State of the debate: is self-custody a solved problem or an unsolved one?

"Not your keys, not your coins" is the most repeated sentence in Bitcoin, and it states a fact: coins are unspent outputs on a public ledger, and what a wallet holds is the key authorising their transfer. Nobody disputes that. The dispute is about what follows from it.

The structural case for self-custody is strong and rests on a genuine difference in kind. A custodian's security depends on access control and on vetting people, so its customers inherit every failure of that vetting. Bitcoin's security rests on proof-of-work instead, which is why the network can be open, its traffic needs no encryption, and a transaction authorises a specific amount to a specific recipient without carrying a reusable secret the way a card number does. A user can verify all of this from the genesis block forward rather than being told it.

What that argument does not establish is that an ordinary person can execute it. The same chapter that makes the structural case closes by saying the ecosystem needs security tools and practices that are easier for nonexperts to use — which is a statement that the available ones are not. Its practical guidance is a list of open problems: physical storage, hardware wallets, balancing risk, diversifying risk, multisig governance, survivability. Every one of those is a decision the custodian used to make. This corpus cannot currently adjudicate the comparison: it holds no entries on custodial failure history, proof-of-reserves, or rehypothecation, so the case against institutional custody has to be argued from outside it.

Live questions

• Does adding keys reduce total loss risk or increase it? Theft risk falls with every added key; loss risk from procedural failure rises. (multisig-reduces-total-custody-loss-risk.)
• Can a nonexpert operate self-custody safely with the tools that exist today, as opposed to in principle? (self-custody-is-practical-for-nonexperts.)
• What happens to the coins when the holder dies or is incapacitated — and why is survivability consistently the last thing any custody scheme designs for?
• Is a seed phrase a backup or a bearer instrument, and does treating it as the former cause the losses attributed to the latter? (seed-phrase-mechanics.)
• Does a public ledger plus address reuse erode what self-custody was supposed to protect, given that custody secures the keys but not the privacy of what they do? (privacy-is-selective-disclosure-not-secrecy.)

Main positions (strongest forms)

1. Self-custody is the only arrangement that makes the asset what it claims to be. A bearer instrument you cannot bear is a claim on someone else, and history's supply of institutions that honoured such claims under stress is not encouraging. The trust model is better in kind, not degree: verification replaces vetting, and the root of trust is a chain the holder can check rather than a counterparty's word. Difficulty of use is an argument for better tools, not for surrendering the property.
2. The risk was transferred, not reduced, and the transfer is usually undersold. Pushing responsibility to the user does not make the risk smaller; it makes it theirs and uninsured. The failure modes simply changed shape — from counterparty default to a lost phrase, a house fire, an untested recovery path, an heir who never knew where to look. Elaborate defences are themselves a documented cause of loss: the source records owners who built complex encrypted backups against theft and thereby created a loss risk out of the complexity. Telling a nonexpert this is straightforward is not honest advice.
3. The binary is the mistake. "Self-custody vs. custodian" flattens a spectrum that includes collaborative multisig, time-locked recovery, inheritance protocols and delegated key management, each with different failure modes. The useful question is which failure a given holder is most likely to actually experience — and for many people that is neither an exchange collapse nor a stolen key, but forgetting, dying, or never setting it up at all.

Related corpus entries

The mechanical floor is keys-not-coins, seed-phrase-mechanics, hd-wallets, hardware-wallet-protections and multisig-basics — all VERIFIED, and all describing how custody works rather than whether it is advisable. The two live claims are multisig-reduces-total-custody-loss-risk and self-custody-is-practical-for-nonexperts, both CONTESTED and both single-sourced to owner research. privacy-is-selective-disclosure-not-secrecy is the adjacent question of what a key protects beyond the coins. Reference points: bip-0032, bip-0039, bip-0044, bip-0049, bip-0084 and bip-0086 (derivation and mnemonics — this brief is the first consumer of the otherwise-isolated wallets BIP cluster), bip-0174/bip-0370 (PSBT, the plumbing that makes multisig workable), and bip-0322 (signed-message proof of ownership). Also lost-coins-estimates, the corpus's only CONTESTED fact, which is the closest thing here to a measurement of custody failure. Absent and load-bearing: nothing on Mt. Gox, FTX, rehypothecation or proof-of-reserves.

Open questions a debate could resolve

• What the actual base rates are: how much bitcoin has been lost to self-custody error versus taken in custodial failure — and whether anyone can source either number rather than asserting it.
• Whether multisig is net-protective for a median holder, or a scheme that only pays off above a level of operational discipline most people do not have.
• What a genuinely good inheritance design looks like, and why almost no wallet ships one by default.
• Whether "not your keys, not your coins" should be taught as a property-rights fact (which it is) or as operational advice (which it is not), and how much of the disagreement in this topic is just that conflation.