Which might be exactly, why nobody wants to talk about this. (This may be especially nasty where archives had been digitized and then the analog originals were disposed of. What can you even do about this?)

PS: In order to have learned anything from this, there should be a rule for the digitization of archives containing core documents to use (at least) two unrelated scanning technologies, with unrelated processors and unrelated software from different sources (much like Airbus does it with flight computers). A sole, single scan should attribute to nothing.

> In order to have learned anything from this, there should be a rule for the digitization of archives containing core documents to use (at least) two unrelated scanning technologies, with unrelated processors and unrelated software from different sources

But what do you do in case of conflict? There's no way of resolving it, so for this type of thing to be more effective, you'd need 3 different scans, and a quorum of 2 to decide the real version.

Knowing that the data is unreliable is something already. Then, you can apply heuristics. And, if there is a systematic error, you should be able to detect this, eventually. (E.g., what is the unreliable series, what is the bug and when is it triggered?) But, yeah, this is why I wrote "at least".

I'm pretty sure that Kriesel said in his C3 talk that at least one company called and told him that they scan and dispose of the originals. You can't do anything at this point. You're probably right – they rather just want to forget about it because the only correct solution would be to scrap everything.

My original comment was more aimed at the point that damage would need to be admitted. But they did admit it, which makes it evermore confusing why nothing really happened after that, at least publicly.

I don't think that TMR or similar overhead would be practical. Just ban compression algorithms that do pattern matching and substitution for archival purposes (like the German BSI did).

There's a scene in the Cryptonomicon, where Goto Dengo's (a lieutenant in the Japanese Imperial Army in WWII) convoy is sunk by a US air attack. This involves a new weapon, which also requires a crucial change in tactics – and he knows that the Imperial Army has lost the war: this is not how the Americans are supposed to attack, how could this happen? Notably, this requires that several high ranking figures observe, comprehend and admit that their tactics are ineffective, that they are wrong – and invent and implement something new and adjust their tactics, instead, involving the entire chain of command. This shouldn't happen. They must have all killed themselves. No warrior of principles would admit to such a fault, commit to the shame, and change tactics in the midst of the war…

Given that science progresses one funeral at a time, the practice of hari-kari would seem to allow the Imperial Japanese officer corps to evolve their tactics more rapidly than most.

The problem being that many of our big corporations and institutions are much like an Imperial Army, but without the culture of seppuku. They have overrun considerable territory and established the Greater Welfare Zone, including its glacis and satellites, by adhering to concise, even gruel tactics. This works. This is how it works. If we just stick to the tactics and principles, this will continue to work. This is how we do things… Admitting fault and accepting shame, just to veer off proven and established procedures, isn't only a serious danger to your career, it also endangers the greater good, we've all committed to… There's too much at stake. (Compare how we are all just forced to throw out even the basics of AI safety out of the window. There's no alternative. Let's suppose that Banzai attacks do work…)

> A sole, single scan should attribute to nothing.

That's not much less than a sole, single repository of paper documents that may not be climate controlled or fire protected (or the fire protection may be sprinklers which will ruin most of the documents anyway).

Document digitalization is a cost saving program, but it doesn't save money if you need to have two vendors, validate the vendors are actually separate and remain actually separate, have a comparison process and an exception process. If the requirements are too high, paper remains and nobody uses the documents because the retrieval costs in labor and time are too high.