Filter Criteria That Survive a Real Morning

A scan built in the evening is built under ideal conditions. The data is complete, there is time to think, and every criterion can be checked against yesterday and found reasonable. The same scan run against a live morning is operating on partial data, under time pressure, with results that change between one refresh and the next. Some criteria survive that transition unchanged and some fall apart, and the difference is not obvious from reading the criteria.

Criteria That Depend on Settled Data

The most robust filters are the ones built on information that was already final before the session opened. Average range over recent weeks, typical volume, the price band an instrument has been trading in, whether it is liquid enough to enter and exit without difficulty. None of these change during the morning and none of them can give a different answer depending on when the scan is run.

These criteria do the heavy lifting and get the least attention, because they are dull. They also define the universe within which everything else operates, and a universe defined badly cannot be rescued by clever conditions applied on top of it. An instrument that is too thin to trade properly does not become tradeable because it appeared on a list.

Criteria That Need Time to Mean Anything

The fragile group is anything defined relative to today. Volume compared to a typical day at the same time, movement compared to a normal session, position within the day's range. All of these are perfectly sound once enough of the session has elapsed and all of them are close to meaningless in the first minutes.

The failure is quiet. Early in a session almost anything will look like unusual volume relative to a full day, so the criterion passes nearly everything and the scan appears to be finding a great deal. It is not finding anything. It is reporting that the day has started. Criteria of this kind need a stated earliest time at which they become valid, and that time belongs in the scan definition rather than in memory.

Criteria That Are Really Preferences

A third group is neither robust nor fragile because it is not measuring anything. Filters that encode a preference for a sector, an aversion to instruments that behaved badly last week, or a rule added after a single memorable loss all fall here. They reduce the list, which feels like progress, and they reduce it on grounds unconnected to how the setup performs.

These accumulate over time and are hard to remove, because each one has a story attached. The test is whether the criterion could be stated to someone else as a general principle without reference to a specific past trade. If it cannot, it is a scar rather than a filter.

How Many Is Too Many

Each additional criterion narrows the output, and the narrowing compounds faster than intuition suggests. A scan with a handful of reasonable conditions can go from a workable list to nothing at all with one more, and the one more usually gets added on a morning when the list felt too long.

The more subtle cost is that a heavily filtered scan produces very few candidates over any given month, which means very few observations, which means it takes a long time to find out whether the criteria are any good. A looser scan that produces more candidates than can be traded at least generates enough history to be judged.

Testing a Criterion Properly

The question to ask of any filter is what it removes, not what it keeps. Run the scan with the criterion and without it, and look at the names that appear only in the wider version. If those names would clearly have been poor candidates, the criterion is doing something. If they look much like the ones that survived, it is only shortening the list.

This has to be done on a stretch of sessions rather than one, and it has to include the quiet mornings as well as the active ones. A criterion that works only on busy days is a criterion that will hand you an empty list on precisely the days when standing aside was already the right answer, and will do nothing on the days a decision was actually needed.