Crypto projects have spent about $638 million buying their own tokens back off the open market so far in 2026 — a record — according to Allium Labs data reported by the Financial Times on August 31 and by crypto.news. That compares with $545 million over the same stretch of 2025 and $366,000 in 2024. Two names account for nearly 90% of it: Hyperliquid, a perpetual futures exchange (a venue for leveraged bets on price), at roughly $370 million, and Pump.fun, a memecoin launchpad (a site that mints new tokens on demand), at nearly $200 million. Both tokens have outrun the market this year — HYPE up 145% and PUMP up 109% year-to-date against Bitcoin down 10% and total crypto market capitalisation down 11.9%.

Set the price comparison aside. The August 21 post argued that a price move is the lightest input a rules-based assessment reads; that holds in both directions, and 145% is not evidence a mechanism works. This is also not the August 27 argument about Ethena, which concerned revenue that had not started flowing. This money is real, booked and already spent.

The scoreable question is not whether a project buys back. It is what pays for it. A buyback is a use of revenue, so it inherits every property of the revenue — which is why it reads in On-chain Usage & Economics, 22 of 100 points, and in Ecosystem & Moat, 12 of 100. The first asks where the money comes from and whether it recurs. The second asks what stops it moving to a competitor. Under both, the concentration is the story: nine dollars in ten came from two businesses that sell the same thing, a toll on trading. Trading volume rises with risk appetite and falls with it. So the buyback is largest when the token needs it least and smallest when it needs it most. That is a property of the business, not of the token.

The three named programmes are three different objects to score.

Hyperliquid's is a protocol rule. Its documentation states that the assistance fund "converts trading fees to HYPE in a fully automated manner as part of the L1 execution," and that the HYPE it holds "is burned, removing the tokens permanently from the circulating and total supply." Roughly $1.3 billion has gone through it since the December 2024 launch; the company reported $169 million of second-quarter revenue on August 6 and put $141 million of it into buybacks.

Pump.fun's has an expiry date. On April 29, 2026 it burned about 36% of circulating supply — roughly $370 million of previously repurchased tokens — and locked 50% of net revenue into buy-and-burn through an irreversible smart contract, in its own words "for the next year." Two facts sit inside that. The prior policy was 100% of revenue; the other half now goes to product, hiring and acquisitions — the company electing to spend on the moat instead, after saying there was "a lack of trust, in the longevity of the business, the certainty of buybacks." And the one-year term ends in April 2027. About $200 million of this record is running on a clock. The August 26 post covered why schedules like this are governance parameters rather than facts; the point needs no re-litigating, only the date.

Lido's proposed mechanism is the one that writes the cycle down. As described in November 2025, it activates only when ether trades above $3,000 and annualised staking revenue clears $40 million, spends half the surplus above that baseline, and caps out at $50,000 a day and $10 million a year — about $4 million at then-current revenue. The coverage called this "an anti-cyclical approach, placing more bids in bull markets and fewer during downturns" — bids that scale with the bull market and pause in the downturn are the cycle, not a hedge against it. On September 1 ether traded around $2,430 to $2,470. The condition is not met.

None of this changes a grade; grades come from the scoring engine, not from a news cycle. And the plain risk is worth stating: a buyback creates a recurring buyer and nothing more. It cannot make revenue durable, and it has never stopped a token from failing.