Advice on a strong GraphBLAS recomputation baseline for VeloGraphX #633

Hi GraphBLAS community,

I’m working on VeloGraphX, a C++20 research engine for exact dynamic graph analytics:

https://github.com/sauravsingla/VeloGraphX

One of our main evaluation questions is when exact localized incremental repair becomes preferable to a strong optimized full recomputation baseline.

I would like to use GraphBLAS/LAGraph in a publication-quality same-hardware comparison and would appreciate guidance on the fairest setup.

Specifically:

• Which BFS/SSSP implementation should be used?
• Which SuiteSparse:GraphBLAS/LAGraph version and configuration would you recommend?
• How should threading be configured?
• Which preprocessing should be excluded or included in measured latency?
• Are there common benchmarking mistakes we should avoid?
• What would you expect to see before considering such a comparison scientifically fair?

The goal is to identify the true crossover between incremental repair and optimized recomputation rather than produce a favourable headline speedup.

Any criticism or recommended experimental setup would be very valuable.

Thank you.