100,110 rows. Decontaminated version of DCAgent/g1_min_episodes_top8_100k_glm47_traces — the dataset originally used to train DCAgent2/g1_top8_100k_32b_optcfg.

What was changed

The source dataset had 2,442 rows in the issue subset whose harbor task derives from a SWE-bench Verified instance (459 of 500 unique Verified instance_ids leaked through into training). Those 2,442 rows were:

Removed from the dataset. Replaced 1-for-1… See the full description on the dataset page: https://huggingface.co/datasets/DCAgent2/g1_top8_100k_decontam_glm47_traces.