On October 25, a Solana co-founder posted a paragraph proposing that validators price assets. No SIMD number. No specification. No reference implementation. Within a day the paragraph had been sorted into a category — "oracle disruption" — and the category did the rest of the work.
I want to slow that down. The interesting property of this proposal is not what it would build. It is what it cannot punish.
The mechanism, as stated, fits in three sentences. Validators vote on the price of tokenized collateral, weighted by stake. New SOL is minted to facilitate network transactions. A tax is levied on the resulting profit and booked as network revenue. Three sentences, four mechanisms, zero parameters. In a bull market that is sufficient to move a narrative. It is not sufficient to move a codebase.
Alpenglow landed four days earlier, pulling finality to roughly 150 milliseconds and simplifying the validator vote path. The proximity is not accidental. A validator set that can vote every 150ms can vote on prices at a frequency lending markets can consume. Feasibility, though, is not soundness. The vote path got cheaper. Nobody asked whether the vote should carry economic weight.
To see why that distinction matters, you have to look at what an oracle actually is inside a Solana lending market. It is not a data service. It is the collateral valuation function, and the liquidation engine is its derivative. Kamino, Drift, Save, Marginfi — every one of them computes health factor as a function of an external price feed, then triggers seizure when that function crosses a threshold. Change the feed and you have not changed a dashboard. You have changed who gets liquidated and when.
The current feed layer is Pyth, with RedStone and Switchboard occupying the pull-based and modular niches. Pyth's model is first-party: exchanges and market makers publish signed price updates with confidence intervals and exponential moving averages, and consumers pay for the update they need at the moment they need it. Chainlink sits adjacent with a node-operator model and staking-backed reputation. The design premise across all three is identical, and it is not decentralization aesthetics. It is trust-domain separation. The party that says what an asset is worth is deliberately not the party that writes the ledger.
That separation is what the proposal dissolves. Validators already secure the ledger; the argument runs that they can therefore secure the prices on the ledger. It sounds like an efficiency. It is an authority merge.
When I traced the leaked FTX interface repository in late 2022, the failure was not a missing control. It was a merged one. A single sign-off path could mutate user balances without tripping the audit trail, which meant the system had no internal contradiction to detect — no second opinion to consult, because the second opinion had been folded into the first. Complexity is not the enemy of security. Combined authority is. A machine that both records a state and appraises that state has no way to notice that it is wrong.
That is the architectural objection, and it is the weaker of the two. The stronger one is cryptoeconomic, and it is fatal to the design as described.
Consensus security on Solana rests on a specific property: the faults it punishes are objectively provable. Double-signing two blocks at the same slot is verifiable from signatures alone. A node that produces that evidence gets slashed, mechanically, by code. The slash is not a judgment call. It is a deterministic function of publicly checkable bytes. That is why the security budget has a number attached to it — the cost of acquiring a third of stake plus the cost of forfeiting it.
Now run the same machinery against a price. A validator that votes $180 for SOL when the market clears at $150 has not committed a provable fault. It has expressed an opinion. There is no on-chain ground truth to diff the vote against. The chain can verify that the validator signed. It cannot verify that the validator was right. Stake-weighted voting is only as secure as its slashing conditions, and there is no slashing condition for a false price.
The consequence is precise. The security budget does not transfer. Under this design, the cost of corrupting the oracle is not the cost of acquiring stake — it is the cost of coordinating voters, which is a bribe, not a purchase. And a bribe is cheap, because the honest voter's payoff under a stake-weighted scheme is exactly the reward for voting with the majority, with no penalty for voting with a well-funded minority.
This is the P + epsilon attack, described by Vitalik Buterin in 2015 against SchellingCoin-style designs, and it is the reason stake-weighted "vote on everything" oracles — Truthcoin, Augur's early constructions — never became primary price sources. The attack is trivial to state. An attacker who profits from a manipulated price offers each voter slightly more than the honest expected payoff. Voting honestly yields the honest reward. Voting dishonestly yields the honest reward plus epsilon. Rational voters take epsilon. The attacker captures the gap between the manipulated price and the true price, which in a lending market is the entire liquidatable collateral base. The bribe is a rounding error against the seizure.
Note the asymmetry that makes this different from every critique of proof-of-stake you have read. To attack consensus, an attacker must acquire stake and accept the slash. To attack this oracle, an attacker must acquire votes and accept nothing, because the slash has no object. The mechanism does not degrade the security model. It routes around it.
I built a version of this problem from the other direction in 2026, specifying a zero-knowledge proof of intent so that autonomous agents could prove a transaction originated from a certified model within a stated confidence interval. The lesson from that work is directly applicable: a proof constrains an assertion to something checkable. A stake-weighted vote on a price is not a proof. It is a poll with collateral attached to the participants rather than to the claim. When agents begin pricing collateral autonomously, they will need attestations they can verify, not committees they must trust. This proposal supplies the latter and calls it the former.
The remaining two mechanisms are underspecified in a way that is itself a finding. "Mint new SOL to facilitate network transactions" paired with "tax the profit" does not read like standard proof-of-stake inflation. It reads like a settlement layer for tokenized collateral: minted units backed by pledged assets, with the protocol taking a cut of the carry. If that is the design, the dilution panic is aimed at the wrong target, and the real questions are who holds the minting authority and what the redemption path looks like under stress. If it is not the design, then it is seigniorage with a public-good wrapper. The text supports both readings. A mechanism that can be read as either a synthetic-asset settlement layer or a dilution engine has not been specified — it has been advertised.
The tax deserves separate attention because it is the part with legal consequences. Solana's non-security posture rests substantially on SOL's use being consumptive: you burn it as fees, you bond it for consensus. Routing DeFi carry through the validator set and booking the proceeds as network revenue changes the character of the cash flow. Under a Howey analysis, "common enterprise" and "reliance on the efforts of others" are the two prongs that get stronger when a protocol captures third-party profit and distributes it to a bonded class. That is not a fatal objection at the concept stage. It is a fatal objection to shipping without counsel.
The competitive framing is where the proposal is weakest rhetorically and strongest politically. Pyth's publishers are exchanges and market makers — the same institutions that would run validators. The proposal does not decentralize the data source. It relocates the fee. I have made this argument before about liquidity fragmentation, and it generalizes cleanly: infrastructure "gaps" in this industry are usually fee-capture opportunities wearing a public-good costume. Solana already has sub-second pull oracles with signed confidence intervals. The marginal latency improvement from validator pricing is approximately zero. The marginal authority concentration is enormous. What is being sold as a performance upgrade is an accounting change.
Here is the contrarian read. The popular objection is that validators should not hold pricing power. That is the wrong axis, and it is the axis the proposal wants you to argue on, because it is a values argument that can be won with a governance vote. The structural objection is that the design is unfalsifiable in the way that matters, and therefore unshippable in the form described. Every version that could actually clear governance will be a degraded version: a fallback feed consulted only when Pyth is stale, or a divergence detector that halts liquidations when the validator vote and the first-party feed disagree by more than a band. Those versions are genuinely useful. They are also not what the post described, and the gap between the two is where the narrative lives.
The ecological risk is underweighted for the same reason. Pyth is not an external competitor. It is Solana's own middleware, and it is one of the ecosystem's few genuine exports. A vertical integration play aimed at your own most successful middleware is a governance problem before it is an engineering problem, and the governance problem is not solved by the founder's credibility. Solana's history is littered with founder-floated mechanisms that either died in discussion or arrived at a fraction of their original scope. Agenda setting is a legitimate tool. It is not a roadmap.
So: watch for the SIMD number. If none appears within a quarter, classify this as signal-testing and move on. If one appears, read the slashing section before the architecture section. If price votes are not slashable — and they cannot be, because there is no objective falsehood to punish — then the design has no security budget and will be reframed as advisory. The only survivable form is validator attestation as a divergence detector feeding a circuit breaker, not as a price source. Lines of code do not lie, but they obscure, and this one is obscuring a fee transfer behind a latency claim. The question worth carrying forward is narrower than the one being asked: if the ledger appraises the collateral it holds, what exactly is left for the oracle market to sell — and who audits the appraiser?