XDRIPACADEMY
Sign in

Curriculum·F104 Wallet Architecture, Done Properly·60 min

Choosing your setup

By the end of this lesson you can

  • Distinguish a guarantee enforced by physics or mathematics from one enforced by a vendor's policy
  • Select a setup from a decision procedure driven by amount, activity and who else is involved
  • Identify the point at which added complexity increases expected loss rather than reducing it
  • Write down your own configuration as a specification you can hand to F104-L1 and to F110-06
AutopsyLedger Recover, May 2023no funds lost, and a widely held belief about the product turned out to be wrong

16 May 2023. Ledger announced Ledger Recover: an optional paid service that would encrypt your seed, split it into three shards held by three separate companies, and restore it to your device after identity verification.

Nothing was breached. Nothing was stolen. The service was opt-in.

The reaction was the loudest in the company's history, and the reason is worth understanding precisely.

To make it work, the firmware shipping on every device had to include a code path capable of packaging seed material and sending it off the device, gated on the user consenting on-device.

That sat against years of messaging that your keys never leave the secure element.

Ledger's chief executive argued the meaningful residual risk was legal compulsion, since subpoenas across the three custodians' jurisdictions could reconstruct a seed. The company delayed the launch and committed to publishing the code.

Whether the risk profile genuinely changed is still argued about, and both positions are reasonable. What is not in dispute is what users learned.

The property they thought they had bought was enforced by the vendor's policy, not by the hardware. And they found out from a press release.

That is the question this lesson exists to make you ask before you buy anything: not what a product promises, but which of its promises would survive the vendor deciding otherwise.

Primary source

You have the taxonomy, the tiers, the setup ceremony, the backup and the drill. This lesson turns them into one decision, and then into a written specification.

Physics, mathematics, or policy

Sort every guarantee you are offered into one of three bins.

Enforced by mathematics. Your seed cannot be brute-forced. Nobody can derive your private key from your address. These do not change, they do not depend on anybody's intentions, and F103-01's arithmetic is why.

Enforced by physics. An air-gapped machine cannot transmit over a network it is not connected to. A metal plate in a safe cannot be read remotely.

Enforced by policy. Everything else. What the firmware will and will not do. Whether the vendor holds a key. Whether the terms transfer title, per F103-03. What the company does when a court asks.

Policy guarantees are not worthless. Most of a good setup is built from them, and a reputable vendor's policy is worth more than a stranger's. They have one property you must account for: they can change, and you will learn afterwards.

So the diligence question is never "is this secure". It is: which bin is each promise in, and what happens to me if every policy promise is withdrawn tomorrow?

Run it on the autopsy. "The key never leaves the device" was believed to be physics. It was policy. Nothing was stolen, and a great many people had built a plan on the wrong bin.

Three inputs

Everything about your choice follows from three facts about you, and none of them is a brand.

How much. Not exactly, but the order of magnitude, and honestly.

How often. Daily activity and a long-term hold are different jobs, which is F104-02's entire argument.

Who else. A partner, an heir, a business, a co-signer. This changes the answer more than the amount does, and F110 is where it is developed.

The procedure

Read down until a row describes you. The tiers from F104-02 apply throughout; this is about what each tier is made of.

SituationVaultDailyBurner
Learning, under a few hundredNot needed yetReputable software walletSame wallet, second account
Meaningful but replaceableOne hardware wallet, paper backup, drilledSoftware wallet on a phoneSeparate software wallet
More than you could replaceHardware or air-gapped software, metal backup in two locations, annual drillHardware wallet, separate seedSoftware wallet
Life-changing, or others depend on itMulti-signature across devices and locations, plus F110-06 written planHardware walletSoftware wallet
A business, or you hold for othersMulti-signature with separated signers and a written policyPolicy-controlled accountNot applicable

Two rows deserve comment.

The first row is a real answer, not a placeholder. A beginner with $200 who is told to buy hardware frequently buys nothing and stays on an exchange. A reputable software wallet with a small amount, used carefully, is a better place to build competence.

The last two rows involve other people, and at that point the technical question stops being the hard one. Multi-signature is a coordination problem before it is a cryptography problem, and F110 exists because most holders never solve it.

When complexity starts costing you

This is the part that the industry, which sells complexity, does not say.

Every component you add reduces one risk and adds a way to lose access. Multi-signature removes the single point of compromise and adds a set of keys to lose and a quorum to coordinate. A passphrase defends against physical discovery and adds an unrecoverable secret. A clever hiding place defends against a burglar and adds the possibility that nobody ever finds it.

The crossover is where the probability you cannot operate the setup exceeds the probability it is attacked.

And F103's survey tells you which side most people are on. 35 percent had lost access. The leading causes were a forgotten password, a lost phrase and lost two-factor access. Not one is an attacker.

Competence before complexity

The correct order is not the one the marketing implies.

A simple setup you have drilled, maintained and can operate under stress beats a sophisticated one you half-configured. A two-of-three multi-signature you have never rehearsed is worse than a single hardware wallet you have wiped and restored twice.

So the honest upgrade path is: get one tier working, drill it, live with it for a few months, and add complexity only when you can state which specific threat the addition removes and confirm you can still recover after adding it.

If you cannot name the threat in one sentence, per F106-01's first question, the addition is theatre with a maintenance cost.

Write it down

The output of this course is not a purchase. It is a specification, and it becomes the input to two other things.

Write down, on one page:

  • Each tier: what it holds, roughly how much, what it is allowed to connect to
  • What each one is made of: device, seed, passphrase or not, derivation path if non-default
  • Where each backup lives, by location rather than by content
  • The drill date for each tier and when the next one is due
  • What triggers a review: a threshold crossed, a house move, a new person involved

This page is what F104-L1 asks you to build with real but small funds, and it is the technical half of the document F110-06 asks you to leave behind.

Store it where you would store the rest of your important paperwork, dated, containing no secrets.

Common misconception

The right answer is whatever the most security-conscious people use.

The people whose setups you read about online have a threat model you almost certainly do not share, and copying their answer without their inputs produces a worse outcome than choosing your own.

F106-01 made this point about threat models and it applies with more force here, because a custody setup has to be operated for decades. A configuration that a security professional maintains comfortably is a configuration you will stop maintaining in month four, and an abandoned setup fails in the direction of lost access, which is the more common failure.

There is a second problem with copying. The public advice in this field is disproportionately written by people who sell components of the answer, and F104-01's commercial disclosure is on this page for the same reason. Notice what a recommendation would cost the recommender if it were wrong.

The defensible method is the boring one. Start from your three inputs, choose the simplest configuration that covers them, drill it, and add only what you can justify in a sentence. That produces a setup nobody will write an article about, which will still be working in ten years.

Closing F104

You now have the architecture that the rest of the Freshman level assumes. F105 is what you sign with it. F106 is who is coming for it. F110 is what happens to it when you are not there.

F104-L1 builds the three tiers. F104-L2 is the wipe drill. Both are components of the Custody Practical, and both are done with real but small funds, because a setup that has never held anything has never been tested.

Key takeaway

Sort every guarantee into physics, mathematics or policy, and remember that policy can change by press release, which is what Ledger Recover taught a great many people about a property they thought they had bought. Choose from three inputs, how much, how often and who else is involved, rather than from a brand, and read down the table until a row describes you. Then stop, because every component you add removes one risk and adds a way to lose access, and the crossover arrives sooner than the marketing suggests: a simple setup you have drilled beats a sophisticated one you half-configured. Write the specification on one page with no secrets in it, and hand it to F104-L1 now and to F110-06 later.

These come back later

What is the question to ask of any custody product?
Which of its guarantees are enforced by physics or mathematics, and which by the vendor's policy. Policy can change by press release, and you will find out afterwards.
What are the three inputs to choosing a setup?
How much you hold, how often you transact, and who else is involved. Everything else, including the brand, is downstream of those three.
When does complexity start costing you?
At the point where the chance you cannot operate it exceeds the chance it is attacked. Every added component adds a way to lose access, and losing access is the more common failure.
What is the correct order of operations for a beginner?
Competence, then complexity. A simple setup you drill and maintain beats a sophisticated one you half-configured, because F103's survey says access loss beats theft.

Sources and review

Confidence medium·Volatility high·Reviewed 2026-08-05·Owner unassigned

Contested

Whether Ledger Recover materially changed the risk profile of devices is genuinely disputed. Ledger's position is that the capability requires explicit on-device consent and that the firmware was always capable of arbitrary updates. Critics argue the existence of an export path is itself the change. Present both positions; the lesson turns on what users believed, which is not in dispute.

Any specific product, model or service named in this area will age badly. The decision procedure is the durable content. Check every product reference at each review and prefer removing a name to updating it.

Track your progress

Create a free account to mark lessons complete and pick up where you left off.