Loading...
Loading...
OmegaEngine · Feedback & RLHF
Feedback Grading lets operators review OmegaEngine decisions and submit corrections. Each grade feeds a continuous learning loop that recalibrates risk and regret scores per domain, reducing false positives over time.
OmegaEngine evaluates a request and returns a decision with a risk score, policy outcome, and audit trail.
A human reviews the decision and rates it GOOD, BAD, or NEUTRAL — with optional labels, severity, and notes.
Grades feed the calibration loop: risk and regret scores are adjusted per domain using recent feedback ratios.
The decision was accurate. Reinforces current calibration for this domain.
The decision was wrong. Feeds recalibration and the RLHF queue with your labels and comment.
Ambiguous or informational. Recorded in the learning log without pushing calibration either way.
# Grade a single decision
curl -X POST https://omegaengine.ai/api/feedback \
-H "Content-Type: application/json" \
-H "x-api-key: $OMEGA_API_KEY" \
-d '{
"decisionId": "1c2a7e0e-5b3f-4c9d-9a41-8f2e6d0c7b1a",
"rating": "BAD",
"labels": ["false_positive"],
"severity": 2,
"comment": "False positive - vendor was already approved in Q4 review"
}'
# → records the grade, pushes it to the learning log + RLHF queue,
# and returns your live Omega grade# Poll your live grade + oversight stats for dashboards
curl -H "x-api-key: $OMEGA_API_KEY" https://omegaengine.ai/api/feedback
# → e.g. "A+ · 94% GOOD · 3% BAD" plus per-topic breakdownsThe feedback summary (GET /api/feedback) computes:
A single S / A+ / A / B / C / D grade over recent feedback — heavy penalty for BAD, bonus for GOOD.
GOOD / NEUTRAL / BAD totals and ratios across your recent decisions.
Per-topic breakdown of where BAD feedback concentrates, with sample comments for tuning.
Suggested policy adjustments generated from the topics with the worst feedback.
Access the full feedback dashboard to review pending decisions, submit grades, view calibration metrics, and export training data for fine-tuning.