I did some data munging to match the latest release of JFK assassination records to their previously released version to produce a pop-out effect of the newly declassified parts. The resulting over two thousand documents, spanning eighty thousand pages are now all posted on github.

As of March 20th, there is a total of 2,343 documents released by the U.S. National Archives and Records Administration (NARA) in JFK Assassination Records - 2025 Documents Release.

The total page count that NARA indicates that there were a total of 77,100 pages released, though in downloading all 2,343 files and summing their page numbers, I get a higher total of 82,864 pages.

Wouldn't it be nice if someone went through to highlight the newly un-redacted parts of previous reports? Well, dear reader, I got you!

But first let me help you get oriented.

*_fresh vs *_highlighted

There are two flavors of PDFs that I've generated - _fresh and _highlighted. The _fresh versions contain only the pages where there is something different, and each page has some text at the top indicate what the corresponding page number was in the original released document. There's also a _highlighted version, which contains whole document, with the highlighting of differences in their entire context. The notes at the top of the highlighted document (and top of each fresh page) are in light blue so as not to be too intrusive, and you can highlight them to read or copy when you end up seeking out the original, because in cases where the diff-jfk version is not legible.

Now you have the necessary information, if you're ready to dive in, just click below:

>> Happy Sleuthing <<

Otherwise, stick around for some more information and examples.

Unredacted diffs

Turns out that Mary Ferrell Foundation is an excellent resource, see their 2025 JFK Records Releases. Unlike the MFF, which is choosing to redact the social security numbers, diff-jfk is not.

The only visual differences in the diff-jfk generated documents that are omitted from counting toward fresh pages are the very top of the first page document. This is where there would be an expected difference in the banner indicating the year of the release. And also the bottom of each page that was used to keep track of document ID and page number but ended up reproducing slightly different in some of the documents, so I discarded that to get fewer false-positives.

These trivial changes are not informative, so they has been explicitly omitted from "fresh" consideration. So a document that was initially released in 2023, and again released in 2025, would have a red mark in the last digit of the year in the banner on top of its first page, but such a page would not be considered "fresh" if there are no other changes on that first page. You will still see that red mark, however, if there were other changes to the page.

Some samples

Here are some examples of the new information that's been released:

2025/104-10301-10001_fresh.pdf - fresh page 1 has previously censored details of CIA's 1961 budget: $41.5 million spent on Cuban operations, $11 million on the rest of Latin America

2025/124-10114-10010_highlighted.pdf shows you the one previously redacted box, which now reveals the word "Embassy" in between "Paris" and "CAS". CAS is one of the terms of art that I learned in preparing these documents - it stands for Controlled American Source - CIA.

When it works, it works brilliantly. 2025/177-10002-10105_fresh.pdf has just the two pages that have newly declassified information which concerns understanding of Soviet submarine fleet in 1966-68, but if you wish you can peruse the entire 14 page document which now highlights the declassification.

Take for example, 2025/104-10307-10006_highlighted.pdf, where you can learn, among other things about how CIA set up a special facility in Miami to intercept electronically internal communications within Cuba,

details of the 14 countries' spy agencies which "have been most cooperative and helpful" (page 12),

where one of the sources was a former protege of Raul CASTRO , one was and ambassador of a North African country in Cuba, and another was a Cuban ambassador in Western Europe (page 5).

Document 104-10178-10000 contains, among other things, the tales of WKSCARLET-3, with some juicy bits about plans for that Venezuelan to bug the Colombian delegation in Rome during Colombian-Venezuelan border talks in 1973.

The full-context highlighted document has 310 pages, whereas the freshly declassified pages add up to 86. The image above is page 72 in the original document, and appears on page 9 of the fresh-only version.

Limitations of Minimal Magic

I didn't do anything fancy to try to get the pages to line up properly. Also, the approach I took with highlighting works pretty well for pages that were previously redacted in white, but isn't as legible for documents which were previously blacked out. So take this with a grain of salt. Hopefully it helps you identify where to look in the original document if you can't make it out in the diff-jfk version.

Slight rotations, rescans, change to the cropping, and so on, make the previously unredacted parts of the document either hard to read or completely illegible. Nevertheless, despite this limitation, it's still makes it easier to scan the highlighted document to identify which pages to look at in the newly unredacted document. For example, scanning through the otherwise atrocious mess in the diff-jfk version of 2025/124-10179-10220_highlighted.pdf, I was quickly able to come to page 37 that previously had censorship boxes, and looking on the same page of the original, I can see that the August 20, 1968 (Warsaw Pact invasion of Czechoslovakia) example of positive intelligence involved the wiretapping of, previously redacted, "the Soviet Mission to the United Nations". On the next page, phrase "the Romanian Embassy" was omitted from a description of another wiretap, and further down the page the previously censored phrase of "intercepts on various missions to the United Nations and embassies" providing many indicators during Arab-Israeli Six Day War and the India-Pakistan War.

There's more to be done, but I wanted to get this out there and not let perfect be the enemy of good enough. Here's an example of what's supposed to be the same document but having different page numbers. On the left hand side is Page 57 of 2023/docid-32404520.pdf and on the right is page 61 of 2025/104-10332-10007.pdf

By the way, that's definitely one of the recurring themes for what I've come across being previously censored but now reveled in these documents: the use of State department cover. It also came up in, to name just a few: 20225/104-10330-10102.pdf and 2025/104-10330-10083.pdf

How diff-jfk came to be

The first thing we can do is just hope for the best and assume that the filename hasn't changed and exactly matches the previous release (2023).

That gets us 849 files, 36.24% of the total file count with exact matches that were previously released in 2023. Not bad for a first pass, and this haul actually gets us the bulk of the raw-page count, because these 849 files contain 44770 pages, so we're already more than half way there!

To continue on, it's a bit of a slog, because whereas in 2025 all files started off with the 3-5-5 pattern, that's not the case for earlier years. We luckily have a spreadsheet from 2023 which gives us a mapping for record number to filename. There are also such spreadsheets for 2022, 2021, and the 2017-2018 releases.

Duplicate ambiguity

Ok, we also have the problem of ambiguity where from the filename, we have multiple files for the same record number. For example, we have ./2025/104-10105-10290 (C06932214).pdf and also ./2025/104-10105-10290.pdf, there's also 2025/124-10273-10007_multirif.pdf and 2025/124-10273-10007.pdf and so on.

Anyway, I'll spare you the rest of the details, except to say that I got pretty close to finding previous versions of all of the files.

2023:       matching files: 873,    left: 1456
2022:       matching files: 1349,   left: 107
2021:       matching files: 0,      left: 107
2017_2018:  matching files: 13,     left: 94

There are 2,221 files from the 2025 release that I've managed to match to a previous version with the same page count.

Another 26 files where the previous and 2025 release have a different number of pages. A handful of those just lost or gained a page or two, I was able to accommodate and fix, but there's more work to be done to completely close those out.

It's not perfect, but it's a pretty damn good start.

Previously unreleased files

Finally, here's a list of files that I believe are entirely new in the 2025 releases. As there is nothing to diff against, I did not generate fresh or highlight files for them. You can these new files here in their own folder.

Issues and Discussion

If you want to leave a comment, or just come across something cool or interesting that you want to share with others, please open a new thread or join one over in the discussions.

Please report any issues you come across in the diff-jfk data. In particular, I'd like to know if I have mislabelled some of the 2025 releases which I marked as new, so do let me know if there's a prior version of one of those documents, so that I can generate a proper diff of those. I'd also like to know if some of the file contents from NARA have changed. This last part might seem like an odd, but here's are some examples of document id confusion from 2018: multiple record identifiers mapping to the same file name, and the same filename having different version depending on when it was downloaded.

Parting thoughts

It took more time than you might expect to tinker around and connect all of the dots to produce diff-jfk. I hope you enjoy perusing it as much as I do. If you end up finding something interesting, why not share it with others by posting in the discussions. For attribution, please link to either this post or the diff-jfk repository, which you can also star if you've got a github account and are feeling generous. I also enabled sponsorship on that repo if you want to support my work financially.

>> Happy Sleuthing <<

Paul Ivanov