Research-led astrology intelligence.

We do not treat generative output as a substitute for calculation. The model receives structured chart facts from the calculation layer, then produces interpretive guidance within defined boundaries.

1. 01
   Data approach
   Anonymized historical predictions paired with structured chart context. Personal identifiers are removed before any review or training. Quality, not volume, drives selection.
2. 02
   Foundational model training
   We are training a foundational astrology model to enable prediction for everyone, and building a framework that lets practitioners define and train their own models on top of it. The model receives structured chart facts from the calculation layer, then produces interpretive guidance within defined boundaries.
3. 03
   Evaluation framework
   Outputs are scored on chart consistency, specificity, empathy, contradiction rate, and refusal behavior — both automatically and through sampled human review.
4. 04
   Safety boundaries
   Defined refusal policies for medical, legal, financial, psychological, and emergency-adjacent topics. Generative outputs are not treated as advice.
5. 05
   Practitioner review
   Senior traditional practitioners review samples for tradition consistency. Their feedback drives data selection and prompt design, not just final filtering.
6. 06
   Public notes
   Findings, limitations, and methodology notes are published as research notes when stable enough to discuss.

Developed with inputs from doctoral researchers, engineering contributors, and traditional astrology practitioners with decades of field experience.