How this page is made

The path of the information

Think of an assembly line with four stations. At each one, the material gets a little more organized and a little easier to check.

1. Capture

Given a request like “Captura Stargate: Sirius”, the AI reads the sources Aileen has already gathered (books, papers, class transcripts) and, when needed, searches the web. It fills in a standard sheet: identification, astronomical data, history, cultures, mythology, astrology, magic and oracle. The format is always the same, for every star and constellation.
2. Review

The sheet goes into Aileen's research notebook, where she reads it, corrects it and decides what stays. Nothing goes to the site without that reading.
3. Turning it into data

The sheet (running text) is sliced into a structured file in JSON format: a list of statements, a list of names, a list of correspondences and a list of references. This file is also reviewed before moving on, and it is kept with a full change history.
4. Loading into the site

A program reads the file and creates the records in the site's database. The page you are looking at is built on the fly from those records.

What the AI does (and doesn't)

The AI helps to

Read a lot of sources quickly and find what each one says about the star.
Separate astronomical data from myth, astrology, magic and oracle.
Break the text into small statements, one per line, each with its own source.
Gather names and correspondences that appear in several works.
Check that every citation has a reference and vice versa.

The AI may not

Make up data, dates, etymologies or references: where there is no source, the sheet stays blank.
Decide on its own what goes on the site: the review is human.
Pass off text it wrote itself as if a person wrote it (see “Gnosis and AI-written text”).
Use a source nobody can check.

How references come together

This is the most important technical part, and the idea is simple: no piece of information travels alone.

One statement per line. Instead of a paragraph with five facts mixed together, each fact becomes its own record (in the database it is called a Claim). That way we can say exactly where each one came from.
Each statement points to its source. In the sheet, the citation is a short key such as [Allen, 1963]. In the database it is linked to the full work: author, title, year, link or file and, when available, the page and the literal quote.
More than one source, one line. If two works say the same thing (for example, the same stone associated with the star), there is a single record with both references attached. Different names for the same value are merged too.
Each work is catalogued once. Frequently used books have a fixed “catalogue key”. So the Agrippa cited in the Sirius sheet and the Agrippa cited in the Spica sheet are the same reference, and the site's resource list doesn't fill up with duplicates.
No checkable reference, no entry. Sources nobody can verify (like an old spreadsheet with no origin) are left out. The loading program rejects the whole sheet if it finds one.
On the page, they become footnotes. Each tab numbers the sources it used (¹ ²…) and shows the full list at the bottom. Click a number to jump to the reference.
When sources disagree. We don't quietly pick a side. Both versions appear, each with its own source, and the disagreement can become a research question.

Confidence levels

When the data is loaded, each statement is stored with two labels. They are used for review and to organize the database; the page doesn't show a confidence badge on every line. What it shows is the source of each piece of information (the numbered notes) and the “personal gnosis”, “generated by AI” and “to review” tags.

How well documented

High: well documented in a solid source.
Medium: partly documented, or with a caveat from the source itself.
Speculative: interpretation or gnosis, not a verified fact.

When in doubt, the sheet always goes one level down.

What kind of source it is

Primary source (ancient texts).
Academic secondary source.
Esoteric tradition.
Modern astrology.
Starlight Laboratory experience.
Personal gnosis and inference.

A fact from an ancient text and a personal impression are recorded appropriately.

Where things appear

The sections of the sheet become, almost one to one, the tabs of the page. Each piece of information lives in a single place, so nothing is repeated.

Tab	What goes in
Astronomy	Identification, technical data (distance, magnitude…), asterisms, history, observation.
Names	Alternative names, each with its culture and source.
Myths	Cultural traditions, mythology and associated deities.
Astrology	Symbolism and fixed-star readings, separated by author.
Magic	Correspondences (stone, herb, color…), grouped by type, and notes from the tradition.
Prayers	Greetings, enchantments, invocations and channelings, with credit.
Personal gnosis	Experiences from people in the group, always with the name of who lived them.

Constellation pages follow the same path, with the same sheet and the same rules.

Gnosis and AI-written text

Research and gnosis are different things. Research brings what is documented. Gnosis brings what people felt and received in practice. Neither corrects the other: if they disagree, both appear, and gnosis carries its own tag.
Always credited. Channelings and accounts appear with the name of who wrote them. The project's author appears as Aileen Daw.
AI-written text is tagged. If a greeting, an enchantment or a message was generated by AI, the page shows “Generated by AI” and never attributes the text to a person.
Some areas require login. Prayers and personal gnosis only appear to people who have logged in.

Technical safeguards

All or nothing. Each sheet is saved in a single transaction: if there is an error (a cited reference that doesn't exist, for example), nothing is saved and the program lists the problem.
Running it again doesn't duplicate. Every line has its own identity; on a new load, what changed is updated and the rest stays the same.
Clean text. All formatted text goes through a filter that strips unwanted code before it is stored.
Private material stays out. The program compares what would be published against reserved files and discards anything that matches.
Community contributions go through moderation before everyone can see them.