> you can't simply search&replace with machine precision
Sure you can. Search and somehow mark the text (underline or similar) to make keywords hard to miss. Then proceed with the manual print, expunge, scan process.
I suppose a global search/replace to mark text for redaction as an initial step might not be a bad idea, but if one needs to make sure it's correct, that's not enough.
Don't bother with soft copy at all. Print a copy and have multiple individuals manually make redactions to the same copy with different color inks.
Once that initial phase is complete, partner up persons who didn't do the initial redactions review the paper text with the extant redactions and go through the documents together (each with their own copy of the same redactions), verbally and in ink noting redactions as well as text that should be redacted but isn't.
That process could then be repeated with different people to ensure nothing was missed.
We used to call this "proofreading" in the context of reports and other documents provided as work product to clients. It looks really bad when the product for which you're charging five to six figures isn't correct.
The use case was different, but the efficacy of such a process is perfect for something like redactions as well.
And yes, we had word processing and layout software which included search and replace. But if correctness is required, that's not good enough -- a word could be misspelled and missed by the search/replace, and/or a half dozen other ways an automated process could go wrong and either miss a redaction or redact something that shouldn't be.
As for the time and attention required, I suppose that depends upon how important it is to get right.
Is such a process necessary for all documents? No.
That said, if correctness is a priority, four (or more) text processing engines (human brains, in this case) with a set of engines working in tandem and other sets of engines working serially and independently to verify/correct any errors or omissions is an excellent process for ensuring the correctness of text.
I'd point out that the above process is one that's proven reliable over decades, even centuries -- and doesn't require exact strings or regular expressions.
Edit: Fixed prose ("other documents be provided" --> "other documents provided").
According to a random dictionary I found:
To trump. Verb. Surpass (something) by saying or doing something better.
> that's very difficult to unintentionally screw up.
You've already screwed up by leaking length and risking errors in manual search&replace
These are roughly equivalent. The point is having a hard copy in between the digital ones.
They're equivalent in security. The digital method is more convenient (albeit more error prone). What confers the security is the print-scan step. Whether one is redacting in between or before doesn't change much.
You'd still want to do a tabula rasa and manual post-pass with both methods.
> point was about the chance of making mistakes in redaction
Best practice is humans redacting in multiple passes for good reason. It's less error prone than relying on a "smart" redactor, which is mostly corporate CYA kit.
They aren't, security is defined as the amount of information you leak. If you have an inferior process where you're substituting the correct digital match with an in incorrect manual match, you're reducing security
> albeit more error prone
The opposite, you can't find all 925 cases of the word Xyz as efficiently on paper without the ease of a digital text search, my guess is you just have made up a different comparison (e.g., a human spending 100hrs reading paper vs some "smart" app doing 1 min of redactions) vs. the actual process quoted and criticized in my original comment
> Whether one is redacting in between or before doesn't change much
It does, the chance to make a mistake differs in these cases! Printing & scanning can't help you here, it's a totally set of mistakes
> Best practice
But this conversation is about a specific blogged-about reality, not your best practice theory!
The fact that this release process is missing this key step is significant too imho. It makes it really clear that the people running this didn't understand all of the dimensions involved in releasing a redacted document like this and/or that they weren't able to get expert opinions on how to do this the right way, which just seems fantastical to me given who we're talking about.
In other threads people are discussing the possibility of this being intentional, by disaffected subordinates, poorly vetted and rushed in to work on this against their will. And that's certainly plausible in subordinates but I have a hard time believing that it's the case for the people running this who, if they understood what they were tasked with would have prevented an entire category of errors by simply tasking subordinates to do what you described regardless of how they felt about the task.
So to me that leaves the only possibility that the people running this particular operation are incompetent, and given the importance of redacting that is dismaying.
Regardless of how you feel about the action of redacting these documents, the extent to which it's done and the motives behind doing it, the idea that the people in charge of this aren't competent to do it is not good at all.
So it's likely there simply isn't the time to do extended multi-step redactions.
What's happening is a mix of malicious compliance, incompetence, and time pressure.
It's very on-brand for it to be confused, chaotic, and self-harming.
It's clearly a superior process that provides ease of use, ease of understanding, and is exceedingly difficult to screw up. Barr's DoJ should be commended for having selected a procedure that minimizes the risk of systemic failure when carried out by a collection of people with such diverse technical backgrounds and competence levels.
Notably, had the same procedure been followed for the Epstein files then the headline we are currently commenting under presumably wouldn't exist.
It has direct relevance since it describes the process as lacking the automated search and replace
> I refuted that
You didn't, you created a meaningless process of underlinig text digitally to waste time redacting it on paper for no reason but add more mistakes, and also replaced the quoted reality with your made up situation to "refute".
> and is exceedingly difficult to screw up.
It's trivial, and I've told you how in the previous comment
> Notably, had the same procedure been followed for the Epstein files then the headline we are currently commenting under presumably wouldn't exist.
Nope, this is generic "hack" headline, so guessing a redacted name by comparing the length of plaintext to unmask would fit the headline just as well as a copy&paste hack
Can't leak a file that doesn't exist.
And the "obvious word needle in a haystack of many thousands of pages" isn't as self-healing as you appear to think it is.