While the full release of Nunes memo was was approved by President Trump, the White House denied a similar request to publish the minority partys response, citing concerns about the release of classified information. The version of the new memo released to the public was therefore heavily redacted.
While the document still lays out the pieces of evidence used in FISA applications for the surveillance of Trump campaign advisor Carter Page, the individual claims details have been omitted. So while, for example, his meetings in Moscow are mentioned, the specific identities of his contacts at those meetings have been redacted.
The document was released as a PDF that appears to be a scan of a printed copy that was manually redacted using a black felt-tipped pen or text marker.Felt-tipped pen redactions in the released memo.
While this method of redaction effectively obscures the relevant sentence fragments, the texts layout remains unchanged. This opens up the tantalizing opportunity to reconstruct top-secret information by correctly guessing fragments that result in the flow of text after each redaction to align to the released version.
While most redactions appear to be too long to guess and manually check, there is one significant single-word redaction where the process described above leads to success with trivial ease.
On page 3, the first paragraph contains the sentence By then, the FBI had already opened sub-inquiries into [REDACTED] individuals linked to the Trump campaign. The size of the redaction and context of the sentence strongly suggest the classified information to be a numeral.Original release: a single, redacted numeral
In the following illustractions, versions of the line with different numerals as the redacted word are overlayed the original release. Trial-and-error determined Times New Roman at 17pt to be a good fit for the original fiint and size. The original was slightly rotated to provide the best possible fit, since it was apparently scanned at a slight angle. A perfect recreation will diverge little from just the original, with the tested text lying directly on top of the original. Wrong guesses lead to clearly visible offsets. The software used is Pixelmator, a Photoshop-like image editor.
Only the numerals zero to ten roughly fit the size of the redacted word, with zero and one being excluded because the resulting text would be nonsensical. Trying the remaining numerals two to ten results in the following.'two' leads to a bad fit past the redaction Larger view showing the clear tracking error after 'two' 'three' 'four' Larger view showing constant fit before and after 'four' 'five' 'six' 'seven' 'eight' 'nine' 'ten'
It is clearly visible that the numeral four gives us far higher accuracy for the text overlays that follow. We have defeated some three-letter agencys tradecraft, and exposed top-secret information with a $30 graphics program.
I am working on automating this process and trying to find similar hits for the remaining redactions.
Mark Felt-Tipped Feb 25, 2018
Building Tensorflow 1.5.0 on MacOS with CUDA GPU Jan 21, 2018
Contact Jan 1, 2018
Add iterm2 marks to long-running outputs Jun 1, 2017
Trump Documents Atbash Decoded May 23, 2017
Polkit? WTF is this again? Actually... May 23, 2017
lighthouse/chromium/alpine/docker image Mar 9, 2017
Don't know how to build task '# Mar 6, 2017
link_directory must be a directory Jan 2, 2017
Verifying NGINX configurations in ansible May 23, 2016
ANSIBLE-MODULE-PATH and more Apr 29, 2016
'credentials-gcloud.sh' is not a git command Apr 23, 2016
I rule (.svg) Mar 8, 2016
React and Relay error scrapbook Mar 3, 2016
Get a Grid, at least vertically Feb 17, 2016