Every file in the PURSUE release carries marks that have nothing to do with the phenomenon: wrong dates burned into frames, scanner noise, an analyst's circle, a censor's bar. These artifacts are usually treated as defects. They are actually evidence, about the cameras, the custody chain, and the release itself. This page teaches you to read them.
Twenty-four of the FBI's photo-submission files share one impossible date.
12-31-1999 23:5xAcross the FBI photo-submission series, 24 of the 32 files carry the same burned-in sensor timestamp: December 31, 1999. The photographs are from the 2020s. The date is not a mystery and not a cover-up: it is what a consumer camera, typically a battery-powered trail camera, stamps on every frame when its internal clock was never set and reverts to the factory epoch.
Read forensically, that one artifact says three useful things. The images came from ordinary consumer field equipment, not from professional or government sensors, which is consistent with their provenance as civilian submissions to FBI field offices. The submitters were not curating for credibility: someone staging evidence tends to fix the obvious tells, and a wrong date on every frame is the most obvious tell there is. And the government released the frames as received, wrong clock included, rather than normalizing or re-dating them, which is exactly what you want a custodian to do with evidence.
It also carries a warning for the reader: the burned-in date on any frame in this corpus is testimony from a camera's settings menu, not from the sky. Dating these images relies on the FBI's case metadata, not on the pixels.
43 historical files are flagged OCR-degraded. The degradation itself has a stratigraphy.
The Cold War documents in the corpus were typed, carbon-copied, microfilmed, photocopied, and finally scanned, and each generation left sediment: halftone moiré from printed pages, hole-punch shadows, skewed lines from a hurried platen, the ghost of a rubber stamp bleeding through from the reverse side. In 43 of the corpus files the accumulated noise is heavy enough that machine text-recognition partially fails, which is why their spreadsheet entries carry bracketed editorial notes instead of clean verbatim quotes.
The forensic value runs in both directions. Degradation certifies age and custody: a 1949 memo that arrived pixel-crisp would deserve more suspicion, not less. But it also means keyword search silently fails on precisely the oldest material, and any claim that "the word X appears nowhere in the corpus" is really a claim about what survived five generations of copying. The full-text searchability of this record improves as it gets younger; conclusions about the historical layer should lean on reading pages, not querying them.
Six Tranche 1 files were modified by the government before release, openly.
Six files carry the release's most easily misread artifact: circles, arrows, boxes, and inset enlargements added to imagery to "highlight areas of interest." These are government annotations, drawn during analysis, and the release notes say so. They are not part of the original capture, and they are emphatically not the phenomenon: a gold circle around a light says an analyst considered that light worth your attention, nothing more.
The reading habit this trains is general: for every marked-up frame in this corpus, ask what the unmarked original showed. Sometimes the marking answers a question; sometimes it manufactures one. Wherever this site reproduces an annotated frame, the annotation stays visible, because the analyst's choice of what to circle is itself part of the record of how the government handled these cases.
32 imagery files were redacted before release. The bars have a grammar, and it is measurable.
Thirty-two Tranche 1 imagery files had black redaction bars applied to the originals before publication, mostly covering case numbers, names, and location strings burned into frames or headers. Documents carry the richer version of the same artifact: every textual removal cites a legal exemption code that states why the material was withheld, which means the shape of the secrecy can be read without seeing inside it.
That analysis has its own home on this site: the corpus-wide census of exemption markers, the finding that the substantive redaction is overwhelmingly operational (military plans and installations) rather than intelligence sources-and-methods, and what that fingerprint reframes. It is the deepest artifact in the release, and the one that behaves most like data.
The full redaction analysis, 791 exemption markers, the 1.4(a) fingerprint, and the 36 files whose black boxes hide substance, is on Key Documents: The shape of the wall.
Not all secrecy is FOIA. Tranche 4 carries a marking with a different legal engine entirely.
The modern military documents redact under executive order and cite FOIA exemption codes. The Pantex incident report (DOE-UAP-D005) carries a different stamp: UCNI, Unclassified Controlled Nuclear Information, withheld under Section 148 of the Atomic Energy Act of 1954, not the national-security classification system at all. It is a reminder that the phenomenon crosses jurisdictions, and each one marks its files its own way. A DoD mission report and a DOE nuclear-plant report can describe the same kind of silent, propulsion-less object and still be governed by two separate secrecy regimes.
The oldest Tranche 4 files pose the opposite problem. The 1955 CIA memoranda (CIA-UAP-D020, D021) are TOP SECRET cover sheets filled out by hand, and their bodies resist machine transcription entirely, not because a censor blacked them out but because the era’s intelligence handling predates the typewritten form. Legibility, here, is its own kind of redaction, and it is why a handful of quotes on this site are flagged for manual verification against the source scan.
Even the file numbers testify. The official count is 330 records; the official files number one fewer.
The government's database lists 330 records: 189 documents, 27 images, 103 videos, 11 audio files. Anyone who mirrors the release, as this site does, discovers the audiovisual holdings come to 94 files, not 95, because two video records, PR057a and PR057b, share a single hosted file. It is a trivial quirk with a useful lesson: the release's records are catalog entries, not files, and reconciling the two is how you verify a mirror is complete. This site's holdings reconcile exactly against the official database, quirk included.
The videos also carry public military identifiers (VIRINs) and platform upload numbers, which is what lets any reader confirm that a clip embedded here is the same object the government published, byte for byte, rather than a re-encode circulating on social media. Provenance in this corpus is checkable arithmetic, and the arithmetic is part of why the collection can be trusted at all.