Data licensing & judgment capture
License expert judgment data, capture your organization’s expertise, or talk to us about your next project.
For AI labs, agent builders, and research leads licensing the Judgement Corpus: a precise 14-day pilot protocol to fine-tune models on real human judgment, not scraped internet text. Judgment capture projects are scoped separately.
Pilot is anchored on Judgement and Decision Making, GLEAC's most defensible skill domain and the one that translates across the widest set of agent workflows.
GLEAC delivers 80 fine-tuning examples within 3 business days. Selection mode chosen by buyer: GLEAC-curated, random sample, or stratified by Bloom's level.
Buyer authors 50 evaluation scenarios in the same skill. These are the sole eval set, never seen by either model during fine-tuning. Optional 30-minute eval-design call from GLEAC on request.
Buyer fine-tunes their selected model on the 80 GLEAC examples using their standard pipeline.
Both the baseline model and the fine-tuned model generate responses to all 50 buyer-written prompts. Blind human preference rating, 3 raters per response pair, baseline vs. fine-tuned.
Results reviewed jointly within 14 days of data delivery. Primary metric: blind human preference rate lift.
Tell us what you’re working on and how we can help.
Prefer a call? Book 15 minutes with Sallyann