qrs detection learning parameters, 'Failed to find * beats during learning'

I am having problems with the xqrs class.
I have to work with 1000 Hz real ecg signals.

I would like to know why can't the qrs detection learn initial signal parameters.

In my linked jupyter notebook (had to clear plot outputs of size problems), the function fails to learn on a generated perfect 1000 Hz signal and the interpolated 360 Hz version of it, yet the learning works on the interpolated 360 Hz version of the noisy signal but fails on the 1000 Hz version with wrong output.

When I tried to figure it out in the source code, it seemed like when the finding of beats happen, the ricker wavelet gets compared to wrong sample sized signal, so the wavelet might not be scaled up well.
( I might be completly wrong here)

Thanks for any help!

my representation of the issue:
MWE notebook
data for the example:
clear ecg data
noisy ecg data

Start with the linked wfdb_qrs_detection.ipynb and run its xqrs examples against the clear and noisy ECG data at both 1000 Hz and 360 Hz. Trace the learning and beat-finding path, then compare the sample sizes used by the Ricker wavelet and signal; done means the failure and incorrect 1000 Hz result are explained or covered by a reproducible fix.