We publish a headline rate of 10.5 percent against a staked book of about one thousand dollars. Anyone numerate who reads those two facts together arrives at the same question within about nine seconds, and it is a fair one: what happens to that number when real money shows up. This post answers it with arithmetic rather than reassurance, because the answer is already computed, already published, and already checkable without asking us anything.
The short version. At the escrow we hold today the full rate holds flat up to roughly fifty-four thousand dollars of staked kUSD. Past that the headline walks down, continuously, toward the carry the strategy actually earns. At one million dollars of staked kUSD it would print somewhere near 3.4 percent rather than 10.5. Holding the full rate at a million for the thirty-day minimum we solve against would take about 6,160 dollars in escrow, roughly eighteen and a half times the 333.43 that is sitting there now.
We would rather write that down ourselves than have it found. The companion pieces are the through-cycle feasibility band, which is where the sustainable long-run rate for this design is derived, and the honesty index, where our own advertised-versus-realized gap is listed first and, at the moment, worst.
The headline is a total, and only one half of it is funded
Start with what the number is, because the rest of the post is meaningless if this part is skipped. The 10.5 percent is not strategy yield. It is a sum of two things that behave completely differently, and /api/apy publishes them separately for exactly that reason.
The first component is organic carry, published as organicAPY, reading 3.00 percent at the block this post is stamped to. That is a live model of the deployed book: a liquid staking yield plus perpetual funding, multiplied by a ratio below one because the hedge is sized one for one against on-chain spot, then haircut for strategy costs and the insurance allocation. Nothing is escrowed against it. It moves every time funding moves and it can fall.
The second is the Genesis incentive, published as incentiveAPY, reading 7.50 percent. That one is paid out of real USDC sitting at an address we name, and it is the only half of the headline with money behind it.
One correction to a description we have seen repeated, including in our own earlier shorthand. The program does not declare a 7.5 point subsidy. It declares a 10.5 percent total, and the subsidy is whatever is left over after carry: the engine computes the top-up as the declared total minus organic carry, floored at zero and capped at 8.5 points. It reads 7.50 today only because carry reads 3.00 today. If carry improves the subsidy shrinks by the same amount and the escrow lasts longer. If carry falls the subsidy grows to cover it, until it hits that 8.5 point cap, at which point the headline prints below 10.5 rather than the escrow being drained faster. That cap binds whenever carry drops under about 2 percent, and a flag called capBinding says when it does.
The two calls that reproduce the whole thing
Both inputs to the arithmetic are on Base and both are readable without our permission. Every figure below was read at block 49362296.
The first is the escrow: USDC held at 0x52d3E450bA6c299B1B07298F1E87DD74732D4877, which is the 2-of-3 Safe that administers the protocol. The second is the eligible book: totalAssets() on the skUSD staking vault. Those are the two numbers the endpoint publishes as fundedUSD and eligibleTVL, and they reproduce exactly.
One trap worth naming before it wastes somebody's afternoon, because we hit it ourselves while checking this post. Calling decimals() on that vault returns 24, and 24 is the wrong divisor. It is the share decimal. totalAssets() is denominated in the underlying asset, which is kUSD at 18 decimals. Divide by 1e24 and you get 0.001011582 and conclude we are lying. Divide by 1e18 and you get our published figure to the last digit.
Two things the eligible book is not. It is not total deposits, and it is not the tvl field on /api/stats, which reads about 1,113 dollars and measures the PSM reserve, a different book. Only kUSD that is actually staked into skUSD earns the subsidy, so only staked kUSD belongs in the denominator. Counting unstaked kUSD would flatter the published runway.
The rule the rate is solved for
There is one invariant and everything else falls out of it. The subsidy rate we publish must be one the escrow can pay for at least thirty days at the staked balance published beside it. Thirty days is not a gate the program passes or fails. It is a constraint the rate itself is solved against:
At today's numbers, 333.43 divided by 1011.582169 times thirty over 365.25 gives 4.013031. That figure is published as sustainableAPY, and it is a raw decimal rather than a percentage, so it means 401 percent, not 4.01. At a book this small the escrow could sustain an absurd rate, which is why the declared rate rather than the escrow is what binds today and why taperBinding currently reads false.
Invert that same equation for the balance at which the two sides meet and you get the number this post is about, published as fullRateToTVL:
Note the declared rate in that line reads 0.0749621 rather than a clean 0.075. The published incentiveAPYPct of "7.5%" is rounded for display, and the engine solves on the unrounded residual. If you recompute the crossover from a flat 7.5 percent you get 54,126.80 and a gap of about twenty-seven dollars against our published figure. That gap is rounding, not an error, and we would rather explain it here than have somebody find it and reasonably assume the worse thing.
What deposits actually do
Below the crossover, a deposit does not move the rate at all. It shortens the date. The full rate holds flat, and what gets consumed is runway, which is why fundedThrough exists as a separate field from the rate. Above the crossover, the escrow rather than the declared rate sets the published number and the rate itself begins to fall, continuously, as a simple reciprocal of the book.
The table holds the escrow at 333.43 and organic carry at 3.00 percent, and varies only the staked balance. Carry is a live model and will not really sit still, so read the last column as the shape rather than as a forecast.
| Staked kUSD | What the escrow sustains | Published subsidy | Headline | taperBinding |
|---|---|---|---|---|
| 1,011.58 (today) | 401.3% | 7.50% | 10.50% | false |
| 10,000 | 40.60% | 7.50% | 10.50% | false |
| 54,154 (fullRateToTVL) | 7.50% | 7.50% | 10.50% | the crossover |
| 100,000 | 4.06% | 4.06% | 7.06% | true |
| 250,000 | 1.62% | 1.62% | 4.62% | true |
| 1,000,000 | 0.41% | 0.41% | 3.41% | true |
So the answer to the question in the title is 3.41 percent, and the honest way to say it is that at a million dollars of staked kUSD this escrow buys about four tenths of one point on top of whatever the strategy is earning. The subsidy does not disappear. It stops being the story.
The 6,160 dollar figure, and what it is not
Holding the full declared rate on a million dollars of staked kUSD for the thirty-day minimum would take 1,000,000 times 0.075 times thirty over 365.25, which is 6,160.16 dollars. Against the 333.43 actually escrowed that is a factor of about 18.5. Put the other way, the escrow funds the full declared rate on a million-dollar book for 1.6 days, not thirty.
It matters that this is a sizing requirement and not a spend, and we want to be precise because the sloppy version of this sentence would be self-flattering in one direction and alarming in the other. Nothing ever pays out 6,160 dollars. It cannot: the rate is capped at what the escrow sustains, so the actual thirty-day outlay at a million-dollar book is 333.43 dollars, which is the whole escrow, which is the point. The 6,160 is the answer to "what would this escrow have to be for the headline not to move," and the gap between it and 333.43 is the honest measure of how much of the current headline is a function of being small.
Why a taper instead of a cliff, and what the previous version did
This design replaced a worse one, and the worse one was ours, shipped on 2026-07-29. That build gated the entire incentive on a single test: if runway fell below the thirty-day minimum, publish no subsidy at all. Binary. At the sizing live at the time, the headline would have held 10.50 percent up to 1,264.79 dollars of staked kUSD and then dropped to organic carry on the very next deposit. Roughly 253 dollars of headroom, and the location of that edge was published nowhere. One depositor arriving would have destroyed the number that attracted them, in front of them, with no warning anywhere on the site.
We replaced it on 2026-07-30 with the rule above, and the replacement is why a number like fullRateToTVL can exist at all. A cliff is a number you hide because publishing it invites somebody to walk you off it. A continuous taper is a number you can publish, because crossing it costs a reader a fraction of a point rather than the whole claim. The regression that pins this is in the repository: with the old sizing, crossing 1,264.79 with a single three hundred dollar deposit moves the headline from 10.50 percent to above 8.6 percent, instead of collapsing it to 2.92.
There is a second decision inside this one that is worth stating because it went the other way from what was convenient. There is a field called tvlCapUSD and it is set to zero, deliberately. Capping the eligible book would have let us keep advertising a high rate by shrinking the denominator. But the subsidy reaches holders as a rising skUSD share price through a pro rata distribution across the entire vault, so there is no mechanism that could pay only the capped portion. A cap would have made the published rate larger than the rate an actual depositor received. It is not coming back.
Five places this is weaker than it sounds
A post that only explained why our design is good would not be worth publishing. These are the limits, and four of the five are things we went looking for specifically to put in this section.
It does eventually switch off. The taper is continuous, but there is a floor: below one hundredth of a percentage point the subsidy is not worth publishing and the block is withdrawn entirely, at which point the headline is organic carry. At today's escrow that happens around forty million dollars of staked kUSD. The step is 0.01 of a point, smaller than the rounding on the figure next to it, but it is a discontinuity and it exists.
Only the subsidy is escrowed. The carry is not. The 10.5 is roughly 3.0 points of unescrowed live model plus roughly 7.5 points of escrowed subsidy. The thirty-day guarantee attaches to the second number only. Organic carry can fall, and in a sufficiently bad funding regime it can go negative, and no escrow protects it.
The word escrow is doing some work. The 333.43 is a USDC balance at a 2-of-3 Safe, read live. It is not a time lock and there is no contract preventing those signers from moving it. The guarantee is arithmetic at read time, not a contractual commitment for thirty days, and anybody relying on it should know which of those two they are getting.
The published rate is rounded, and rounding can round up. The internal rate always satisfies the thirty-day rule exactly. The wire figure is rounded to four decimal places, so a reader who recomputes the obligation from the published rate can find the escrow slightly short of thirty days. At a million-dollar book that works out to about 29.7 days rather than 30, on a difference of about three dollars. We would rather state the size of that discrepancy than let somebody derive it and wonder what else is rounded.
The program has a hard end date that the taper never solves against. The incentive ends on 2027-07-24 regardless of how much escrow is left. There is a field called runwayDays that currently reads 1606, and it would be wrong to read that as the program lasting four years: it measures the escrow only and ignores the end date entirely. fundedThrough is the field that tells the truth, and it reads 2027-07-24, which is 358 days out. A consequence we have not fixed: inside the final thirty days the engine will publish a thirty-day funding promise the program itself will not outlive.
The number moves, so read it rather than quoting this post
The crossover is not a constant and it is not purely a function of the escrow. Because the declared subsidy is the residual after carry, fullRateToTVL moves whenever funding moves, with the escrow untouched. It read 53,599.89 on 2026-07-30 and 54,154.13 on 2026-07-31, a drift of about one percent in a day with nobody depositing anything. If carry fell to 2.5 percent the same escrow would put the crossover near 50,700, and under the 8.5 point cap it would sit near 47,800.
Which is the practical instruction. Every figure in this post is stamped to Base block 49362296, read at 16:05 UTC on 2026-07-31. Any of them may be stale by the time you read this. The live values are one request away at /api/apy, in the incentive block, which also carries the exact calls above and the block they were read at, so the endpoint documents its own verification rather than relying on this page.
Why publish this at all
The standard practice in this category is to publish a rate and to treat the conditions under which it stops holding as a private matter. We have argued for a while that the interesting question about any yield-bearing dollar is not what it advertises but what it would have to admit if asked precisely, and it is not much of an argument if we exempt ourselves. The point at which our headline stops being achievable is a field in our own API, and this post is the prose version of that field.
None of this is a claim that 10.5 percent is a rate a large book could earn here. It plainly is not, and the arithmetic above is the reason. It is escrowed for at least thirty days at the balance it was published against, at every escrow size and every book size, and that is a smaller and more specific promise than a headline rate usually implies. We think the smaller promise, stated exactly, is worth more than the larger one.
All figures read from https://kerne.fi/api/apy at 2026-07-31T16:05:39Z, Base block 49362296, and reproducible with the two calls above. Organic carry is a forward model of the deployed book, not a record of distributions; realized distributions are published separately at the honesty index and were unavailable at the time of writing. The escrow address, the eligible vault and the block are published live in the incentive block of /api/apy. Nothing here is investment advice or an offer of any token, no rate here is a promise, and past or modeled yield does not predict future yield. Our first external audit, by Hexens, delivered its initial report on July 20, 2026 with remediation underway, so we are not an audited protocol and do not describe ourselves as one. On-chain reserves can be checked without us at /verify and /api/por.
Verify it yourself
Run the same check on any reserve, or have it run for you.
Paste any issuer's signed attestation into the free verify tool and recover the signer, rehash the figures, and check freshness in your own browser. For a machine-signed, point-in-time read of an address you name, delivered on the page in about two minutes, the instant self-serve read is $29; a human-reviewed read is $149. A teardown like this one, commissioned on any target you name, is $499. An independent read of a counterparty you hold or allocate to is $2,500. Attestation tooling, not an audit, and not a solvency opinion.