Hello, I'm Jameson Lopp, the CTO and co-founder of CASA, and today you're gonna get a brain
dump from the past four years of experience that I've had while researching and stress
testing over 70 different metal seed phrase backup devices.
I've come across quite a few different variations of designs for these things, and after putting
them all through the wringer, I've come to a number of different conclusions about which
ones are in fact the best for you in terms of robustness and in terms of user experience.
So Satoshi Nakamoto had an excellent quote many years ago, where they made the point
that if you lose your coins by which you lose your private keys, and of course there's a
million ways to do that, then you're essentially making a donation to everyone else in the
ecosystem who has not lost their money.
This is of course something that I've heard from a few people is in their plan to do when
they pass away to essentially let that value go out to the rest of the ecosystem, but I
suspect the vast majority of people do not want this to happen to their bitcoin.
So we're gonna go into some of the things that you should think about if you're trying
to prevent that from happening.
We should be specific here, and we're not gonna cover absolutely everything that could
go wrong with managing your private keys, but rather we're going to talk about how
to best store a backup of your private keys, your seed phrase, so that if something happens
to your regular wallets, whether it's software wallet, hardware wallet, whatever, you can
reconstitute that wallet from the backup.
And this of course means that if your device is destroyed, bricked, whatever, you'll be
able to load the seed phrase into a new wallet.
We're not gonna be talking about issues of physical attacks, it's a very different type
of problem set, and I have presentations and extensive writing about that, but rather what
we're worried about is the more common form of loss.
Some sort of disaster that can result in destruction of data, and we want to be able to store our
data in a way that is essentially disaster proof.
Most commonly of course, as you can see here, fire, flood, earthquake perhaps, causing collapse
of your house, these threats of course will vary depending upon your own living situation.
So I'm trying to be as general and widely practical as possible here.
So what am I specifically gonna be focusing on?
We're gonna talk about fire, house fires are actually quite common around the world.
We're gonna talk about corrosion.
This can usually be a result of flood or some sort of water leak, or in some cases there
may be various chemical agents that can cause that.
And of course we'll talk about crushing, which could happen as a result of perhaps a building
collapse in which your device is stored.
So if we're looking at the threat of house fires, this is a pretty well understood problem.
And we can very safely say that a house fire is probably not going to get hotter than about
1500 degrees Fahrenheit.
And so the stress tests that I do generally are going to aim for about 2000 degrees Fahrenheit
because if your device can survive that, it's gonna survive any natural house fire.
There may of course be other things that could happen like if you have jet fuel laying around
that may burn at a hotter temperature, but I don't think that's a common threat that
the average person needs to worry about.
Now if we're looking at the different material design and the best materials that we should
be using to protect ourselves from common threats, this is also a fairly well understood
issue.
Metals have been vigorously tested for all the different properties throughout the course
of human history.
And if we're looking at melting point, which is of course what we would be worried about
if we have some piece of metal with some data imprinted upon it and we want to make sure
that can survive a house fire, we of course want something with a melting point higher
than the average house fire and probably even higher than that, you know, you want a nice
safety margin.
So if we're looking for something with more than 2000 degree Fahrenheit melting point,
you might be able to get away with copper or gold.
Of course gold is going to be incredibly expensive.
Nickel does really well, steel does really well, tungsten does extremely well, though
tungsten is not something that you're going to come across very frequently and probably
has other properties that would be less interesting to us.
So what if we then look at corrosion resistance?
Well, there's something called the Galvanic scale, which gives us a pretty good idea of
which materials are more likely to react with chemicals or corrode as a result of water
or other liquids being passed over them.
Once again, gold does pretty well here, but probably not something you want to use due
to the other material properties, similar with graphite and silver, they tend not to
be as resistant to crushing and pressure.
So we see that stainless steel actually does quite well when it comes to being resistant
to corrosion.
So that's another good point for stainless steel on that factor.
Finally, what if we're looking at crush resistance?
Well, there are actually a number of different ways that you can measure hardness or flexibility
or whatnot, but in general I would go with something like the Rockwell scale.
We can see here if we're looking for something that's particularly durable, at the high end
of the scale we've got cast iron, we've got stainless steel, we've got titanium.
And cast iron is probably not something you're going to want to use because I believe it's
not very corrosion resistant.
So once again, stainless steel keeps popping up in all of these categories.
If we really want to narrow it down, the best material choices that you're going to be left
with are stainless steel and titanium also works really well.
Both of these materials have very high melting points, have high corrosion resistance, high
tensile strength, and high hardness.
Now you can start going down a rabbit hole because there are dozens of different types
of stainless steel and types of titanium that have slightly better corrosion resistance.
If you really want to get into it, I would recommend getting a marine grade of stainless
steel.
But in general, these are going to be far better than almost any other common metal
that you might come across.
So how did I get to the conclusions that we'll be coming to throughout the rest of this presentation?
Well I did a lot of independent testing.
And one of the first things that I have done is this heat and fire testing.
Basically use a very large propane blow torch that was creating a flame that was probably
five, six inches in diameter and would be able to heat up over 2,000 degrees Fahrenheit
based upon the temperature probe that I got and used to once again test even my testing
equipment.
So here you can see a couple of examples just to show you these stainless steel devices
getting red hot as a result of me putting them directly in this fire for about 10 minutes
or so based upon my experience.
If one of these devices can last more than 10 minutes in 2,000 degrees, it's going to
last basically indefinite amount of time at that level of heat.
But while a lot of these devices performed well, a number of them, I would say probably
30% in fact failed.
So what happens when you have a device that starts to fail, these devices that I'll show
you the photos of, they only lasted a couple of minutes and then I saw them starting to
fail and thus pulled them out so that there was at least something left for me to get
a photo of, otherwise it would have been just a heaping pile of slag.
So here we have a device that claimed to be durable to house fires, but upon stress testing
and closer inspection, we can see it's not made of stainless steel.
In fact, the reason this device was so cheap, I think it was only $30 or $40, is that it
was made out of aluminum, which is a more common, cheaper metal that's easier to manufacture.
But the flip side is its melting point I think is only around 1,300, 1,400 degrees.
It's less than your average house fire temperature, not something that you would want to put all
of your life savings on.
Here's another one that actually claimed to be stainless steel, and it was half right
in the sense that the tiles were made out of stainless steel, however, the device itself
was once again made out of aluminum, the whole thing melts into slag and you can very easily
lose all of your data as it sort of disperses.
Now here is a device that's actually 100% stainless steel, however, there's just a problem
and this is a very common design that you've probably seen, but there's a problem in the
way that the data is secured, and you're basically sliding these tiles into some racks that are
sandwiched into the device, and the problem arises due to the fact that the heat of course
can cause metal to expand and contract, and one of the things I do on the stress test
is in fact I dump it into a bucket of water at the end to simulate a firefighter coming
in with a fire hose to make sure we get the full experience, and this expansion contraction
can cause the device to loosen up and create gaps, and if it's a tile device, the tiles
may in fact fall out and you essentially have data loss, even though the data itself is
not technically destroyed, it's just moved around and you can no longer necessarily tell
what it is.
Here's another example of a similar design, really I think I've seen a half a dozen different
designs of these tile based products and only one or two of them managed to survive all
of my tests and what it basically came down to was the tightness and tolerances of the
tiles and the rails they were going into.
If it's not extremely tight with almost no give and no area for there to be expansion
and contraction, you may have problems if it's not extremely tight.
So what's the next set of tests I did?
Well, corrosion, which you would normally think of as rust or oxidation or whatever,
but I don't have time to be waiting weeks or months to see how long it will take for
rust to oxidize or destroy these devices.
So I actually use muriatic acid, which is a common concrete etcher and pool cleaning
supply to basically do an accelerated corrosion test.
This is extreme, but once again, for doing stress tests, we want to know, can this device
tolerate an extreme environment?
If we know it can tolerate extreme environments, then it's going to do fine in any more common
scenarios.
So what's some of the issues that happened here?
This particular device actually survived the heat test fairly well, but upon corrosion
testing, I discovered all of the screws and the fasteners in the device were not made
out of the same type of stainless steel.
So all the stainless steel pieces survived.
Unfortunately, the fasteners dissolved and the device, as a result, completely fell apart
and all the data was scattered.
So you have to worry about every piece of the device, making sure all the materials
are the same level of robustness.
Here was another one.
This was aluminum device again.
Once again, I kind of fished it out, otherwise it would have been completely consumed by
the acid.
It was already and mostly consumed here.
So aluminum, once again, not very good on the galvanic scale.
This is the same aluminum device that I showed a few slides earlier where, once again, did
not do well on this test either.
So aluminum is a very bad choice overall.
Here is a stainless steel grid plate design.
And in general, the grid plate designs, of which there are quite a few out there, tend
to do the best on my test due to their simplicity and robustness.
However, one thing that's worth noting here is that while the data itself, which is this
center punched divots, which I'm also a big fan of, the data is still there.
However, the problem, as you can see, is that the template that is etched on the device
is fading away.
And so you also have to be very careful that even if the device itself survives and the
data survives, you need to be able to actually read and interpret the data.
You need to have that template that's on the device also be legible.
So yet another issue that you need to worry about when you're looking at these devices
and trying to decide the best one to use.
Finally, I do a crush test.
This is, once again, fairly extreme.
I use a 20-ton hydraulic press, a shop press, and just try to figure out what's the weakest
point or which way can I potentially deform it to cause some sort of data loss.
Here's an example of a device that actually did quite well.
Even though you can see it deformed, the data itself is still quite legible.
So if we're looking at devices that don't do well, once again, the rail and tile-based
devices, if you deform them, they are most likely going to separate and you're going
to have data loss.
Finally, here's a device that performed extremely well, and there was no technical data loss.
But what happened was that I realized that you actually also, from a usability standpoint,
need to worry about how do you read the data from the device after it's deformed.
And so this is a tube, and you basically have a rail that you slide into the tube and it
screws in.
Well, if it's deformed in this way, you can imagine you can no longer unscrew that cap
because the rail goes all the way down the length of the device and it essentially gets
completely seized up.
So data is still there, but you can't open the device without power tools.
You need some sort of vise and extremely high-end saw that would be able to very delicately
saw through this capsule so that you could pry it open and actually get the rails of
tiles out of the device without destroying them.
So yet another consideration that should go into your thought process.
So what are some of my takeaways?
Well, size and form factor certainly seem to be one of the things that you should be
concerned about.
Particularly, it's not like the total volume or total size, but a smaller, more compact
size does seem to be less likely to deform.
But it's not just about the total size.
It's about sort of the thinness.
Look at the width versus the height versus the thickness.
And the wider and higher the device is, the thicker you're going to need it to be so that
it's less flexible, less likely to be deformed.
And if you're doing a device out of stainless steel or titanium, what I found is you probably
want a minimum of five millimeters of thickness for the most robust metal that is not going
to deform.
Here, I have one of the best performing devices I've ever come across, though I think it was
over 300 euros.
This is a very large, hefty stainless steel pipe that gets sealed on both ends.
And this thing did not even flinch when I put it under 20 tons of pressure.
So cylinders and rods tend to do very well in general.
Also, just a single really, really thick plate of metal will do quite well.
Even if it deforms some, you'll be able to read the data off of it.
Also, as you saw in one of the acid tests, the fasteners, if this is a device that's
comprised of many different pieces, you're probably going to have some sort of fastener
that keeps the pieces together.
People need to worry about what are they made of and what are their various properties.
If they dissolve, will the whole thing fall apart?
I've also seen fasteners that will melt or fuse to the device because they're made of
cheaper different materials.
Really most commonly, what I've seen is just the fasteners completely seize up on the device
and prevent you from reopening it and being able to read the data.
That's more of just a usability issue.
But I would say the best practice is to avoid devices that require fasteners.
If you want to follow the KISS principle, which is keep it simple, stupid, then you
should not have fasteners at all.
But if you do need them, use fasteners that are just a screw rather than a screw on one
end and a nut on the other end.
It's definitely better if you have drilled threaded holes that go directly into the device
that the fasteners screw into rather than having yet a second piece on the other end
that could fail.
There's also some dilemmas when it comes to the material choice for fasteners.
For example, titanium will gall with itself and potentially seize up, like I mentioned
earlier.
Steel fasteners are more likely to galvanically corrode over time if they are mixed with a
different material like titanium.
You generally don't want to mix different types of metals if they have different corrosion
resistance.
Also, if you're using a threaded fastener, then I would recommend using one of a hardened
material where the head of the screw is as large as possible.
And the reason for this is that I've seen many dainty fasteners end up just getting
stripped when you try to open it back up.
I've seen this happen for Phillips screws, hex screws, star screws, pretty much everything.
So if you really, really have to use a screw, I would recommend using a flat head and making
sure that the head of that flat head screw, like the groove in it, is extremely deep so
that you'll still have something to grab onto even if it experiences some melting or corrosion.
So another issue to worry about, as you've seen in some of the photos, is just the total
number of pieces.
More pieces means more potential points of failure.
We want to keep it simple.
So you have to think, how bad would it be if this device was damaged in a way that the
data itself got jumbled up or separated?
And this is sort of a combinatoric issue.
So think about it in terms of the number of tiles you have.
If you have 24 tiles that are just the seed words and you lose the ordering, then you're
going to have potentially up to six times 10 to the 23rd combinations of seed phrases
that you might have to try to brute force in order to find your actual seed phrase.
That's if you don't know the ordering.
If you do know the ordering but you have lost tiles, then for every tile that you've lost,
you'll have to try 2,000 combinations.
And of course, for each one, you're going to have to multiply that.
So if you've lost half a dozen tiles, then that becomes so computationally challenging
that you're probably never going to be able to brute force it and find the words that
were lost.
On the other hand, if you're using one tile per letter, which is a more common thing,
and those get jumbled up so you'd have like 96 tiles total for 24 words, that's 9.9 times
10 to the 149 possible combinations.
It's pretty much impossible for you to ever brute force all of that to find your actual
data.
So long story short, if you're using many metal tiles, then you probably only want to
use one tile per seed word, and preferably you want to have the words order number on
the tile itself.
In case they get jumbled up, you can easily put them back in order.
Now what about all the different ways that we can inscribe or store data onto metal?
I've tried quite a few different ones.
This order is in terms of usability.
So you can, of course, get the manufacturer to engrave stuff.
I would certainly not recommend that just due to a security and trust issue.
These pre-engraved tiles are quite common.
They're very easy to slide in and out, but on the flip side, they're very easy to come
apart.
Same thing with the tiles that are stamped.
My two favorite methods are electrolysis, which you can see on the right here, which
is basically using a saltwater solution with a battery to etch your data into a block of
metal.
It's very easy to do and kind of fun, though I've only ever come across one kit that does
that.
Of course, you can always create your own if you know what you're doing.
The center punch method, which is commonly used for the grid type devices, is also very
easy to do and tends to hold up quite well.
Then there's also different types of electric engraving.
It's kind of annoying and loud to do and it's hard to get it etched in particularly deeply.
And then a really common form of data inscription that you see is stamping.
And I really hate stamping.
It hurts your ears, your hands, there's potential to injure yourself while you're doing it.
It's not user friendly because if you don't have the stamp aligned perfectly and strike
it perfectly, you can end up with light strikes or double strikes or stamp bits flying across
the room or even in some cases, just I've had the stamp bit aligned incorrectly and
had a stamp, a letter that ends up being on its side or upside down or something.
So stamping is not user friendly.
If you do it right, it can be very robust, but I do not recommend it for the average
person.
Now in terms of legibility as well, one thing that we're seeing is shrinkage in these devices
to make them more compact and easier to hide.
But you have to be aware that in some cases, such as this device, it can be difficult to
inscribe the data because you have a very small target that you're trying to hit.
And it can also be very hard to read the data back if you don't have great eyesight.
One conclusion that I came to in my recent round of tests with this particularly compact
data was that the result of corrosion can actually cause those divots to enlarge slightly.
And on the vast majority of devices, that's fine because you have plenty of margin around
the grid where you're putting these divots in with your center punch.
But once you start to cluster your data points so closely together, you can see that there's
actually some ambiguity if the holes enlarge and they start overlapping and taking up multiple
possible data points.
So this is, once again, creating a situation where you may have to brute force a number
of different possible combinations of your seed phrase if you get into this situation,
which would not be fun.
So if you want full results of all 70 tests, I have them at seedstoragereviews.bitcoin.page.
The vast majority of the devices I've tested have done quite well, but as we can see here,
there are a number that have had catastrophic failures and you would not want to entrust
your life savings to.
So what are my TLDR conclusions to this half hour presentation?
Well, you want to go with either stainless steel or titanium device.
You want a small but thick form factor that will resist deformation, preferably a single
solid piece of metal so you don't have to worry about fasteners dissolving or melting
or coming apart and having data get strewn all over the place.
And you either want to use a straight punch or electrolysis etching method to inscribe
the data on the device.
And in the case where you need a template to decode the device, you need to make sure
that that template is similarly inscribed deeply into the device so that it won't get
erased if something happens to it.
So that's my takeaways from quite a few hours of testing various backup devices and I hope
that you will find that helpful when you're looking to devise your own backup solution.
Thanks for listening.