TextInsightBench / protocol.json
erwinmsmith's picture
Rebuild open exploration tasks and corpus-grounded evaluation
149da17 verified
Raw
History Blame Contribute Delete
1.07 kB
{
"version": "textinsightbench",
"evaluation_unit": "evidence-backed downstream finding",
"discovery_mode": "agent_selected",
"scoring_version": "finding-quality-discovery",
"task_families": {
"group_difference": 20,
"temporal_change": 15,
"compound_association": 15
},
"tracks": {
"task_only": "Use the supplied task corpus.",
"unlabeled_pool": "Learn from the released pool before evaluation."
},
"reference_policy": "No fixed conclusions for redesigned tasks; previous references belong only to historical tasks.",
"semantic_review": "Bounded assignment-stratified and corpus-wide sample; exhaustive arithmetic, not exhaustive semantic verification.",
"independent_validation": false,
"historical_exposure": "Task documents were drawn from a previously public learning pool. Not an unseen or independent validation set.",
"cross_task_adaptation": "Freeze global prompts, learned parameters and thresholds; no evaluation-feedback transfer.",
"security": "Untrusted corpus content. Process runner is not a sandbox."
}