DominikKamp

This is a new issue because it is a dual fail with consistent numerics settings and without assertion fail, so thank you!

An alternative fix is available with f56365291834c487edba45abce4b0038d5ebe137.

Solution read correctly, variable names should just not contain whitespaces because they are required as delimiters.

Sounds dangerous if there is a comment behind containing `(obj:...)`. So we never know which of the `(obj:...)` are part of a variable name. And LP files would still be...

I just believe that supporting whitespaces goes along with a lot of painful code, which will not resolve that standard file formats do not support them. One could also think...

Triggered by disabled scaling in SoPlex as explained in https://github.com/scipopt/scip/issues/135#issuecomment-2842957642.

Here is a simplified instance by [MIP-DD](https://github.com/scipopt/MIP-DD) for SCIP `v92-bugfix`. [seed_17.zip](https://github.com/user-attachments/files/18670401/seed_17.zip) The issue seems to lie in linear presolving since it occurs for multiple heuristics.

As already mentioned, it is not the shifting heuristic because it also fails on the simplified problem with only `heuristics/shifting/freq = -1` so rather try `presolving/maxrounds = 0` as reliable...

The presolved problem is actually correct and solved tolerably so this is really only due to tolerances relative to increased sides caused by fixings. There is already an unmerged approach...

It is usually hard to reproduce LP results from scratch also due to internal numerical deviations up to a warm start. Enabling PaPILO (presolving/milp/maxrounds = -1) works around this issue....