Values, pluralism and paradigm physiology
Contested-direction loading predicts human rater disagreement
inconclusive registered
The reading
-
contested-direction loading predicts per-item human rater disagreement
contested_raterspread_spearman > 0.2 0.55 not measured inconclusive
-
single-model contested loading comes within 0.05 of the 5x-cost ensemble baseline
ensemble_minus_contested_spearman < 0.05 not measured inconclusive
Provenance
- card
- VALUES-CONTEST
- frozen_at
- adjudicated_at
- spec_hash
- spec:2a1ead40948d8a58ae00bf095cad321f
- study_id
- study:campaign-values-contest@v1#2a1ead40
- Evidence records
- e036326c47c7a0f52330559479e03790
- spec_file
- campaign-values-contest.json