Sky cut its buyback to an eighth. No supply chart moved

“Is this token deflationary” is a question with a shelf life, and the answers people rely on rarely carry a date. Sky changed the answer twice inside a year. The first change was silent. The second was voted on in public, executed on-chain, and is still absent from every chart that claims to measure a token’s supply.

The burns stop in January

Our ledger holds 7,529 burn events for SKY, running from September 2024 to the sixteenth of January 2026. After that date there are none. Across the trailing twelve months the window holds three burn rows and they move zero tokens between them — which is a different statement from “we could not measure it”, and is where this started.

The purchases carried on, into a contract with a name

The programme itself did not stop. Sky buys SKY with protocol revenue — the purchases arrive from a Uniswap pair, 0x2621…9e3c, so they are ordinary market buys — and they land at 0xBE8E…98FB, which the governance proposals name as the Sky Pause Proxy. Over the trailing year there are 7,535 such transfers, moving about 805 million tokens.

What happened in March

On 12 March an executive proposal went to a vote. It passed on the 13th and executed on the 14th, and among four unrelated actions it made two parameter changes to the buyback.

The reasoning sits in the Atlas edit for that week and the mandate in governance poll 1620. None of it was hidden, and all of it was published before it happened.

Atlas Edit Weekly Cycle Proposal - Week Of 2026-03-09On behalf of @atlas-axis, we are submitting the Atlas Edit Weekly Cycle proposal below. This proposal includes the following edits: Reduce SKY Buybacks - Reduces the allocation of protocol revenue to SKY buybacks from 300,000 USDS per day to 37,600 USDS per day. Remove JAAA From Sky Direct Exposures - Removes JAAA from the list of current Sky Direct Exposures. Add Sanctioned Locations To Full Block List - Adds Afghanistan, Belarus, Burma, Russia, and Venezuela to the Flagged IPs list for Ful...Sky Forum

Two readings of one number

This is the part worth dwelling on, because it is rare. A parameter set by a vote and a count of what arrived on the chain are entirely separate things, and here they can be laid side by side.

The spell’s parameters imply 37,601 USDS a day. Our count of the transfers arriving at the Pause Proxy through April gives $37,468 a day, across 6.267 purchases a day — the figure the new interval implies, to three decimal places.

Sky@SkyEcosystem

Sky Frontier Foundation has published its inaugural Monthly Financial & Operational Update. This new recurring briefing brings institutional-level transparency to Sky Ecosystem. Subscribe to receive the next updates directly in your inbox: https://t.co/cYrLaVynkD

An agreement like that is worth more than either source alone. The spell could have been executed and then overridden by another; our count could have missed transfers, and on one day in August it had. Two independent readings landing 0.35% apart is what it looks like when a programme does exactly what its parameters say.

What Sky says it is doing with the money

The spell says what changed; it does not say why. Sky’s own April update does, and in plainer terms than a parameter can: governance has moved, in its words, “from capital distribution to capital retention” — rerouting protocol surplus away from programmatic buybacks and staking rewards and toward a $150M reserves target, on the stated grounds that protecting USDS holders comes before near-term distributions to token holders.

Sky Ecosystem: April Financial & Operational Update | Sky Ecosystem InsightsWelcome to the second Sky Ecosystem Financial & Operational Update, published by the Sky Frontier Foundation (SFF).insights.skyeco.com

That is a coherent policy and this article takes no view on it. It is quoted because it answers the question a reader reaches next — the money did not stop, it went somewhere else — and because it is the project’s own account rather than an inference drawn from the chain.

What the fall looks like month by month

March is a half-month at each rate, which is exactly what a spell executed on the fourteenth should produce. The four months at the lower level afterwards are not a lull in a lumpy series; they are the new parameter, running.

And then they leave again

This is the part we got wrong first time, and it is the part that decides what the programme means. The tokens do not stay. Asked what the Pause Proxy holds today, the contract answers about 40 million SKY — against roughly 805 million that arrived from purchases in the same year. A treasury accumulating its buybacks would hold the sum of them.

Counted over the trailing year, 1.31 billion SKY arrived at that contract and 3.46 billion left it. The largest recurring destination, 0x6756…99ee, takes 1.21 billion across 56 transfers, the most recent on 17 August, and asking Ethereum what that contract is returns its name: VestedRewardsDistribution. The bought tokens are paid out to stakers.

These are counts rather than estimates, and the check that makes them so is one anybody can repeat. Every SKY transfer this contract has ever received or sent — 16,852 of them, back to September 2024 — adds up to exactly the balance it reports today, to the last of its eighteen decimals. A set of transfers that reproduces the balance is a set with nothing missing from it, which is the difference between counting a programme and sampling one. Read on 24 August 2026, at block 25,820,502.

Why no supply chart moved

Here is what makes a public, voted, on-chain change invisible anyway. A supply metric measures what was created and what was destroyed. This programme does neither: it moves tokens from a market to a staking contract and back into somebody’s hands. So it contributes nothing to that measurement — not a small number, not a falling number, nothing at all — and cutting it by seven eighths changes a supply chart by exactly zero, in both directions and at every scale.

The purchases themselves have a shape, and it is the shape of the vote. Here is what Sky bought each month, in dollars, from our own record of the transfers:

The cliff is March, and the four months after it sit flat at the new parameter. Now hold that beside the chart nobody draws: our record of every burn, every issuance and every unlock for SKY across the same thirteen months is zero in all three. A programme this size, moving this much money, with a change this large in the middle of it, contributes nothing at all to a measure of supply — and a token that contributes nothing does not show up as a flat line on those charts. It does not show up.

What we publish for SKY, which is nothing

Our index carries no net figure for this token, and the reason is the article in one line: nothing in the window moves the tradeable supply. Burns of zero tokens and purchases that destroy nothing leave no numerator, and a percentage assembled out of that would be an artefact of the arithmetic rather than a measurement of a programme.

The quantities are measured and they are the ones quoted above. Publishing them without a percentage is the honest shape of what we know: we can tell you what Sky bought and when, and we cannot tell you what its supply did as a result, because in this window its supply did not do anything this programme caused.

How to check this yourself

Open the Pause Proxy on any block explorer and list incoming transfers of the SKY contract. Count them by month and add the amounts. Divide April’s count by thirty and you get about 6.27 a day; divide 86,400 seconds by the hop the spell set and you get 6.267. There is no model in that, and no step that requires trusting us.

We ran it twice, through two providers that share nothing — an explorer’s API and a public node — for January and for April. The two readings agree on the count and on the total to the last base unit. Where our record and the chain disagreed while this was being written, on one day in August that held eight of its twenty-one transfers, the record was repaired to match the chain, and the day is named here rather than quietly corrected.

How to read any buyback programme

Three questions, in order, and the first settles most cases.

Where do the tokens end up, and do they stay there? An address with no key is a burn. A contract the protocol controls is not, and the second half is the half we skipped: compare what that contract holds today against what it received. If the balance is a fraction of the purchases, it is a conduit and the tokens went on somewhere.

At what rate, today? A programme is a rate rather than an announcement, and where the rate is a parameter it is public and exact. A programme running at an eighth of its former size is still, technically, an active programme.

What is happening on the other side? Purchases are one half of an arithmetic. Issuance, unlocks and emissions are the other, and a programme can buy steadily while the supply it buys from grows faster than it shrinks.

Every figure of ours in this article was measured on 23 August 2026, and every one can be recounted from a public chain. What we publish, what we withhold and why is written down in the methodology; the four checks that separate a real burn from a reported one are a piece of their own.