Inconsistent "blunder" server analysis for study chapters starting from position.

[Siderite]

Exact URL of where the bug happened

https://lichess.org/study/wvNVfl58/ubiitST3
https://lichess.org/study/wvNVfl58/0jVeylQx
https://lichess.org/study/wvNVfl58/9gMbkk8O

Steps to reproduce the bug

1. Create a study and use a full game PGN for the first chapter
2. Get the PGN of the chapter starting from a specific position (in my case I started from move 7...Nxe5)
3. Create a new chapter with this PGN
4. request server analysis for both

What did you expect to happen?

Same moves marked as mistakes, blunders

What happened instead?

Some blunders did not appear in the from position chapter

Operating system

Windows 11

Browser and version (or alternate access method)

Brave

Additional information

Look at the chapter "Start from white move" (62) and then "Start from further black move" (63)
Look at move 13...Rad8

For the full game, it is marked as a blunder. For the subset, it is not. The server eval is the same for the before and after positions (+0.3 and +2)