"Transparent" is overloaded. Glass offices are transparent. Privacy policies nobody reads are transparent. In Chinese metaphysics marketing, it often means "we feel trustworthy" without any test you can run.
At Daovian, transparent is a checklist — three operations a user (or reviewer) can perform on any strong claim. If a product fails one of them, we treat that as a product bug, not a philosophical disagreement.
1. Recompute
Chart and calendar facts must be reproducible from inputs.
That includes pillars, solar terms, day notes tied to the shared calendar engine, and structural relationships the engine derives by rule. You should not need to trust our servers' mood. Export or screenshot the fact panel, re-enter the same birth data and timezone policy, and you should land on the same result.
Recompute also implies documented rule choices. Calendar systems have forks: early/late rat hour, solar-term month boundaries, optional true solar time corrections. A transparent tool states which fork it implements. Silent forks are how two "accurate" apps disagree while both claim authority.
What recompute does not require: pretending one global standard ends all disputes. It requires showing your work so disputes become inspectable instead of rhetorical.
2. Cite
Traditional explanations must open to algorithm rules or classic passages.
When we attach language to a fact — "this branch clashes with day," "this element is weak in season," "this day note comes from almanac tradition" — there should be a pointer:
- an algorithm rule identifier or documented transformation step, and/or
- a source passage (preview on Free, fuller text on Pro where licensing and curation allow).
"Cite" does not mean "GPT mentioned Zi Ping." It means a user can read the underlying material and judge fit for themselves. If we cannot support a strong classical claim, we show facts without forcing an interpretation. Empty space is better than fabricated lineage.
3. Bound AI
The model explains within engine facts and retrieved sources. It does not invent new math.
Bounded AI still does real work: translating archaic phrasing, summarizing long entries, answering "what does this term mean?" in plain language, comparing two cited views without merging them into a fake consensus. The boundary fails when the model adds novel destiny conclusions — hidden stems "revealing" wealth, invented penalty rules, dramatic predictions with no trail.
We design prompts and UI so the default path shows facts first, sources second, AI last. Optional AI copy is labeled optional. Pro may expand depth; it does not unlock secret truths behind a paywall.
What transparency is not
Transparency is not infinite verbosity. We will not dump raw JSON at every tap. It is not false certainty either — we label outputs as life reference, not guaranteed fate, medical, or legal advice.
It is also not hostility to human masters. Practitioners bring judgment, interview context, and responsibility we do not claim. Daovian optimizes for auditability across users and time. A skilled reader in the room optimizes for nuance in a single case. Both can coexist.
A quick vendor test
Before you trust any metaphysics AI, ask four questions:
- Show me the chart facts without interpretation.
- Show me the rule or passage behind your strongest sentence.
- Change nothing and reproduce the same facts.
- Turn off AI — what remains?
If the vendor only shows a score and a vibe, you are buying persuasion, not reference.
How this maps to Free vs Pro
We do not hide the chart behind payment. Free keeps facts and core rules visible, with passage previews where possible. Pro pays for interpretation labor and infrastructure: fuller citations, AI explanation in-app, auspicious date selection, higher history limits — not for the privilege of seeing your own pillars.
That split aligns incentives. We are not paid to scare you into upgrading because basic truth is locked away.
Why we publish this
Method pages and blog posts like this one are part of the product surface. They set expectations before anyone emails us about a "wrong" reading that was actually an ungrounded model sentence, or before a journalist compares us to black-box fortune apps.
If we cannot explain how a feature passes the three checks, the feature waits. That slows marketing. It speeds trust.
Transparent is verb-like here: recompute, cite, bound. Everything else is decoration.