Release 4 broke a three-release streak: PDFs now carry agency provenance in the /Author field
Every PDF in Releases 1, 2, and 3 had a scrubbed (empty) /Author metadata field — 244 documents, zero authorship attribution. Release 4 breaks the pattern: all 14 R4 PDFs carry an /Author value naming the originating agency ("CIA", "Department of Energy", "Department of War", "FBI"). This is a small but real policy change in the DoW's declassification pipeline.
For three releases, the DoW consistently blanked the /Author field on every PDF — a metadata-hygiene practice consistent with the R1/R2 findings of scrubbed /Producer strings and empty /Author bytes. R4 abandons that: it starts stamping agency names into PDF metadata. Small change, but load-bearing. It means every R4 file can now be programmatically grouped by originating agency without parsing the catalog CSV, and it makes downstream re-hosts (Internet Archive mirrors, academic search) surface agency provenance in their own indexes automatically.
By the numbers
Evidence
Method. Ran fitz.open(path).metadata['author'] across every PDF in each release's local mirror.
Counts:
| Release | PDFs analyzed | PDFs with non-empty /Author | Distinct /Author values | |---|---|---|---| | R1 (May 8, 2026) | 110 | 0 | 0 | | R2 (May 22, 2026) | 6 | 0 | 0 | | R3 (Jun 12, 2026) | 52 | 0 | 0 | | R4 (Jul 10, 2026) | 14 | 14 | 4 (CIA, Department of Energy, Department of War, FBI) |
R4 /Author values (all four distinct):
- CIA — used on 2 files (CIA-UAP-D020 and D021 — 1955 memoranda on unconventional aircraft)
- Department of Energy — used on 2 files (DOE-UAP-D004 Los Alamos 1949, DOE-UAP-D005 Pantex 2015)
- Department of War — used on 9 files (Project Sign, Project Blue Book review, various analyses)
- FBI — used on 1 file (FBI-UAP-D014, 1967-1974 correspondence)
Also new in R4: the /Creator field now shows a single vendor name — Highland Technologies, Inc. — on every R4 PDF. In R1 the /Creator field varied across 13 different pieces of software (HP scanners, PFU/Fujitsu scanners, Photoshop, LuraDocument, PScript5.dll, etc. — see [pdf-creator-fingerprints](/findings/pdf-creator-fingerprints)). In R3 that variety collapsed to 3 values. In R4 it collapses to one: a single named vendor. Highland Technologies is a real U.S. government contractor (there are multiple companies by that name; the DoD context and PDF-processing profile most closely fits the Virginia-based document/records-management firm that services federal agencies). This is the first time a specific vendor's identity appears in the PURSUE metadata.
Cross-reference. See also [tradecraft-cleanup-in-release-3](/findings/tradecraft-cleanup-in-release-3) for the previous cleanup step — R3 removed the historical scanner-fingerprint variety without adding author attribution. R4 goes the other direction: adds attribution, keeps the vendor stack minimal.
PDF metadata scrubbing profile across all four releases
| Release | PDFs | Non-empty /Author | Distinct /Creator | Distinct /Producer |
|---|---|---|---|---|
| R1 · May 8 | 110 | 0 | 12 | 9 |
| R2 · May 22 | 6 | 0 | 1 | 1 |
| R3 · Jun 12 | 52 | 0 | 3 | 2 |
| R4 · Jul 10 | 14 | 14 (100%) | 1 (Highland Technologies) | 1 (Acrobat Paper Capture) |