Research atlas · 29 October 2026
What I was trying to keep apart
2,229 entities, 2,271 attributed claims, 259 document records, 121 research sources. The live site is static. The dataset is one JSON file.
The conversations are long, name-dense, and full of claims that point at archives, books, and other people’s testimony. A transcript dump cannot show which sentence is an interview assertion and which sentence is a later note about a document. Once those get mixed, a reader cannot tell what the speaker said from what a file actually contains.
The atlas is the split. Conversation claims are attributed to a speaker, a video, and a timestamp. They include assertions, hypotheses, reported testimony, questions, and advertisements. Documentary claims cite a book, archive, paper, or catalog record and say what that source contains. They do not promote an interview allegation into a finding. Edge numbers count distinct supporting claims. They are support counts, and they are not independent corroboration.
Full transcripts stay with their publishers. This project stores locators, paraphrases, and short
attributed statements. meta.transcript_redistribution is false. Original application code is
MIT. The research compilation is CC BY 4.0. Recordings, captions, and third-party documents are not
relicensed. A citation of the atlas does not endorse the interview allegations.
The three conversations
The corpus starts from three captioned videos. Each one is a source record with its own theme. Filters on every page can isolate one conversation or combine all three.
| Source | Theme in the atlas | Recording |
|---|---|---|
| Part 1 | Religion & reality | YouTube |
| Part 2 | History & power | YouTube |
| Part 3 | Covert programs | YouTube |
A second filter switches the evidence layer: conversations, documentary research, or both. Documentary rows are additional statements supported by the cited archival or institutional source. They are not attributed to the video.
The rooms
The interface is the same shape as the rest of this site: static HTML, CSS, and vanilla JavaScript, no
build step, no framework. data/atlas.json is about 26 MB. The app loads it with
fetch, so the page has to be opened through a server. scripts/validate.py checks
referential integrity before a publish. offline/atlas_explorer.html is the same edition in
one file and can be opened from disk.
| View | What it is for |
|---|---|
| Overview | Counts, a graph preview, and the most referenced entities |
| Knowledge graph | Directed neighborhood. Each edge opens the claims behind it |
| Claim register | Searchable attributed statements, with status and locator |
| Entity directory | People, places, institutions, works, events, concepts, aliases |
| Timeline | Historical time, kept separate from playback time |
| Source documents | Primary records and referenced works |
| Coverage & method | Rules, growth, and what the atlas still does not establish |
| Final verification | Ledger of the source checks |
| Historical evidence | The boundary dossiers |
| Research notebook | Local notes in the browser. Export them if you want to keep them |
Six research trails enter a focused neighborhood instead of the whole graph: SS and postwar networks, World Commerce Corporation, Puharich and contact networks, Tesla and technology claims, religion and controlling factions, and contactees and esoteric organizations. A trail is a way in. It is a slice of the index.
How the editions stacked
The public file is Edition 9. Earlier reviews stay in the dataset so a later pass cannot quietly replace them. Edition 3 is the coverage audit: dispositions preserved, typed relationships added, native-caption anchors checked. Editions 4 through 6 carried the remaining-limits ledger forward until every known item had a disposition. Edition 6 is the accounting snapshot later editions are measured against.
| Pass | What changed |
|---|---|
| Edition 7 | Source verdicts on six exact-source leads. The Edition 6 counts were held. Ninety Part 3 name windows were listed. None had been listened to. |
| Edition 8 | Six narrow documentary claims and fourteen source-registry additions. Six known document gaps stayed open. |
| Edition 9 | Historical-boundary review. Fourteen documentary claims, twelve source records, fourteen entities, twelve document cards, six dossiers. Six historical originals still required. |
| Publication | No new counts. Blank entity cards received a summary quoted from one attributed claim. Two documentary entities that had no links — Admonitio generalis (789) and the December 1947 Horten intelligence memoranda — now point at claims, sources, and document cards already in the file. The growth table and schema name the fields the dataset already contained. |
The Edition 9 conclusion, kept in the file, is narrow: known source trails are more precise, and claims beyond authenticated originals remain conditional. The existence of an archival file does not authenticate the extraordinary claim that cites it.
What this edition still leaves open
These limits are on the Coverage page and in meta.known_gaps. They were left open because
closing them would have meant guessing.
| Limit | Where it stands |
|---|---|
| 90 Part 3 name windows | An independent machine transcript of the source audio exists. 36 match the indexed spelling, 42 are phonetic variants, and 12 still do not secure the indexed name. Human listening is zero. A matching window identifies a spoken form. It does not authenticate the allegation around the name. |
| Five historical closure conditions | Still missing as records in this archive: a Bahnson intelligence file, a medieval aerial-reporting edict, the speaker’s exact 1947 printout, Vatican or Magenta custody, and a Stephenson–Josephine Brown telegram. The Puharich scan was retrieved and described. The March 1997 Klass comparison of the Schulgen memo was read. |
| 221 claims with no typed relationship | An edge is added only when the wording names both endpoints. |
| Six document cards with no exact public URL | What would close each one is an item-level catalog record for that work, rather than a category page. |
The graph currently holds 3,455 typed relationship records and 3,385 unique directed relationships. Different roles and directions stay separate. A claim that supports more than one role is counted once in a pair total.
How to read it
Hosted edition:
jorjani-atlas.vercel.app.
Source:
github.com/Lucastil2212/jorjani-atlas.
Open Coverage and read the section titled “What this atlas still does not establish” before treating any
neighborhood of the graph as a result. The same edition can be checked locally with
python3 scripts/validate.py and python3 -m http.server.