To avoid some real noise in indicators we are
applying a outlier-detection technique (1.5xIQR)
and proceeding to normalize them using a pseudo avg
that will fit always within the expected values.
Note we don't apply the outlier algorithm recursively
neither any other detail. Just 1-pass normalization
of noise values.
While 1.5 is a generally accepted value, if we detect
that too much is being normalized, we can always go up
to 2.3 to just normalize "very rare" outliers.
By default, it's disabled, so everything continues
working 100% the same. Only under demand it will become
enabled.
You can enable it by adding n=1 or n=true to the url.
At the same time, a bunch of rounding errors have been
fixed, display is always with 2 decimal digits but all
average calculations are performed with full resolution.
To avoid some real noise in indicators we are applying a outlier-detection technique (1.5xIQR) and proceeding to normalize them using a pseudo avg that will fit always within the expected values.
Note we don't apply the outlier algorithm recursively neither any other detail. Just 1-pass normalization of noise values.
While 1.5 is a generally accepted value, if we detect that too much is being normalized, we can always go up to 2.3 to just normalize "very rare" outliers.
By default, it's disabled, so everything continues working 100% the same. Only under demand it will become enabled.
You can enable it by adding n=1 or n=true to the url.
At the same time, a bunch of rounding errors have been fixed, display is always with 2 decimal digits but all average calculations are performed with full resolution.
Other small changes are: