Personal Fortune-Telling

A quiet hour with the stars, and with yourself.

How the overall reading is made — what the AI is given, and what it is not

updated · 6 min read

SoraSora here. The overall reading at the end is the one place this service uses AI. What is handed to it is a fair thing to wonder about, so let me be concrete.

A reading here falls into two stages. The nine methods in the first half are calculated by a fixed procedure with no AI at all. Only the overall reading in the second half is generated.

The division is partly about cost, and not only that. The calculations of a method ought to come out by a fixed procedure; a different answer every time would be a problem.

Only symbols are handed over

To produce the overall reading, what the AI is given is symbols: card numbers, calendar signs. Names and dates of birth are not given. Neither is the text of the readings shown on screen.

ItemGiven to the AI?Note
Name and date of birthNoNot kept on the server in the first place
Card ids and orientationYesTurned into card names on the server before use
Signs, stroke counts, star numbersYesNumbers and symbols only
Text of the readings by periodNoTo avoid producing a paraphrase of them
Face and palm analysis textYesOnly what the server has signed

After the card ids arrive, what card each id refers to is looked up in a dictionary on the server. Text sent from a browser is never passed straight to the AI.

It is built this way so that nobody can use this service's API to have arbitrary text processed by an AI. On a service with no account system, that is where the weakness usually lies.

The face and palm analysis text is the one thing passed as text. It carries a signature applied when this service's own server produced it, and text whose signature does not match is refused. Nothing can be injected from outside.

The instruction is to find what is shared

The instruction to the AI is not to summarise the nine results and list them. It is to look for tendencies that come through more than one method.

Where methods with different material and different systems point the same way, that can be taken as a quality showing strongly. Conversely, anything appearing in only one method is left out of the summary.

The output is one sentence drawing the whole together, plus a few sections. Each section carries the names of the methods that produced its content.

The AI is called once

Across the whole flow, the AI is called once, for the overall reading. Adding the photograph readings adds one more call. Nothing else calls it.

StepDoes it use AI?
Calculating the nine methodsNo
Readings by period (one hundred and eighty)No
The compatibility readingNo
Reading the oracle decks by subjectNo
Face and palm from photographsYes
The overall reading at the endYes

Limiting the calls is about cost, and it makes the service faster as a result. The calculations finish inside the browser or the server, so the only waiting happens at the overall reading.

There is also a mechanism for reusing a result so the same material gives the same overall reading. The same conditions are never sent twice.

Why the source methods are always named

Each section of the overall reading shows badges naming the methods it rests on: tarot, the Four Pillars, Lenormand. They are not decoration; we take them to be the part nothing else can replace.

Without the grounds written down, there is no telling where a passage came from — whether it is the result of a method or simply plausible-sounding words. The reader loses the ability to check.

The AI is handed a list of the method names it may use. Only methods actually supplied as material appear, so no name of a method that was not used can turn up.

The same badges appear in the letter PDF, so that when you read it again later you can still trace which method produced which content.

SoraEven where the overall reading cannot be produced, the nine results remain as they are. It is built so that a fault on the AI side does not stop the reading itself.

Common questions

Whose AI is used?
The system of OpenAI, a provider based in the United States. The photograph readings use the same provider, and the screen before sending says so.
Does the same input give the same overall reading?
With the same reader and the same material, there is a mechanism for reusing the result. Since the cards drawn change, a different day gives different material and different text.
Does reading only the overall section make sense?
It does. It is the overlap of the nine picked out, so the whole picture is gathered there. Go back to any individual method that interests you.

The nine results are crossed and the shared themes drawn together.

Go through to the overall reading

Related reading

The readings on this service are reference material drawn from the traditional interpretations of each method. They do not guarantee future events or outcomes. For decisions concerning health, law or money, please consult a qualified professional.