Chapter 02 · the game
A stablecoin desk proves its reserves by pulling balances from the venues that hold them. Every balance arrives as a zkTLS attestation, and every one of them below is real: signed in your browser, real keccak digest, real signer recovery. All of them verify. Some of them are still lying to you, and the difference is a check you either wrote or did not.
Three venues report in.
The desk above is not hypothetical. Primus shipped Proof of Reserves twice this year with Brevis. In April 2026 for Perena's USD* on Solana, covering on-chain positions across six protocols, with Binance and Hyperliquid coverage planned for a later phase. In May 2026 for Unitas and USDu, with both legs live from day one and the exchange hedge leg carrying the larger share by design.
Be precise about which layer does what, because the stack has three of them and only the first is zkTLS. Primus sits at the data retrieval boundary and proves a balance came over a genuine authenticated TLS session with the real exchange, without exposing API keys. A TEE then keeps those figures confidential and carries them to the proving layer. Brevis's Pico zkVM does the aggregation and proves that the published total was computed correctly from the authenticated inputs.
So the game above models the first layer and the accounting around it. zkTLS answers did this balance really come from that exchange. It does not answer whether the balance is current, whether you already counted it, or whether the endpoint was the one you meant. Those are the three switches, and in a live reserve feed they are the difference between a dashboard and a claim.
One failure mode has no switch. If the same money is genuinely visible at two venues, every proof is fresh, unique and correctly endpointed, and the total is still wrong. That is an accounting problem, not a cryptographic one, and it is why the real stack proves the aggregation as well as the inputs.