GroundLM: all covered results

claude-experiments.zip: all six earlier Claude runs (360 slots), with exact saved responses, scoring, settings and reports.
family_comparison.zip: 456 new model slots plus 48 Claude references copied from replication01.
groundlm_hard_v1.zip: 4800 slots on the fresh Hard v1 dataset.
paper: earlier reported study; original raw generations were unavailable in the checkout.

There are 5616 collected inference slots across the saved runs. The 48 reused Claude slots are already in the 360 Claude slots; do not add them again. Failed provider attempts and token cutoffs remain explicit. Studies differ in tasks, seeds, models, reasoning and token limits; do not pool these into one capability ranking.
