Research · Candlesticks
The short answer
How often a formation appears is a property of the thresholds used to detect it, which is why published counts for the same pattern differ by multiples.
The per-formation frequency figures came from the same export as the hit rates, and they were withdrawn with it. They were computed on a superseded engine, and the candlestick rows included the still-forming bar.
They will return with the re-run. In the meantime it is worth being clear about what such a figure even is, because frequency is the statistic most often quoted and least often qualified.
How often a formation appears depends entirely on the parameters used to detect it. A hammer requires a body no larger than a third of the range and a lower wick at least half of it. Loosen those to 0.4 and 0.45 and the count rises substantially; tighten them and it falls.
This is why published occurrence counts for the same formation differ by multiples between sources. They are not measuring different markets; they are using different definitions and not saying so.
Worth knowing
Every formation page here publishes its thresholds and its strict and loose variants, so a frequency quoted from this data can be reproduced. That is the part that makes the number meaningful — not the number.
Classical definitions of most reversal formations require a prior trend — a hammer after a decline, a hanging man after a rise. Applying that requirement typically discards the large majority of raw shape matches.
Most published occurrence counts quietly drop it, which is why their numbers are several times larger. A count that ignores context is counting a different formation from the one the tradition describes.
A formation appearing on one bar in eight cannot carry much information, whatever it is called: if it is present that often, its presence barely narrows the range of what happens next. A formation appearing a handful of times across a decade cannot support a rate at all — the interval around it is wider than any effect it could have.
Useful formations live between those extremes, and where the boundaries fall is an empirical question that needs the measured run. Until it lands, the candlestick patterns index carries the rules the counts will be counts of.
If you want a number rather than a quoted one, the calculation is simple and the honesty is in the disclosures around it.
A formation with tens of thousands of detections does not give you tens of thousands of independent observations of what follows it. Detections cluster, and clustered detections share forward windows.
For frequency this does not matter — you are counting occurrences, and overlapping ones are still occurrences. For any statement about what happened afterwards it matters enormously, because the effective sample can be a third of the raw count.
Worth knowing
This is the reason a frequency and a hit rate cannot be quoted with the same denominator, even when they come from the same run. They are counting different things.
When a frequency is quoted anywhere — here or elsewhere — two questions separate a usable number from a decorative one.
A source that cannot answer both is quoting a number it did not compute. That is common enough that the absence should be treated as informative rather than as an oversight.