Orientation

How to read this archive

Before drawing conclusions, it helps to know what this dataset is — and what it can and cannot support. The single most important point: the reliable signal here is in the record and the disclosure process, not the phenomenon. The footage is too low-resolution and the sample too biased to settle what the objects are; but the documents tell us a great deal about how the U.S. government collected, withheld, and released this material.

Six things to keep in mind

the dataset's character

It is disclosure-driven and bimodal

The 212 records cluster in two eras — 118 before 1960 and 52 after 2000 — with a ~40-year gap when no standing program collected. The shape of the data follows releases, not sightings.

It is a U.S.-government corpus

64% of geolocated records sit in the Western hemisphere, concentrated where the U.S. operates. The map is a footprint of where the government looks and mandates reporting, not where the phenomenon is.

Rich shape is testimony, not measurement

Mid-century eyewitnesses describe discs and saucers; modern sensors resolve only points. Shape detail falls as instrumentation rises — it is a property of the account.

Almost nothing was resolved

60% of records are unresolved; the agencies explained only 19. Most of this material was never closed.

Redaction tracks non-resolution

The most blacked-out cases are the unresolved ones — the record is most opaque exactly where it is least explained.

The phenomenon side is underdetermined

Single-band, uncalibrated infrared at unknown range cannot give absolute size, temperature or distance. Resolved-object work surfaces shape and motion where the optics allow — never beyond.

What happened to the cases

resolution funnel

all records

pre-1960 · 118

1960–99 · 29

2000+ · 52

explainedunknownunresolvedredactedunstated

The modern (2000+) set is almost entirely unresolved — the better-instrumented era leaves more cases open, not fewer. Note the selection effect: the modern records are a curated military UAP collection, biased toward exactly the cases that were not explained away.

212 records is not 212 events

same-event composition

The corpus includes compilations and cross-referencing documents, so some records describe the same encounter or closely-related ones. By text similarity, 50 records fall into 22 clusters of two or more — read those as related, not independent, so a pattern isn’t double-counted.

6 records · 1x UAP observed via FMV during NAVCENT ISR mission — Arabian …3 records · UAP sighting 1 — Persian Gulf/Arabian Gulf area, 1830Z 16 Ju …3 records · Apollo 17 crew reports cosmic-ray light flashes in cabin — t …2 records · Football-shaped object with throbbing sound — Sterling, Utah …2 records · Multiple round flying objects at ~6,000 mph — Mt. Adams, Ore …2 records · Maury Island 'flying disc' fragments and B-25 crash — Maury …

A known gap — the big files are under-read

document coverage is a floor, not a census

The largest mid-century compilations — the FBI HQ 62-83894 case-file sections, roughly 2,500 pages — mix typed memos, newspaper clippings, cursive correspondence, and faint scans. Automated OCR garbles clippings, handwriting and poor scans, so these files are under-extracted: the 212 structured records undercount the reports and material the documents actually contain, especially the handwritten mid-century pages. (Use full-text search to see the raw text directly.) A vision-based transcription pass is planned to recover them. Until then, treat document coverage as a floor, not a complete census.

In short: trust the archive about itself — its timing, its geography of collection, what it withheld, what it left unresolved. Treat everything about the objects as provisional and resolution-limited. The deep views (Objects, Patterns, Behavior, Map) carry their own confounds inline.