Arithmetic should be reproducible
A planetary degree, ascendant, house number, or nakshatra is the output of defined inputs and rules. The same inputs under the same method should produce the same result. That makes software calculations testable: we can compare them with independent references, write boundary tests, and identify which release produced a value.
A language model is good at explaining text, but it is not a trustworthy ephemeris. Asking it to calculate a chart directly invites invented positions or inconsistent arithmetic. Kalika therefore keeps astronomy and astrology calculations in scripts.
Educational material is not a chat answer
The learning library is made from maintained content records and static pages. Names, ranges, and the structure of a whole-sign chart are published as inspectable reference material. We can update an error in one source and rebuild the pages; the content does not change unpredictably from one visitor to another.
Static educational pages also work when the future AI service is offline. They can be linked, read with assistive technology, and checked by subject-matter editors.
What AI may do later
A future chat interface may help a person explore traditional interpretations or ask follow-up questions. It should receive the chart facts from the calculation layer, identify which facts it used, and distinguish tradition from certainty. It should not quietly alter the chart, invent a scripture citation, or offer frightening guarantees.
The chat layer is not live in this beta. We will refine it separately rather than using AI output to cover gaps in the deterministic foundation.