Score timestamped predictions against outcomes.
A free zero-dependency Node loop that accepts sensors and a scorer you supply, records falsifiable predictions, grades mature predictions against outcomes, and reports calibration metrics.
How it actually works
Exactly what it does — and when.
A workflow records whether a scheduled check will pass with a confidence and future test time. After the real result arrives, the supplied scorer grades it and updates calibration. This measures recorded predictions; it does not establish consciousness or train model weights.
- An agent asserts accuracy without testing it
- The predicting model grades itself
- Confidence is never compared with outcomes
- Require confidence and a test-after time on predictions
- Grade with ground truth supplied through your scorer
- Inspect per-domain calibration and surprise records
Use it to…
Real ways people run SENTIENCE-LOOP — Predictive-Processing Cognitive Loop.
Record a falsifiable prediction
Save the claim, confidence, domain, and the time after which it can be tested.
Score a mature forecast
Use the external outcome and scorer rather than letting the predicting model self-grade.
Inspect calibration
Review Brier score, hit rate, and surprises by domain before changing confidence policy.
What's inside
Who it's for
Agent developers
want claims turned into testable predictions
Forecasting workflow owners
can provide timestamped outcomes
Evaluators
need transparent calibration metrics
Questions, answered
Ready to put SENTIENCE-LOOP — Predictive-Processing Cognitive Loop to work?
Free download · no account needed · yours forever