What have you actually learned about crypto risk and custody?

This course asked where the value actually goes when it goes. It closes the Crypto & DeFi domain — which began with blockchains as market infrastructure, moved through the assets and how they trade, then through DeFi's attempt to rebuild finance in code.

The four movements, in one breath

• Custody models. A chain stores spending conditions, not accounts, so ownership is the capability to sign. That leaves two arrangements: you hold the key, or you hold a claim against whoever does. Self-custody swaps counterparty risk for operational risk, and its failure directions — loss and theft — fight each other, so redundancy counts only when the copies are genuinely independent. A 2-of-3 quorum breaks that trade-off: any single key can be destroyed or stolen without loss. QuadrigaCX is the reminder that "we lost the keys" is a claim to verify. Institutional custody is a stack of controls resting on one legal sentence: are assets held for you, or owed to you? Insurance is a limit against named perils — $500m of cover on $8bn is 6.25%.
• When custodians fail. The freeze always precedes the filing: Mt. Gox by three weeks, Celsius by a month, FTX by three days. Mt. Gox failed because nobody reconciled the chain against the ledger — coins drained from 2011 while customers saw correct balances, with no segregation and no independent verification. FTX had better branding and the same substance: commingled customer assets, a risk engine exempting the one account large enough to matter, and collateral consisting of a token it had issued itself. Circular collateral evaporates exactly when it is needed, because asset and liability move on one variable. Proof of reserves answers half the question — liabilities are self-declared, the snapshot is a moment, control is not unencumbered ownership, and off-chain debts are invisible.
• Smart contract and protocol risk. An audit reviews a named commit against a named scope: check it against the deployed bytecode, read the resolution column not the badge, and note that economic design is often out of scope. Five families recur — reentrancy, access control, flash-loan price manipulation, accounting errors, key compromise. Immutability means bugs cannot be patched; upgradeability means someone can rewrite the protocol; no configuration removes both. Bridges concentrate this structurally: the pool grows with adoption while the verification mechanism stays a fixed set of keys, and in the largest incidents it was the quorum or the upgrade process that failed, not the chain. Governance is checkable in fifteen minutes — proxy admin, timelock delay, top-ten voting concentration — and Beanstalk was the limit case.
• Regulation, records and risk framing. Regulation attaches to activities and economic substance, sorted into pre-existing categories that differ by country — so the same token is classified differently in different places. Stablecoins attract payment-style rules; the travel rule asks that identifying information accompany transfers between regulated providers. In many jurisdictions the taxable event is a disposal, not a bank withdrawal — and platforms take your transaction history with them when they fail. On sizing: recovery arithmetic is asymmetric, drawdowns of 77–93% are in the documented record, total loss at the token level is ordinary, and the "uncorrelated" claim has repeatedly failed in stress — bitcoin fell roughly 50% in two days in March 2020.

The one sentence to keep

The chain secures the asset; everything that loses the asset lives in the arrangements around it — keys, counterparties, contracts, quorums, jurisdictions and records — so the risks worth analysing are the ones no block explorer will show you.

The non-negotiable framing

Everything here is education, not advice. No custody product, platform, exchange, protocol or allocation has been recommended, and none should be inferred from any example.

And two points need stating explicitly. This is not legal advice, and it is not tax advice. Classification, licensing, disclosure, reporting and tax treatment differ substantially between jurisdictions, differ by activity, and change frequently. Where the stakes are real, the answer comes from qualified professionals in your own jurisdiction.

Before you sit it

Each of these is a minute at your desk. Any one that is not names the lesson to reopen first.

• Say what "not your keys, not your coins" means in terms of what you hold a claim on — What does "not your keys, not your coins" actually mean?
• Say what proof of reserves does not prove — Why is proof of reserves weaker than it sounds?
• Say what a smart-contract audit proves and what it leaves open — What does a smart-contract audit actually prove?
• Say how you size a position that can go to zero — How do you size something that can go to zero?

Try it now

1. From memory, name the two custody arrangements available to anyone and the failure mode each concentrates — then the sentence that decides what a custodial balance is worth in an insolvency.
2. Write a one-page risk sheet for anything you hold: where the keys are and who else could sign, which platforms hold claims and on what terms, which contracts can be upgraded and by whom. Date it.
3. Take one published proof of reserves and answer the three questions — liabilities included, verified by whom, on what date — then verify your own inclusion in the tree if you can.
4. Measure the largest drawdown on the chart below and the recovery it required, then complete the sizing sentence with a number. Then take the checkpoint quiz.

A note on what we do here. EODHD Academy teaches how markets and market infrastructure work, using real market data as a laboratory. Nothing here is a recommendation to buy, sell, hold or custody anything, and nothing here is legal, tax or investment advice. The point is not to make you avoid the asset class — it is to make sure that when you take a risk in it, you know which risk you are taking.