WIP<\/em> My attempt at pulling useful information out of a large dataset.<\/p>\n High-level loop:<\/p>\n artifact -> frontier -> bounded claim -> probe family -> evidence bundle -> route -> reasoner -> confidence update -> next-evidence request -> validation -> frontier advance<\/p>\n What it is trying to do is take stage-1 corpus outputs and turn them into a controlled evidence loop: pick one bounded question, gather support or contradiction around it, decide what evidence to ask for next, and only move forward when the result holds up.<\/p>\n This HF link is one visible slice of the larger system, not the whole thing. Here it shows repeated bridge and contrast structure plus one stable rare finding across follow-up validation. <\/p>\n partial artifact, more on the link below <\/p>\n “cluster_ids”: [ 406 ], “member_count”: 1, “observation_count”: 3, “source_bundle_ids”: [ “tier1_bundle_mixed_region_0410c17c2515bf49” ], “source_result_ids”: [ “finding:406” ], “primary_terms”: [ “anisotropy”, “coherent”, “high-confidence”, “rare”, “centered”, “gmims-related”, “text”, “magnetic”, “medium”, “film” ], “supporting_results”: [ { “result_type”: “finding”, “cluster_ids”: [ 406 ], “bundle_id”: “tier1_bundle_mixed_region_0410c17c2515bf49”, “summary”: “Cluster 406 is a coherent, high-confidence rare cluster centered on GMIMS-related text about anisotropy, magnetic medium, and film\/samples, with 22 rows and a strong mean probability of 0.814.”, “confidence”: 0.76608, “support_cycles”: 3, “next_probe_name”: “neighbor_cluster_comparison”, “next_probe_cluster_ids”: [ 406 ], “next_probe_reason”: “The card lists centroid neighbors 285 and 282 with very high cosine similarity, making them the most direct follow-up for checking whether the GMIMS\/anisotropy pattern extends to adjacent clusters.” } ],<\/p>\n link to hf with artifacts : https:\/\/huggingface.co\/datasets\/cjc0013\/reasoningovercorpusartifacts\/tree\/main<\/a><\/p>\n link to previous dataset post : https:\/\/old.reddit.com\/r\/datasets\/comments\/1sej8ro\/fused_patent_arxiv_clustering_dataset_9m_raw_388m\/<\/a><\/p>\n<\/div>\n submitted by \/u\/Either_Pound1986 <\/a> <\/h1>\n
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
