SOURCES & METHOD
Read the evidence. Keep the uncertainty.
Gateway Explorer is an independent project that makes Monroe-related material on Internet Archive easier to navigate and discover. It is not affiliated with, sponsored by, or endorsed by the Monroe Institute.
Archive sources and editorial context
The archive experience organizes material published on archive.org and links back to the original items. Introductions, groupings and discovery tools are our editorial additions; they are not statements from the Monroe Institute. Source access does not establish permission to reuse a work or endorsement of its claims.
Internet Archive is the primary collection. Selected YouTube videos and their existing transcripts enrich it, with their source clearly labeled.
What is indexed
14789 passages across 1210 Markdown documents. General conversations combine an AI explanation with supporting passages from the indexed collection. Source-specific questions use the selected material. No new OCR or transcription is generated.
Coverage varies by source. A transcript can be incomplete, and a recording can be available without a transcript. Gallery captions are derived from filenames and should not be treated as verified identifications.
What a citation establishes
Each citation resolves to an immutable passage with exact source text, a source document hash, and a page or Markdown line locator. Historical OCR page markers may differ from printed page numbers. Sections without reliable page markers use line numbers.
Exact quotation checks establish that quoted words occur in the retrieved source. They do not prove that the model’s interpretation is fair, that OCR is accurate, or that the source’s claims are true. Read the adjacent context to assess the answer.
Audio timestamps remain unverified. Earlier #5/#6 timing candidates are available in an internal review screen, but they are excluded from answers and are not exact audio citations.
Who is speaking
Transcript labels are preserved. A label such as SHE or ROMC does not establish a real-world identity. Reported Explorer and channeled accounts are attributed accounts. Unknown attribution stays unknown.
Erickson’s compilation is secondary editorial material, including where it reproduces quotations. It becomes eligible for answers when you ask about compilation, interpretation, or provenance, or explicitly select the commentary comparison option.
Reviewed background references
Basic questions can also use short, authored summaries checked against official publisher pages. These citations are labeled Reviewed background reference, show a review date, and link to the official page. They are separate from archive transcripts and do not independently validate claims about consciousness. These reviewed references are separate from the model’s general knowledge.
How answers work
For general questions, the AI explains the topic using its general knowledge, then identifies supporting archive passages and other media in our collection. The explanation appears first. Each linked source has a short description of the particular point it supports. A linked document does not validate the entire answer, and general knowledge can be mistaken or outdated. This is not live web research.
Keyword and alias retrieval supplies a bounded selection of excerpts without model-based search expansion or ranking. One model request writes the explanation and selects support. The server checks that selected excerpt identifiers exist and attaches their exact text and known source links. Missing source matches do not prevent a general explanation; the answer says when no supporting passage was identified.
Questions explicitly scoped to a source, session or named work retain a stricter path: answers must be supported by the selected excerpts, with a separate model support review and exact citation validation. This additional review can take longer. A refusal on that path means the retrieved excerpts did not establish the answer, not that the entire historical record has no answer.
How search works
Lexical search finds words and phrases. Reviewed aliases broaden vocabulary matches. Library search uses keyword and alias matches without paid model calls. General Ask uses the same keyword and alias retrieval. Source-specific questions can use additional model ranking and review. No paid embeddings are used. Every result shows its match reason. Lexical search and source browsing remain available when the model is offline.
Site introductions, collection descriptions, and terminology are indexed as site editorial guidance, separately from archival evidence. Relevance rationales and diagram OCR are excluded. Concept pages use reviewed orientation text; their passage lists are lexical matches, not reviewed annotations.
Where processing happens
The website runs on Vercel. A private Supabase database holds the source evidence and reading snapshots. Audio, video and gallery images load from Internet Archive; optional YouTube players load from YouTube.
The AI feature uses gpt-6-luna through the paid OpenAI API. Only bounded questions, context and selected excerpts are sent, with response storage disabled. Provider abuse-monitoring retention can still apply. A shared allowance and request limits bound AI usage. Answers become unavailable when the allowance is exhausted; browsing and search remain available.
The app keeps conversations in tab memory and clears that page history on reload. Opening an original-source link takes you to its publisher or Internet Archive.
To limit automated requests, the server stores a keyed hash of your network address with request times. Records older than two days are deleted when the next question is admitted. It does not store chat messages in that rate-limit ledger. Hosting and model providers may retain operational logs under their policies.