A deterministic AI astrology reading service

Grew from an astrology skill he had already built. Core idea: a meaning graph. Computed natal and transit configurations map by stable ID to an authored corpus of meanings, and that structured grounding is passed to an LLM for prose, so the same chart state always selects and suppresses the same facts: constrained improvisation, not free generation. Differentiators: determinism over the resolved graph, and continuity across time (the arc of a life, not point readings), which he recognised is a memory problem (a "said-log" of prior claims as constraint) rather than a chart problem. His learnings: specificity creates falsifiability and falsifiability produces resonance; deceptive grounding (fluent prose bound to the wrong astrological entity) is the dangerous failure with a literate audience; the diviner tradition (Cornelius) is the dominant serious camp and the I Ching is the precedent for a fixed deterministic corpus used for divination. Settled: subscription model, trial not freemium, per-person transits, single pre-cached corpus rather than retrieval. Unsettled: consumer brand versus professional tool; distribution, which he identifies as the binding constraint. He is building before selling, knowingly, and intends a tripwire date because building is the enjoyable part and selling to strangers is the aversive part.

Live: re-check with James before relying on this.

Provenance: James, in conversation with Claude, Sept 2026 (unverified); body restored from raw/notes/hermes-wiki-seed-2.md on 2026-10-18 after it was found missing. Disposition: live.

Detail: astrology-product-design-learnings

Part of thread: thread-judgement-in-machines