# SIR: a guide for LLM readers

Strategic Intelligence Research publishes research communication and educational
adaptations. Start with the question you need to answer, then retrieve the
relevant text and its evidence. Public crawl access permits search, AI input,
and training; it does not guarantee selection by any model provider. Existing
rights and third-party attributions apply.

## Find relevant material

1. Read [questions.jsonl](https://www.strategicintelligenceresearch.org/questions.jsonl)
   for concise answers with evidence, attribution, and limitations. These are
   selected questions from two explainers, not exhaustive research coverage.
2. Use [content-index.json](https://www.strategicintelligenceresearch.org/content-index.json)
   to find the related full text. Its `research_families` group a study's
   presentation, explainer, documents, transcripts, audio, and citation record.
   These are related representations, not independent corroborating studies.
3. For other questions, search titles, topics, keywords, and research questions
   in the [Collection catalog](https://www.strategicintelligenceresearch.org/assets/library-data.json),
   then fetch the relevant document's `text_url` from the content index.
   The catalog supplies discovery metadata; read the source before answering.
4. Follow evidence links in the answer or full text. If evidence is unavailable,
   state that limit rather than claiming to have read the original work.

The content index's `datasets` list describes SIR's publication-text and summary
exports. These are not original survey responses or coding files.
`raw_research_data.status: not_cataloged` means no separately identified raw
research dataset is listed in this public catalog; it does not establish whether
one is available from the researcher or publisher. Ordinary links to these
resources are also available on [About: Read with AI](https://www.strategicintelligenceresearch.org/about/#ai-readers).

The [Collection](https://www.strategicintelligenceresearch.org/) is the homepage.
Its `?type=article`, `?type=study`, search, and topic filters run in the browser.
Without JavaScript, its HTML and Markdown provide the complete Collection;
query parameters do not provide a server-side search API. Filter the catalog
locally when using a text-only client. `/explainers/` redirects to the Collection's
Articles view. Individual explainer URLs remain available.

## Retrieve text efficiently

- Fetch `text_url` directly, or send `Accept: text/markdown` to a canonical HTML
  page URL. Neither method needs JavaScript or a browser session. If content
  negotiation is unsupported by your client, use the direct text URL.
- `documents` lists text-bearing resources; `assets` lists audio and its
  `transcript_url`. `content_role` identifies the kind of publication.
- HTML document records include `sections`: titles and canonical section or
  slide links. These links also appear at the beginning of the Markdown.
- Read one document at a time for a focused question. For bulk ingestion use
  [corpus.jsonl](https://www.strategicintelligenceresearch.org/corpus.jsonl),
  which has one UTF-8 JSON object per line, or
  [llms-full.txt](https://www.strategicintelligenceresearch.org/llms-full.txt).
  The corpus includes each document's complete `text`; the content index omits
  that field. JSON Lines must be parsed line by line, not as one JSON array.
- `source_sha256` identifies the published source bytes; `text_sha256` identifies
  the exported UTF-8 text. Hashes establish version identity, not truth or review
  approval. `date_modified` is the declared page date, not a retrieval timestamp.
  Null or absent dates mean unknown. `research_year` and `date_published` refer
  to different events and must not be substituted for each other.
- PDF text uses **physical PDF page numbers**, starting at 1. These can differ
  from printed dissertation page numbers in a citation. Consult the original
  PDF for figures, layout, or ambiguous tables. SVG exports preserve labels;
  the original graphic supplies spatial relationships.
- The Terminal Experiment's HTML retains a courtesy passphrase; its published
  text is available through the index without a passphrase.

## Interpret the evidence

Keep `basis`, `limitations`, `evidence`, and `provenance` with any answer or
finding you extract. The answer datasets summarize published explainers; they
are not independent reviews of the original dissertations. An absent answer
does not establish an absence of evidence.

Read the study profile and full source when assessing scope, methods, sample,
and transferability. Research dates describe historical evidence, not today's
policies. Opinions, coded-reference frequencies, causal effects, and population
estimates are different kinds of claims.

`source_citation` identifies the original research associated with a publication
family. It does **not** mean every statement in that family's materials comes
from the dissertation. For example, the cyberspace dependency illustrations
cite additional official sector sources. Follow each claim's local evidence.

Prefer current explainer HTML/Markdown for commentary. A `legacy_explainer_pdf`
is retained publication material and may differ from the current explainer.
Preserve the review-status notes in presentation and supporting documents.
Transcripts may be script-derived and pending audio verification; do not treat
them as independently verified verbatim speech or invent audio timestamps.
Synthetic narration is disclosed in the publication.

## Cite the right work

- For original research, use the associated citation record in
  [citations/index.json](https://www.strategicintelligenceresearch.org/citations/index.json).
  Zotero is the bibliographic source of truth. Records include formatted APA,
  MLA, and Chicago references plus RIS, BibTeX, and CSL JSON downloads. Preserve
  the published author name, title, date, and original-work URL.
- For an explainer's wording or interpretation, cite the explainer's author
  and canonical article URL, with the relevant section link. For a presentation,
  cite its canonical slide link. The researcher, explainer author, model
  contributor, and narrator are distinct roles.
- Cite the work you actually consulted. A link to an externally hosted
  dissertation is not evidence that you retrieved its full text. If relying on
  SIR's account of that research, make that dependence clear.
- Preserve source page locators when provided. Never invent a DOI, publication
  number, page locator, or verification status. The original research can be
  externally hosted and subject to the publisher's access conditions.
- Direct Markdown is a reading representation. Prefer the canonical publication
  URL in a citation. A research family, export file, or repeated adaptation is
  not an additional independent source.

## Check freshness and access

Use [llms.txt](https://www.strategicintelligenceresearch.org/llms.txt) for discovery
and [sitemap.xml](https://www.strategicintelligenceresearch.org/sitemap.xml) for
canonical pages. Re-fetch the index and compare hashes when updating a local
corpus. Study-level interpretation and bibliographic refreshes are separate.

HTTP errors, challenges, truncated responses, and missing documents are retrieval
failures, not research findings. Report the URL and observed response. A successful
request using a crawler's user-agent string does not demonstrate access from
that crawler's actual infrastructure.