Most people read a yield table from the top down. They see CAKE at 14.4% adjusted earnings yield. They see HYPE at 9.8%. They conclude the top of the list is where the value lives. They are reading the wrong column.
The number worth remembering is not 14.4%. It is 18%. That is the share of its own earnings CAKE actually routed to holders, and it is the single most quoted-yet-unquoted figure in the entire report.
Grayscale Research published two rankings, not one. The first measures token holder yield — what holders actually received as a percentage of circulating market capitalization. The second measures adjusted earnings yield — what the protocol itself earned. The lists overlap on nine of ten names. That overlap is the most information-dense figure in the report, and it went almost entirely uncovered. The gap between the columns is not statistical noise. It is value capture architecture, rendered as a spreadsheet.
The two-column structure is itself the finding. A single-column ranking of ten names tells you who paid. A dual ranking tells you the ratio between who earned and who paid, which is a fundamentally different question — and one that only becomes visible when the same asset pool is measured twice.
The ledger remembers what the bubble forgets.
What Grayscale actually measured
Token holder yield: HYPE, LIT, CAKE, AERO, UNI, RAY, PENDLE, AAVE, JUP, CRV. Adjusted earnings yield: CAKE, LIT, HYPE, RAY, AAVE, JUP, SKY, UNI, AERO, PENDLE. Nine names in common.
One methodological caveat first. The report carries a dateline that resolves to October 2026, which cannot be right. Either the dataset was mis-entered, or the price and emission environment it describes does not match the present. Every directional judgment below should therefore be read as a framework claim, not a mark-to-market claim. That is not pedantry. In a bear market, a ranking built on stale emission data is a map of a country that has since moved.
Note what the report does not measure: whether the payout is automatic, whether the parameters are timelocked, whether the treasury can redirect the buyback in a single vote. Those are the questions that determine whether a yield survives contact with a bad quarter. The rankings count the payout and skip the plumbing.
Now the set difference, where the real signal hides. If both tables draw from the same asset pool — and a nine-of-ten overlap makes that near certain — then SKY appears only in the earnings ranking, which means its token holder yield sits below CRV, the floor of the holder list. CRV appears only in the holder ranking, which means its adjusted earnings yield sits below JUP's 3.5%, the floor of the earnings list.
Read that twice. SKY, the former MakerDAO, rebuilt around a Smart Burn Engine and a migration incentive program, is now the textbook case of high revenue paired with the most aggressive dilution on the board. CRV is simultaneously worst-in-class on earnings and worst-in-class on holder return. Two failure modes, both invisible in a single-column chart.
The ratio nobody computed
I ran the lists against each other. Token holder yield divided by adjusted earnings yield gives value capture efficiency — for every unit of economic profit a protocol generates, how much reaches the holder.
HYPE: 8.0% over 9.8%, so 82%. LIT: 7.9% over 14.1%, 56%. CAKE: 2.6% over 14.4%, 18%. JUP: negative over 3.5%, below zero.
CAKE ranks first on earnings and captures 18% of them. Hyperliquid captures 82%. The number one name on the leaderboard is the least honest name on it.
Where does the missing 82% go? Treasury, liquidity incentives, market-maker subsidies, gauge bribes. Every line item defensible in isolation. None of them are the holder's pocket. The protocol earned 14.4%, distributed 2.6%, and the press release described the first number.
Mechanically this is a design choice, not an accident. HYPE routes protocol fees into an on-chain Assistance Fund buyback — automatic, parameter-light, difficult for a team to intercept quietly. AERO distributes fees to veAERO lockers at contract level. CAKE runs periodic buyback-and-burn sized by governance, with a veCAKE gauge layer steering emissions. That difference, automatic execution versus governance-determined execution, is the largest single variable in the table, and the methodology does not grade it.
Sort by that trust grade and the order scrambles entirely. HYPE and AERO sit in the hard tier: the contract pays before the committee meets. CAKE, PENDLE, AAVE, and JUP sit in the soft tier, where the payout depends on a vote, a proposal, or a treasury decision. LIT and UNI sit in the unknown tier, which is the most expensive place to be holding a yield thesis.
In 2017 I built a Python script to reconcile Golem's claimed emission schedule against live pool balances. It surfaced a 15% discrepancy between what the distribution mechanics promised and what the chain delivered. The lesson stuck: the announcement and the transfer are different events, and only one of them settles on-chain. The CAKE line is that lesson at scale.
There is a softer version of the same risk. A buyback sized by governance is not a mechanism, it is a motion that passed. No timelock survives a sufficiently motivated forum thread, and no published yield table accounts for the vote that has not happened yet. Call it soft value capture. The parameter exists; the parameter can move.
LIT deserves separate suspicion. A protocol in its first year rarely produces a clean yield; it produces a blend of fee share and emission subsidy, and only the fee share is durable. Second place on a ranking is a claim about sustainability, and on a two-year-old protocol that claim is mostly untested.
The UNI anomaly
UNI sitting fifth on holder yield is, on its face, a category error. For most of its life UNI has been the canonical example of a governance token with no claim on cash flow — the fee switch discussed, deferred, discussed again. If UNI genuinely ranks fifth, the fee switch is almost certainly active, and that is a paradigm-level change in the asset's fundamentals rather than a cyclical one. If it is not active, then Grayscale's definition of holder yield is doing undisclosed work. Both possibilities matter enormously, and they point in opposite directions.
JUP is the mirror case. A buyback program, a positive 3.5% earnings yield, and a negative 2.6% holder yield. That combination means one thing: emissions exceeded repurchases. The buyback exists, runs, and is announced — and still loses to the printing press. Performative buyback. The public relations value of the announcement exceeds the economic value of the transfer.
Contrarian: the yield was never the point
A 9.8% to 14.4% earnings yield, against assets that routinely post 60% to 100% annualized volatility, with a risk-free rate near 4%, is a risk premium of five to ten points. That is not compensation. That is an insult with a coupon attached.
Liquidity is not depth, it is just delayed panic. These yields are quotes against a market that has not yet been forced to be deep. When I modeled a 30% ETH drawdown against Aave V2 in 2020, roughly 40% of user positions came out undercollateralized. Model the same stress against a yield table and the structure collapses into two columns again. Suppose the drawdown arrives, emission programs get cut to protect runway, and buyback budgets get re-allocated to liquidity defense. Every name in the soft tier loses its yield almost simultaneously. The hard tier keeps paying, at a lower base.
In a drawdown, the buyback producing HYPE's 82% capture keeps executing; the governance-sized buyback at CAKE gets re-proposed, and the treasury behind it gets redirected to ecosystem growth. Same table, different trust grade.
This is what the fragmentation narrative keeps dodging. Ten new venues slicing the same small base of users do not create depth — they create more places for the same exit to clog. A yield table with ten names and one liquidity base is not diversification. It is one position wearing ten tickers.
Takeaway
By the next cycle the question will not be which protocol pays the highest yield. It will be which protocols encoded the payout in contracts no governance vote can quietly revoke. Two years out, the honest measure of a DeFi protocol is not earnings yield. It is the ratio between what it earned and what it could not take back.
The ledger will keep the record either way. The only open question is whether the entry next to your ticker was written by a contract or by a committee.