Anomalies in audio detection

I have been trying to create something in a sort of ambient drone style, and have noticed some anomalies in audio detection, which I have mentioned in a previous post.
So I set up what is a more definitive test to verify the issues.

I created an 8 bar sequence in Live thus

Dropping this into scaler and detecting midi, the result is this

Nothing to argue with there, although there are other interpretations

Next, I rendered this to audio using a pad from Omnisphere, and fired up Scaler Audio. Dropping the WAV into the detect region resulted in this

call scaler audio detect

Something odd here. However, Performing the detection again produced this

call scaler audio  2

Which is weird.

Restoring state “as was” produced repeated the two detection results

I then tried the following sequence

with the first 4 bars being Cm, legato. as expected, midi detection found

Audio detection (inexplicably) found
c legato audio 2

The oddity of running again producing a different result (what I think Einstein described as madness, but then again he never subscribed fully to quantum mechanics) produced
c lagato audio run 2

All in all, odd. I shall investigate further, but so far I conclude that for audio detection

1 it is affected by the attack part of a pad, even though this might be quite small

2 it takes no account of the length of the audio at all, and will detect 4 bars of audio pad without any attack as one bar.

3 it appears to not reset it’s internal state (in spite of warning otherwise) after a detection, and the residual state impacts on a subsequent detection run.

I have alternative means to detect audio key/notes, so it’s not critical for me, but I have provided this for info for the board, and potentially @Ed1 to put on the list to look at for some later update.