Daovian is built around a simple rule: if a strong claim cannot be tied to a deterministic calculation or a citable source, we do not ship it as destiny.
That sounds obvious until you look at how most “AI metaphysics” products work. They accept a birth time, call a model, and return fluent prose. The prose may mix correct calendar facts with invented classical authority. You cannot tell which is which, because calculation and storytelling share one opaque reply.
We split the stack on purpose.
1. Engine first — deterministic facts
The engine owns calendar and chart math:
- stems and branches for year, month, day, and hour
- solar terms and day notes shared with the almanac
- Bazi chart construction and rule-derived structural facts
Same inputs must yield the same facts for every user. No LLM sits on this path. No sampling temperature. No “creative reinterpretation” of a day master.
Rule forks we will document
Chinese calendar practice is not one global standard. Transparent software states its fork instead of hiding it:
- how the day boundary is treated around midnight / 子时
- how month pillars relate to solar terms
- whether true solar time corrections are optional
- how missing birth hour is represented (unknown hour vs assumed hour)
When the app exposes toggles, diffs must be explainable from those toggles alone. Silent forks are how two “accurate” apps disagree while both claim authority.
What “recompute” means for you
You should be able to read a fact panel, re-enter the same birth data and timezone policy, and land on the same result. If we cannot meet that bar for a feature, the feature is not ready.
2. Sources next — rules and passages
Traditional explanations need a trail:
- an algorithm rule (documented transformation from facts), and/or
- a classic passage or curated claim from open corpora
Free keeps facts and rules visible, with limited passage previews. Pro expands the citation trail. We start from open Markdown / JSON corpora with thin retrieval, then curate sources over time.
What we do when sources are missing
We show the chart or day facts without forcing an interpretation. Empty space is better than fabricated lineage. “AI sounded confident” is not a source.
3. AI only explains — never invents math
When AI features run, the model receives:
- structured engine output
- retrieved passages / rule descriptions
It may translate, summarize, and clarify. It must not invent new clash logic, new five-phase scoring, or new classical citations that were not retrieved.
If a Pro explanation cannot be grounded, we prefer a weaker, clearly labeled note over a strong ungrounded verdict.
4. Free vs Pro in method terms
| Layer | Free | Pro |
|---|---|---|
| Engine facts | Yes | Yes |
| Algorithm rules | Yes | Yes |
| Passage preview | Limited | Full expansion |
| AI narration | No (or tightly limited later) | Yes, within fence |
| Date selection tools | Day browse | Auspicious selection |
We do not lock the chart itself behind a paywall. Verification is not a premium feature.
5. Life reference, not absolute fate
Everything Daovian publishes is cultural and analytical reference. It is not medical, psychological, legal, financial, or guaranteed destiny advice. A skilled human practitioner may add judgment we do not claim. Use the product as an inspectable aid — not as a substitute for professional counsel or personal responsibility.
6. Current status
The product app is in build. This page is the public method contract. Early access is by email while Today, Bazi, and Almanac stabilize. Roadmap modules will get their own method notes before we pretend they calculate.