Decision systems that learn your judgment.
Cyclotron wraps a decision model with human feedback, learned decision heads, calibration, and an inspectable optimization trace. It turns a scorecard into a system you can improve without hiding how it changes.
- Human feedback
- Trusted labels
- Model output
- Calibrated decisions
The loop
1
2
3
An inspectable lab
Watch the loop earn its improvements.
This playback uses Cyclotron's actual classifier-monitor component. It steps through the same decision, feedback, optimization, and evaluation states that the lab records for a real run.
Recorded lab playback
The decision model answers the current scorecard before the person votes.
Request
Should this research item be included in the knowledge base?
Recorded outcome
include · 62% confidence
Papyrus
- Recall
- 40.0%
- Precision
- 50.0%
- Accuracy
- 58.0%
12 reviewed in latest 200 · positive: include
Confidence calibration
No calibration snapshot recorded at this point.
Raw decision model vs final classifier
No matched raw decision-model and final-classifier predictions recorded yet.
Each classifier uses its most recent 200 human-labeled items, or all available if fewer. Predictions were made before your votes. Not a held-out evaluation.
Choose your level of help
Start open. Add help when it helps.
No cost
$20/month
$20/month + $100 once
Quoted