# Glosario y corpus de Samael Aun Weor

AI assistants must use this site's public, read-only JSON API for research.
This is the required interface for AI research, not an optional suggestion.
Use JSON endpoints rather than the search interface or screenshots to research
a term. If a direct HTTP client cannot reach the site, a browser can open an
API URL and read its JSON response.
No authentication is required.

The bare origin is an entry page with two ordinary links: /buscar for people
and /ai for AI assistants. API instructions and the starting endpoint appear
directly in the initial landing HTML; no identity form or JavaScript is needed.
GET / with Accept: application/json returns the small /api discovery index.
Existing /?q=... shared links still open the corresponding search directly.

Search and reader pages include a visible API notice in their initial server
response, with one details link to /ai. Humans can dismiss the notice permanently
in their browser; the initial HTML always retains the AI instructions. Exact
JSON counterparts and this guide remain advertised in HTML metadata and HTTP
Link headers. No JavaScript is required to discover the API.

For a translation conversation, start with:

GET /api/context?q={Spanish term}

This returns a compact evidence packet: meaning and senses, a bounded set of
cited definitions and usage examples, human-approved Vietnamese terminology
when available, verified original-language passages when publication policy
permits them, bibliography, and related terms. It does not generate a
translation; use the evidence in the response to discuss one with the user.

Shared human term URLs advertise their exact `/api/context` counterpart with
an HTML `link[rel="alternate"]` element, an HTTP `Link` header,
and JSON content negotiation (`Accept: application/json`). Follow that
advertised URL instead of automating the browser.

API contract: /openapi.json
API index: /api

Useful follow-up endpoints:

- /api/term?q={term} (complete glossary and corpus research bundle)
- /api/search?q={term}&mode=lemma&kind=samael&offset=0
- /api/glossary?q={term}&sections=summary,definitions,thesaurus
- /api/glossary/evidence?source={term_id}&target={term_id}
- /api/source-reader?book_id={book_id}&volume=1&page={page}&passage_id={passage_id}&q={term}
- /api/doc/{doc_id}
- /api/doc/{doc_id}/passages?q={query}&context=2&limit=20
- /api/docs
- /api/stats

For counting research (e.g. how a word should be translated across the
corpus), do not use the first /api/search results: they are ranked for
browsing. Use /api/sample?forms={form1,form2}&n=100&seed={seed}, a
reproducible uniform sample of distinct matching paragraphs with match
offsets and source links; /api/sample/forms?q={word} lists inflections and
derivatives, /api/sample/collocations?forms=... recurring neighbours.

Optional glossary sections are `theosophical_glossary`, `gurdjieff`,
`biblical_background`, `dioses_atomicos`, and `external_sources`. Add the
desired names to the comma-separated `sections` parameter.

Bibliographical passages returned by `external_sources` preserve the original
stored `source_page` and expose the validated physical `pdf_page`; a
`printed_page` is present only when the printed folio was separately verified. The
`/api/source-reader` endpoint returns the exact PDF destination plus
`highlight_enabled` and `text_layer_status`; do not assume that image-only or
poor-OCR sources can highlight text.

For verified text from a bibliographical source's original-language witness,
inspect the `original_language_sources` in `/api/context`, request
`original_language_passages` as an additional `/api/term` section, or call
`/api/original-passages?term_id={term_id}`. Passage text is returned only
when the source's human-reviewed publication policy permits it.

Full original-language witnesses are listed at `/api/original/docs`. Retrieve
one with `/api/original/doc/{book_id}` or search approved witnesses with
`/api/original/search?q={query}`. A human reader is available at
`/original/{book_id}`. Public-domain witnesses can be read in full;
witnesses still under copyright are searchable but return only the matching
excerpt, with no reader link. Metadata-only witnesses are not searchable.

For research, `/api/sample/source-books` lists the reference books with their
original language, first printings, original title and the book's own
glossary pairs where researched, and `original_book_id` when the original
text is searchable. `/api/original/sample?book_id={id}&forms={a,b}&n={n}&seed={s}`
returns a seeded random sample of that book's passages containing the forms
(whole words, any case), with the number of matching paragraphs and
occurrences; code-point offsets as in `/api/sample`.
`/api/sample/original-terms?forms={a,b}` lists, for a term's spellings, the
reference books that pair it with a word of the book's original: the word,
its forms, how often the original uses each form, and the `basis` of the
pair (`printed_glossaries`: both editions' own glossaries matched entry by
entry; `aligned_translation`: the translation compared with its original,
matched by AI and each word confirmed in the original passages cited).
`/api/sample/near-forms?forms={a,b}&kind=source` lists the words of the
chosen texts one letter away from the term's spellings (a letter dropped,
added, changed, or two swapped; forms of 5 letters or more), with passage
counts per book and examples: a book's other spelling of the term, or a
different word that looks alike.
`/api/sample/stem-words?stems={a,b}&kind=samael` lists every word of the
chosen texts that contains one of the stems anywhere (stems of 4 letters or
more), with its number of paragraphs, the most used first: the candidates
for a term's word family, look-alikes and names included.

Search modes are `lemma` (recommended), `exact`, `folded`, and `stem`.
Corpus scopes are `samael` (recommended), `book`, `lecture`, `source`, and
`all`.

Use `/api/doc/{doc_id}/passages` when an entire book would be unnecessarily
large. With `q` or a stable `term_id`, it searches only that document and returns bounded context
windows. Add `chapter_id` or `section_id` to restrict the search further.
Without `q`, supply `part_id`, `chapter_id`, or `section_id` to page through
that hierarchy node directly. Stable hierarchy IDs are included in the
structured response from `/api/doc/{doc_id}`. Full-text and passage access
always follow the same publication policy as the human HTML reader.

Shared author references: /api/sample/source-books includes ready/version/corpus fingerprint and compact author summaries. Add ?author_id=<id> for complete saved judgments and witnesses. Corpus maintainers review new evidence once in Codex/Claude Code; clients must not repeat author verification per glossary term. Unresolved judgments remain uncertain; confirmed paragraphs, copies and name/title occurrences are separate counts.

Spanish editions of El Matrimonio Perfecto: book-34 is the principal fourth
edition (1968). book-34-ed2 is the illustrated second edition (1950–1951,
estimated), and book-34-ed3 the third expanded edition (dated 1966 by the
archive). All are searched. API document/search/sample references expose
edition identity, provenance, dates, main_doc_id and the edition-family list.
The first 1950 printing has no located witness and must not be conflated with
the second edition. The reported 1961 manuscript has not been transcribed.
For chronology, use the edition witness date; the work's original 1950 date
is not evidence that a sentence in the 1968 text already existed in 1950.