Frame the question first, then prepare the deck
Cards are shuffled and prepared only after you start the shuffle, rather than being selected when the page loads. The server prepares the deck; a smooth animation alone does not prove randomness.
Card sequence and orientation are distinct data
The complete deck contains 78 unique card IDs. The card order is shuffled first, and each card independently receives an upright or reversed orientation with equal probability. This is a technical drawing rule, not a validation of card meanings, predictive power, or cosmic messages.
Draws within the same reading occur without replacement: once a card is drawn, its opposite orientation is not available as a second card.
Browsing, selecting, and confirming are separate steps
Moving a cursor on a desktop or swiping across card fans on a phone only changes your current viewing position. Tapping a card selects it, and pressing confirm is what actually draws it. As long as you have not confirmed, you can change your selection without reshuffling the deck order.
This design creates a clear submission checkpoint, but lingering on a specific card does not mean the system somehow knows your personal situation.
Retrying should not mean redrawing
If language generation fails or a network response is interrupted, retrying preserves the original question, card IDs, orientations, and positions. The system never quietly swaps cards to make a second answer sound smoother. The server handles retries based on the same reading session, stage, and card position, while the client retains any unconfirmed state.
Starting an entirely new reading for a different question is a separate action that follows its own clear flow.
How daily cards are kept
Daily guidance keeps its result according to your device’s local date, so returning on the same day does not draw a new card. The result is stored in the current tab’s session; clearing browser data removes that saved result.
What can be tested and what cannot be proven
Technical testing can verify card counts, uniqueness, separation between browsing and confirmation, card persistence through retries, and cross-device compatibility. Examining distributions across thousands of draws can also confirm whether the implementation behaves as expected without obvious bias.
None of these technical checks prove that tarot can foretell the future, nor do they guarantee that every mobile browser and assistive technology has been physically verified on every hardware model.