Draft
What AI interpretation may and may not change
AI writes from fixed facts; demo mode is labeled and cannot redraw results.
Original draft for local preview. Human editorial review is still pending; this article is not published in production.
Cards, lots, hexagrams, numbers, and themes are fixed before any model call. Payment, language changes, and follow-ups reuse the same factsHash.
In demo mode the interpreter is simulated and the report says so. Production must not hide that distinction when mocks are active.
Generated text cannot invent medical diagnoses, wealth guarantees, relationship success odds, or deity identity. Private pages stay noindex.
A report has two sources of input: saved rule facts and optional context. A question can guide emphasis, but it cannot instruct the interpreter to replace a card, change a date, or reveal private answers. The system validates the report structure and references against the saved reading. Those checks reduce certain errors; they do not establish that every sentence is insightful or that a suggested action will succeed.
When checking an interpretation, compare names, orientations, and numbers with the fixed result first. Then identify which sentences are observations, cultural metaphors, or suggestions. Ask for clarification if a claim has no visible basis. Failed generation retries reuse the report reservation and original facts; they are not fresh draws. A paid report purchases the stated writing service, without buying certainty or changing the original result. Keep any disagreement and uncertainty visible in your own decisions.