Staking gets described as earning interest on a holding, which is the one thing it is not. You are being paid to perform a task, you can be penalised for performing it badly, and the payment comes from three sources with nothing in common except that they arrive in the same wallet.
A distributed ledger has one hard requirement: everybody must agree on the order of transactions, without a central authority deciding it. Agreement is easy when everybody is honest. The design problem is making it expensive to lie, in a system where anybody can join anonymously and create as many identities as they like.
Both major mechanisms answer this the same way at the level of principle. They require participants to put something at risk that a dishonest participant would lose. The disagreement is entirely about what the something is, and everything else follows from that single choice.
Consensus is not a technical detail. It is the question of what an attacker has to destroy in order to lie, and the two answers give you two different systems.
Under proof of work, the thing at risk is spending: participants buy hardware and consume electricity to produce blocks, and an attacker must outspend the honest majority for as long as the attack lasts. The cost is external, continuous, and denominated in the physical world.
Under proof of stake, the thing at risk is capital: participants lock units of the asset itself, and misbehaviour destroys part of that stake through a penalty written into the protocol. The cost is internal, one-off, and denominated in the asset being secured.
The consequences are less symmetrical than the summary makes them sound. A proof of work network's security budget must be paid every day out of issuance and fees, which is why the long-term fee transition is a genuine open question there. A proof of stake network's budget is a threat of destruction rather than an ongoing expenditure, which makes it cheaper to run and introduces a different question about whether capital concentrates in the hands of those who already have the most of it. There is a second asymmetry that matters after an attack rather than during one. Recovering from a successful attack on a proof of work network requires out-spending the attacker again, indefinitely; recovering from one on a proof of stake network involves destroying the attacker's capital, which is a one-time event they cannot repeat without buying back in at a price their own attack has moved. Whether that difference is decisive is contested, and it is the strongest technical argument made for the newer design.
The job is unromantic. A validator runs software that stays online, receives proposed blocks, checks them against the rules, and signs an attestation saying it saw a valid block at a given time. Occasionally it is selected to propose a block itself. That is essentially the whole task, and it is why the hardware requirements are modest.
Rewards accrue for doing it correctly and promptly. Being offline does not destroy the stake, it simply stops the rewards and applies a small penalty for the missed work, roughly symmetrical with what would have been earned. This is the ordinary failure case and it is not dramatic.
The severe penalty exists for a narrow set of provably contradictory actions: signing two different blocks for the same slot, or making attestations that contradict each other. These are not mistakes a correctly configured validator makes by accident. They are the signature of either an attack or of running the same keys in two places at once, which is why the most common real-world cause of a slashing is a badly executed migration between machines.
The number quoted as a staking yield is the sum of three flows that behave nothing alike, and separating them explains most of the variation people find confusing.
The protocol creates new units and distributes them to validators. This is the base layer of the return, it is set by the protocol's rules rather than by market activity, and it is the component that scales inversely with how much total stake is competing for it.
Users pay extra to have their transactions included sooner, and that premium goes to whoever proposes the block. It rises with network congestion and falls when nobody is transacting, which means part of a staking return is a bet on activity rather than a fixed payment.
Whoever builds a block chooses the order of the transactions inside it, and that ordering has value to anybody trading. A market exists in which block builders pay proposers for the right to supply the block contents, and that payment flows through to validators. It is the most variable of the three, it is invisible in most yield summaries, and it depends on trading activity rather than on the protocol.
The issuance component is divided among all active validators, so the per-validator share falls as the total stake rises. This is not a fee change or a policy decision, it is arithmetic built into the schedule, and it means the return on staking declines mechanically as participation grows.
The practical implication is worth stating because it inverts an intuition. A rising staking rate is generally a sign that fewer people are staking or that network activity has increased, and a falling one often means the opposite. Reading the rate as an indicator of the asset's health gets the direction backwards about half the time.
It also means comparisons between periods are comparisons between different total stakes, different congestion levels and different trading activity. Two identical headline numbers a year apart can be composed almost entirely differently, and the composition is what determines how durable the number is.
A staking yield is three payments added together. Comparing two yields without knowing their composition compares two things that share only a number.
Locked stake cannot be traded, which is inconvenient for anybody who might want to. Liquid staking answers this by issuing a token representing the staked position, which trades freely while the underlying stays locked. It is genuinely useful and it introduces exposures the underlying position does not have.
The token's value depends on the market accepting it as equivalent to the stake it represents, and that acceptance is a market condition rather than a protocol guarantee. Under stress these tokens have traded below the value of what they represent, for the same reason any redemption-dependent instrument does: the path back to the underlying is not instant, so the discount reflects the wait and the doubt.
There is also concentration. A small number of providers hold a large share of total stake through these arrangements, which means a technical failure or a governance decision at one of them affects a meaningful fraction of the network. That is a systemic consideration rather than a personal one, and it is a genuine argument that the design has costs beyond the individual holder's risk.
Proof of stake removed the large ongoing issuance that paid miners, replacing it with a much smaller payment to validators. Separately, a portion of every transaction fee is destroyed rather than paid to anybody. The result is a supply that grows when activity is low and shrinks when activity is high.
This is a genuinely different property from a fixed schedule, and it is frequently described with more confidence than it deserves. Whether net supply increases or decreases in any given month depends on network usage, which is not predictable, and the two forces are close enough in magnitude that the sign flips regularly.
The honest framing is that the supply is activity-dependent rather than deflationary. Anybody presenting it as a one-way mechanism is describing the months in which it went one way.
Three practical consequences matter more than any of the theory above.
Entering and leaving the validator set both pass through queues designed to keep the set stable, and those queues lengthen exactly when many people want to move at once. Capital committed to staking should be treated as unavailable for a period you do not control, which is the opposite of how a trading balance needs to behave.
Blocks reach a state where reversing them would require destroying an enormous amount of staked capital, and that state arrives at a defined point rather than probabilistically. It is a stronger guarantee than accumulating confirmations, and it takes a defined number of minutes rather than being immediate.
A yield paid in the asset is exposure to the asset. Measured in your home currency, a return on a holding that fell by more than the yield is a loss, and the fact that the unit count went up does not change that. This is obvious when stated and is regularly forgotten when a rate is advertised as a percentage.
No. Interest is payment for lending, with the borrower's credit as the risk. Staking is payment for performing validation work, with penalties for doing it wrong and lock-up conditions attached. The return is also paid in the asset, so it is exposure to that asset rather than a return measured against it.
Two different mechanisms, with very different magnitudes. Being offline costs you the rewards you would have earned plus a small penalty, which is the ordinary case and is minor. Slashing destroys a substantial portion and applies only to provably contradictory actions, most often caused in practice by running the same validator keys on two machines at once.
Because it is three separate flows added together. New issuance is divided among all validators, so it falls as more people stake. Priority fees rise and fall with network congestion. Payments from block ordering depend on trading activity. Two identical rates a year apart can have almost entirely different compositions.
They are secured differently rather than one being weaker in a way that can be stated simply. Proof of work makes attacks expensive through continuous external spending; proof of stake makes them expensive by destroying the attacker's capital. The real debates are about capital concentration on one side and long-term fee funding on the other, and neither is settled.
It issues a tradeable token representing your staked position, so the capital is not locked from your point of view. The cost is that the token's value depends on the market treating it as equivalent to the stake, which is a market condition rather than a guarantee. Under stress these tokens have traded below what they represent.
Not reliably. Issuance to validators adds units, a portion of transaction fees is destroyed, and which one dominates depends on how busy the network is. Supply grows in quiet periods and shrinks in busy ones. Describing it as deflationary is describing the months where it went that way.
Not immediately, and not on a schedule you control. Both joining and leaving the validator set pass through queues designed to keep the set stable, and those queues get longer precisely when many participants want to exit at once. Staked capital should be planned as unavailable rather than as slow to access.
The two uses conflict directly. Trading capital needs to be available in seconds and staked capital is committed for a period you do not set. Anybody staking a balance they might need has accepted an exit queue as part of their risk management, usually without noticing that they did.