Percentiles and the P-chip
Distinguish a counted position in the record from a transformed feature value.
A value and its percentile answer different questions: what was measured, and where does that value sit within a specified comparison set?
The counted P-chip
The Research snapshot percentile compares a numeric reading with the same feature’s present buckets over that symbol’s record at the reported revision. It counts buckets strictly below the probe value and divides by all eligible present buckets. Missing values are excluded; ties are not counted as below.
A displayed P97 is a rounded record rank. It does not mean a 97% chance of a future outcome. The comparison can include history after the probed moment, so it is not automatically information available at that historical time.
A feature named pctrank can be different
The book imbalance rank feature applies a normal cumulative distribution function, Phi(z), to a rolling standardized measurement. It is not an exact count of past observations. The OI velocity rank also uses this type of transform.
A transformed feature value of 0.94 therefore does not certify that precisely 94% of recorded values were lower. A P-chip attached to that feature is a separate, counted rank of the feature itself.
Use the comparison carefully
Check the field, symbol, record revision and missing-data coverage before comparing ranks. Extreme ranks identify unusual values within that comparison set; they do not establish usefulness or an edge. State a question, then inspect the outcome distribution and its baseline.
Looking for a calculation? Open the Reading library. For recorded exercises, browse guided lessons.