I suspected this behavior early, but now that I actually started recording, I finally caught the "smoking gun". After playing a while, some (not all) voices in fixed track of arpeggio somehow start to change pitch. I haven't isolated what exactly is causing pitch change: sometimes it happens when a chord is played, sometimes when I accidentally slip my finger onto other key.

Seems, perhaps, your understanding of how the Arpeggio Convert Types work is a bit faulty. There is no bug.

The Convert Type is used when you initially setup to create the User Arp. It is not something you set on the Part selected to play it back.

MIDI data that is converted to an Arp is done by selecting data in the MODX sequencer.
If you set the Convert Type = Fixed, the Arp Phrase will playback exactly the notes you input.

You seem to be saying that for playback that you can set a track to be fixed... no, that’s not how it works. The Arp Phrase is either Fixed, OrgNote, or Normal... that is determined when you first create it.

Once you convert MIDI data to an Arp, nothing alerts you to the nature of the Arp Type except the Name, Main, and SubCategory that you define for it.

So if you didn’t use Convert Type = Fixed, the only way anyone would know what Convert Type was used is to observe the Phrase’s behavior when assigned to a Part and triggeredwith the keyboard. This is why naming and setting the attributes of your User Phrases is very important.
This is why posting an example file will tell use nothing about HOW the use Arps were created. There is no way to tell, after-the-fact.

Seems, perhaps, your understanding of how the Arpeggio Convert Types work is a bit faulty. There is no bug.

Sorry, but yes, there is: the track type is Fixed - I just recreated the arpeggio last night, and it seems to be coonsistent behaviour (in fact, the source sequence is included as part of the backup as Pattern 007).
The building block of the "wavesequence" I created is a drum voice/part, where some elements are mapped to fixed note for a drum-like pattern (bangs, pings and other crap), and the rest is mapped chromatically - one sample across across one octave. I then created a two track arpeggio: one is Fixed - the drum part, the other is Normal - that's the bass part of wavesequence. It works exactly as advertised, but over time and various sessions, I started noticing some oddity in drum part. The notes transmitted match, the key matches, but pitch of some drum samples is shifting.

This is why posting an example file will tell use nothing about HOW the use Arps were created. There is no way to tell, after-the-fact.

Isn't that a bit of a design flaw?

Isn't that a bit of a design flaw?

No, it’s not.

Your mistake: “I then created a two track arpeggio: one is Fixed - the drum part, the other is Normal - that's the bass part of wavesequence.”
You can argue with me or you can learn how the Arpeggiators work... there are fourTracks in the Convert To Arp tool. All four Tracks must be applied to the the same sound — they are used to create one Arpeggio phrase to be played by one Part (only). The articles on Arpeggio making go over how the four Tracks can be used... you need to review these articles.

__ Arpeggio Making 101 Part 1
__ Arpeggio Making 101 Part 2
__ The 4-Track Arpeggio — 4 different Tracks each recorded at different velocities to trigger specific Elements. One use of the 4-Track Arp maker.
__ MONTAGifying Motif XF: Free Fall — another example of how the 4-Tracks use “OrgNote“ Convert Type for the chords, and “Fixed Note” for the finger zings and string raking noise.

These articles will serve as a resource about Arpeggio making, and the last two go in-depth into examples of the use of the 4-Track Arpeggio maker... which trust me, you are misunderstanding.

OK, just answer me one thing: is "fixed" mode arpeggio track ever supposed to change pitch regardless of the notes played - especially in a drum set?

And to get you a clear(er) idea: this only happens when "Hold" is enabled.

No, not ever! Drum Set or Sound FX, it matters not - an Arpeggio created as “Fixed” Note will always play exactly the Notes that were originally played. Exactly.

Now let me ask you ... what is a “‘fixed’ mode Arpeggio Track”?
The fact that you think there can be one is wrong thinking

You assign an Arp to a Part.
The Arp that was recorded using Fixed Notes as the Convert Type will always play the exact same notes, with the exact same timing no matter to what Part you assign it.

Now let me ask you ... what is a “‘fixed’ mode Arpeggio Track”?

My shorthand for: An arpeggiator track imported with "Fixed" convert type.

But, the "bug" is solved! And I laughed - you will probably too. It's not arpeggio: it's the waveform itself! It turns out that some waveforms I chose for the wavesequence (for example "Repique2 St Rim", "Tanak Dom1") are multisamples: velocity dependent. When I recorded music with sequencer, I would not notice it, because I'd normalize note velocity. Turns out the samples for high velocity have slightly different tuning. It might not be apparent at normal audition (although some are audibly different for fff), transposed octaves down - they start to be, even several semitones. That's where I noticed that when performing live and I thought the arpeggiator has a bug.

You can limit arpeggio velocities after-the-fact if you want. I guess it's good to learn how nothing going wrong sounds wrong -- and how you've got more to check or take care of when crafting/using arpeggios. As this sort of thing could happen in non-fixed arpeggios too. Different velocities causing different sounds to come out. Possibly pitched differently.

You can limit arpeggio velocities after-the-fact if you want.

Well, for arpeggios, yes, but the velocity switching is also "hardcoded" into multisample waveforms that use it. So you can imagine my "whaa?" moment, when I discovered that the samples for high velocity were differently pitched.

Different velocities causing different sounds to come out.

I kind of wish Yamaha would give us more control on waveforms and also ability to "extract" a sample to be played inside an element from multisampled waveform. I actually figured out the source of the "bug" while lamenting that the only satisfying male voice sound I would love to use as "lead" within the "WS Memories" performance (shameless plug: try it, I'm really happy about how it came out, even though itself being a "proof-of-concept" thing) was part of a multisample waveform, and worked at low velocity, while for high velocity it was more subdued ("ooh" like).

Again, your mistakes should not be called “bugs” - there is no bug! You can define which sample plays within an Element by changing its KeyBank parameter settings (Note and Velocity Limits) within the Waveform.

Again, your mistakes should not be called “bugs” - there is no bug! You can define which sample plays by changing its KeyBank parameter settings (Note and Velocity Limits).

But I only can do that for user waveforms, I cannot modify factory supplied ones - am I right?