DISCOURSE INDEXA MACHINE READING OF THE DISCOURSE
SOURCES ▸ VIDEO · DOCS · PAPERSUTC --:--:--Z

VIDEO

DOCS

PAPERS

BIG PICTURE

A machine reading of the field's collective voice — thousands of hours of testimony, declassified files, peer-reviewed papers and archival records, with every attributed claim extracted and mapped where the discourse agrees, splinters, and shifts.

We index what sources say, not what happened — every figure a resolved count, weighted by how independently a claim recurs.

HOURS ANALYZED

774

1,075 VIDEOS TRANSCRIBED

SOURCES INGESTED

11,386

1,075 VIDEOS · 10,311 DOCS

CLAIMS EXTRACTED

11,009

ATTRIBUTED & GRADED

RECURRING DETAILS

TRACKED ACROSS THE ARCHIVE

STRONG LINKS

659

DETAILS THAT APPEAR TOGETHER

02 / THE FIELD MAP

How the details connect

Each dot is a recurring detail that keeps coming up in the reports — a shape, a being, an effect. Dots are grouped and coloured by theme; a line joins two details that tend to show up together, and the heavier the line, the stronger that link.

FR-SYNTH · WHICH DETAILS APPEAR TOGETHER

SIGHTINGSBEINGSABDUCTIONCONSCIOUSNESSDISCLOSUREPHYSICAL EFFECTS

DETAILDISCLOSURE

Multiple witnesses

Comes up in 1,418 reports — about 24.8% of cases in the archive (give or take a little). It's strongly linked with 27 other details that tend to show up alongside it.

SEE TAGGED CASES →

HOW OFTEN IT COMES UP24.8%

SHADED BAND = OUR MARGIN FOR ERROR (18.1–32.6%)

COMES UP IN

1,418 reports

STRONGLY LINKED TO

27 details

OFTEN SHOWS UP WITH

03 / BY GROUP

Pick a group to explore

Each group gathers related details. Choose one and the “most common” list below narrows to just that group.

04 / MOST COMMON

The details that come up most

65 details across every group, ranked by how often they come up.

#DETAILHOW OFTENLINKS

— = NO PAIRING CLEARS THE SIGNIFICANCE BAR (FDR q < 0.05)

05 / HOW IT WORKS

How the archive is built

From raw sources to the maps and rankings on this site — the same steps, applied to everything, with nothing asserted as true.

OUR OWN ANALYSIS ENGINEPurpose-built, not off-the-shelf — machine reading plus custom statistics, applied to every source the exact same way.

01INGEST

What goes in

Interviews and documentaries, declassified government files, peer-reviewed papers, news and archives — pulled in and transcribed word-for-word.

11,386

SOURCES GATHERED

02MACHINE READING

We read every word

Our language models read every source against the same template — pulling out each claim, who made it, where, when, and what they described.

11,009

CLAIMS PULLED OUT

03RESOLVE & SCORE

We connect the dots

Our own statistical models merge duplicate accounts into one case, weigh how independently it recurs, and test each pattern against a control.

15,727

CASES PIECED TOGETHER

04MAP & RANK

What you see

Everything is graded, ranked and mapped — the field map, case files, timelines and studies on this site — never asserted as true, only counted.

RECURRING PATTERNS MAPPED