[BUG] Simplex appears to return status `Optimal` despite violating requested tolerances

Describe the bug
For one big LP model, cuOpt returns Status: Optimal even though the final reported residuals appear inconsistent with the user-specified tolerances.

In this run, I set all main primal / dual / gap tolerances, as well as infeasibility tolerances, to 1e-12. However, the final log reports:

• Primal residual (abs/rel): 2.88e-01 / 6.29e-10
• Dual residual (abs/rel): 9.02e-08 / 4.59e-16

So it is unclear whether I am misunderstanding how these tolerances are enforced, or whether cuOpt is incorrectly reporting optimal termination on this LP instance.

Steps/Code to reproduce bug

1. Unzip the attached file. It is double-compressed to meet GitHub requirements.
2. Run cuopt_cli big_lp_obf.mps with the following tolerances set to 1e-12:
   • --absolute_primal_tolerance
   • --absolute_dual_tolerance
   • --absolute_gap_tolerance
   • --relative_primal_tolerance
   • --relative_dual_tolerance
   • --relative_gap_tolerance
   • --primal_infeasible_tolerance
   • --dual_infeasible_tolerance

Expected behavior
I would expect one of the following:

• cuOpt returns a solution whose reported residuals satisfy the requested tolerances, or
• cuOpt reports that the requested tolerances could not be met, instead of returning Status: Optimal.

Environment

• cuOpt version: 26.2.0
• git hash: f73da24d
• CUDA version: 12.9
• GPU: NVIDIA H200
• CPU: Intel(R) Xeon(R) Platinum 8480+
• RAM: 427.29 GiB

Problem summary

• LP instance
• 424118 constraints
• 1367797 variables
• 4512958 nonzeros

Coefficient ranges reported by cuOpt:

• Objective coefficients: [1e-03, 8e+05]
• Constraint matrix coefficients: [1e-18, 8e+05]
• RHS / bounds: [5e-12, 3e+07]
• Variable bounds: [4e-15, 4e+07]

Additional context
This is the same large LP instance as in the #1045
I understand that the model has serious numerical issues. However, since it is handled by other solvers, I believe this is a solver termination-criteria issue rather than just a model-quality issue.

Attachments

• Full solver log

big_lp_obf.log