar at all.
The cost of minute data is volume. A decade of one-minute bars across thirty instruments is tens of gigabytes, and the archive has to be checked for gaps, spikes and timezone drift before it is trusted, because a minute record with a bad hour in it is worse than an hourly record that was clean.
When do you need ticks?
Tick data is needed when the strategy's stop or target is comparable to a minute's range, which is the case for scalping rules, for very tight stops on any timeframe, and for rules whose entry and stop are placed inside the same short burst of movement.
For those rules the minute bar is as ambiguous as the hourly bar was for a swing rule. A scalp with a three-pip stop and a five-pip target on a pair that moves four pips in a busy minute has both levels inside the entry minute regularly, and only the tick sequence can order them. A test of such a rule on minute data is guessing on a large share of its trades.
Tick data has its own problems. It is broker-specific, since each broker's feed is its own set of quotes; it is enormous; and it usually records only the bid or only one side, so the spread at each moment has to be reconstructed or assumed. A tick backtest is more precise about order and less certain about the price actually available, and the second uncertainty can be as large as the first.
For most rules that are not scalps, ticks add precision on a small number of trades at a large cost in data handling, and minute data with an honest rule for the residual ambiguity is the better trade.
What should a backtest do when it cannot know?
When the data cannot establish the order of events, the backtest should book the loss, label the trade as unresolved, and report how many trades were decided that way, so the reader can see how much of the result rests on assumption.
Booking the loss is the pessimistic choice and it is the only defensible one, because the alternative, booking the win, produces a result that improves as the data gets worse. A strategy whose apparent edge grows when tested on coarser bars is a strategy whose edge is the assumption. The pessimistic rule makes coarser data hurt, which is the correct direction.
Labelling is what makes the pessimism honest rather than merely harsh. A result that says a quarter of its trades were resolved by assumption is a result the reader can weigh; a result that hides the count is either flattering or punishing the strategy by an unknown amount.
The count itself is diagnostic. A rule with a high unresolved share on the available data is a rule that needs finer data before it can be graded, and that is a useful thing to learn before trading it.
How does QuantParadox resolve intrabar events?
QuantParadox grades on a decade of one-minute history across thirty instruments, walks the minute bars inside any bar that contains both a trade's stop and its target to establish which was reached first, and where a single minute contains both levels it books the loss and labels the trade as unresolved rather than assuming.
The label is per trade and the count is reported with the result, so a rule that leans on the assumption is visible as such. A rule whose trades were nearly all resolved by the minute record has a result that rests on data; a rule with a large unresolved share has a result that rests on the pessimistic rule, and the platform says which.
The limitation is the absence of a tick archive. Rules whose stop and target regularly fall inside one minute, which is most scalping rules, remain pessimistic on the platform, and their results should be read as a floor rather than as a grade. That is stated rather than hidden, and it is the reason the platform is a poor fit for testing scalps and a good fit for everything above them.
Session and time-based rules use the minute record to place events within the hour, which is what makes opening-range and clock-window rules gradable at all on a platform whose entry timeframe may be hourly.