Characterize the long-rangedness of existing graph benchmarks

Determine whether the LRGB, GLoRa, and ECHO benchmark suites genuinely require long-range interactions, rather than permitting accurate prediction from shorter-range information or other correlations.

Background

The paper develops four axioms intended to certify that a graph-learning task requires information from a specified range. It applies these axioms to several established benchmarks and identifies violations or limitations, including the possibility that short-range predictors perform well because of dataset-specific correlations.

Although the audit identifies concrete failure modes, it does not establish whether the benchmark tasks are genuinely long-ranged under a principled, topology-independent criterion. This leaves unresolved whether empirical gains on these benchmarks can be attributed to long-range modeling.

References

In conclusion, the long-ranged nature of benchmarks LRGB , GLoRa and ECHO is unknown.

— Get a GRIP, this will be a long TRIP: A Quantifiable Long-Range Framework for Verifying Over-squashing  (2610.03556 - Caralt et al., 2 Oct 2026) in Section 3, Auditing Existing Long-range Benchmarks

We note that, eccentricity, for instance, is the longest shortest path from a node, which is plausibly correlated with the degrees of the immediate neighbors of that node -- but the achievable accuracy from such correlations alone is not characterized in .

— Get a GRIP, this will be a long TRIP: A Quantifiable Long-Range Framework for Verifying Over-squashing  (2610.03556 - Caralt et al., 2 Oct 2026) in Appendix A, Section “Intuition behind the axioms” (also discussed in Section 3, paragraph “ECHO”)

Nevertheless, the example is illustrative rather than a general proof, and we believe there remains a not-yet-fully-understood interplay between homophily and range that merits further investigation.

— Get a GRIP, this will be a long TRIP: A Quantifiable Long-Range Framework for Verifying Over-squashing  (2610.03556 - Caralt et al., 2 Oct 2026) in Appendix, Section “Long-rangedness and homophily”