AI Disclosure

1. AI’s role in Oracely

An Oracely reading has two stages. First, our deterministic engine computes a chart from traditional rules. Then AI writes a natural-language explanation of that structured result.

AI is instructed to read the computed JSON and write section copy such as personality, relationships, and work. It is not asked to generate or revise the calculated values.

2. What AI does not do

We do not send a birth date to a model and ask it to cast a chart. Planetary positions, pillars, Mayan kin, and element counts are already fixed by the engine.

If an Ascendant or hour pillar is missing, the prompt tells the model not to guess it. AI does not guarantee future events, admissions, income, or health outcomes.

3. How a reading is produced

Birth information you enter is sent to the calculation engine first. Each system uses a fixed engine version (for example mayan-v1, western-v2, saju-v1, bazi-v1) so the same input yields the same chart.

After calculation, the interpretation service receives the chart JSON and a language code and produces a narrative document. We store interpretation version, prompt version, and model name so a later visit in the same language can reuse the saved reading.

4. Data sent to AI

The user message for a reading interpretation currently contains (1) the tradition identifier, (2) the interpretation language, and (3) the engine’s calculation JSON.

That JSON includes values such as Mayan kin and day sign; Western Sun, Moon, planetary placements, houses, and aspects; or Four Pillars, Day Master, elements, and Ten Gods.

Email, account identifiers, phone numbers, payment data, IP addresses, place names, and raw birth date or birth time strings are not placed in the AI request body. An account identifier may be used only for internal usage logging and is not sent in the OpenAI message.

Western chart JSON may include the timezone and latitude/longitude used for the chart. Four Pillars JSON may include timezone at birth. These are coordinates and timezone used in calculation, not a street address or place name.

Purposes, retention, and international transfers of personal information are described in the Privacy Policy.

5. AI provider

Oracely currently may use the OpenAI API to generate AI-assisted interpretations. If no API key is configured, we use the template interpretations described under errors and fallback.

The provider or model may change for quality, safety, or operations. This page does not hard-code a model name. A stored reading may keep the model identifier used at generation time.

6. Data and model training

Under OpenAI’s published API data-usage policy, inputs and outputs submitted via the API are not used to train OpenAI models by default.

This repository does not enable an OpenAI training or data-sharing opt-in in application code. We cannot verify the live OpenAI organization console from the codebase, so we treat confirming that data-sharing / feedback opt-in remains off as an operational check.

A provider may still process data as needed to fulfill the request, prevent abuse, or meet legal duties. We do not claim that data is never stored in any circumstance.

7. Accuracy and limits

AI interpretation may include generic wording, inconsistent wording, incorrect inference, or an incomplete explanation.

Readings are for entertainment and self-reflection. They are not medical, legal, financial, investment, or mental-health advice, and they do not guarantee future events or specific outcomes.

8. AI output and version changes

Engine calculation is designed to be deterministic for the same input. AI wording may not be.

Even for the same chart, phrasing can vary with model version, prompt version, or interpretation document version. AI does not change the underlying calculated data.

We may change models or interpretation methods as the service improves. Internally we distinguish engine version, prompt version, interpretation version, and content version.

9. Automated decision-making

Oracely’s AI does not make significant automated decisions about credit, hiring, insurance, legal rights, medical eligibility, or eligibility to use the service.

AI is used only to generate personalized content. This notice helps you see when AI is involved. It does not classify Oracely as a high-risk AI system.

10. Quality review and human oversight

Operators may periodically review prompts, sample outputs, failure cases, and quality metrics.

An individual AI reading may not have been reviewed or approved by a person in advance.

11. Errors and fallback

If the OpenAI API key is missing, a request fails, the response is empty, or the document is not in the expected shape, we may serve a pre-defined template interpretation.

API calls are retried a limited number of times. Mayan, Saju, BaZi, and Western readings can then fall back to templates. If an interpretation is already stored, a later visit in that language can reuse the cache. If a stored interpretation is too short or uses an older document format, we may regenerate it.

12. How readings are labeled

On the result screen, calculated figures are labeled “Calculated by Oracely Engine.” Narrative sections are labeled “AI-assisted interpretation.”

The label links here. The goal is that you can tell engine math from AI wording while you read, not only if you find this page.

13. Contact

Questions about AI use, data, or interpretation quality: info@univting.cc

14. Changes to this notice

If the pipeline, provider, or data we send changes, we will update this page. Material changes are shown with the effective and last-updated dates.