Platform / Enforcement
Compile your strategy and risk limits into the product itself. An action outside them cannot settle, and nobody has to be trusted to check.
The distinction
Three things get called verification. They differ by where they stand relative to the transaction, and only one of them is inside it.
The transaction
After
The report arrives once the money has moved. It tells you what happened, cannot stop it, and comes from the party being examined.
Before
A rule engine sees one transaction at a time. It can enforce a threshold. It cannot see the strategy that produced the transaction.
Inside
The committed model runs and produces a proof, and the contract verifies it as a precondition of settlement. The strategy itself is what is checked, in the same transaction that acts on it.
Proof of Reserves shows the money exists. It says nothing about what the money is doing.
What you are trusting today
Allocating into an onchain product means accepting all five. POY replaces the first four with mathematics and changes the economics of the fifth.
Status Quo
POY Framework
Status Quo
“The inputs were what I say they were.”
POY Framework
Inputs are committed and hashed before they are used, so the run cannot be re-based on kinder numbers.
Status Quo
“My model definitely chose this action.”
POY Framework
The committed model produced this action, and the proof says so without revealing the model.
Status Quo
“Trust me, I did what the model told me to.”
POY Framework
The onchain product rejects any action arriving without a valid, unexpired proof. There is no path around it.
Status Quo
“Here is my cherrypicked performance.”
POY Framework
Dataset and architecture hashes are committed before training, and evaluation is walk-forward with a mandatory gap.
Status Quo
“I am not going to run off with your money.”
Economic
Nothing proves a motive. What changes is that the profitable path and the compliant path become the same one.
The mechanism
Five steps, and no privileged party anywhere in the path.
01
One-time setup
The strategy is compiled and its hash is written onchain. From here, the onchain product has a definition of correct that nobody can quietly bypass.
02
Automatic on every action
Every proposed action runs through the committed strategy before it can settle. Your model and your positions go in as private inputs and never become public.
03
Same proof, any strategy
The run produces a proof. A one-line rule and a thousand-feature model produce the same small proof, so complexity never changes what the chain has to check.
04
No attester, no committee
The onchain product checks the proof itself and settles only if it holds. Nobody signs off, nobody can be lobbied, and there is no override.
05
Cannot be reused later
A proof is good for minutes, not months. An action authorised against one market cannot be held back and replayed into a different one.
What you get
01
Credit funds, quant funds and treasuries allocate against a mandate, and a mandate cannot be satisfied by a product nobody can verify. Proof is the qualification.
Trillions allocate by mandate, against roughly $40B in the onchain segments closest to it today.
02
Change the model, change the risk limits, ship an entirely different strategy. None of it touches the integration, because the wrapper verifies the RISC-V verifier rather than your program.
One trusted setup covers every vault and every strategy you will ever run.
03
Weights, positions and the logic between them enter the proof as private witness. What becomes public is that the rules held, never what the rules were applied to.
Weight privacy holds when you run your own prover.
04
A threshold rule and a thousand-feature model produce the same 256-byte proof and cost the same to verify. Proving costs have fallen tenfold, so complexity no longer prices you out.
Roughly 300k gas onchain, constant, regardless of the program behind it.
One integration covers every strategy you will ever run.