Products · Proof of Reserves

One book. Many regimes. One continuous proof.

Digital-asset custodians must attest solvency and institutional standing in each jurisdiction they operate. Snapshot Merkle sums are the wrong instrument. We replace them with a continuously maintained classified obligations state.

The problem

A custodian active in Switzerland, Liechtenstein, and Abu Dhabi runs three separate attestations over the same books under FINMA, MiCA, and ADGM FSRA. Costs grow with the number of regimes; overlap reconciliation grows with entity pairs that share pools, clients, or treasury.

We identify five gaps: (1) attestation cost multiplies with regimes; (2) nothing is attested between snapshots, so assets can be borrowed for the photograph; (3) valid snapshots prove a sum, not that assets available at each redemption horizon cover liabilities callable then; (4) the same artifact over-reveals to banking partners and under-reveals to supervisors; (5) when coverage fails, there is no shared breach definition, tolerance, or escalation path.

Worked instance · 100 ETH owed on demand

Assets on books: 10 unencumbered · 30 unbonding · 60 staked. Scalar PoR reports fully reserved. The ladder fails at the on-demand horizon.

h=365
Staked exits into cover
h=90
Unbonding clears
h=7
Partial liquidity
h=0
10 < 100: shortfall

What we propose

OPSIS (On-going Proof of Solvency and Institutional Standing) is a continuous attestation protocol that:

  1. maps books into a single classified obligations state;
  2. commits every transition, making inter-snapshot borrowing visible;
  3. defines mobility classes so assets cover callable liabilities accounting for haircuts and delays;
  4. lets each reader derive only their permitted predicate, preserving privacy;
  5. defines a breach taxonomy, tolerances, and escalation.

One classified obligations state, read many ways. The client checks inclusion. The banking partner checks coverage. The supervisor reads its slice. Sibling entities become derivations from the same proven state; that is where analytical models project attestation cost to invert.

What is new

Privacy for proof of liabilities was solved in 2015 and never deployed at regulated scale: no assurance criteria, no breach rules, no multi-jurisdiction economics, no continuity between snapshots. We contribute the coverage ladder, regime profiles, disclosure governance inside the attested object, an adversary model, cost models, and hash-based proofs from day one, because a proof archive, unlike a signature archive, would have to be rebuilt wholesale under a stronger assumption. Only the proofs are hash-based today. Reports are still signed and delivered with classical keys (Ed25519, X25519); post-quantum signatures are a target.

Stage

  • Built (prototype): custody proofs with breach history, client checks, per-reader openings, witness cosigning, an action gate, and a masked mode. All runs so far use synthetic books.
  • Measured (14 October 2026, one Apple M3 Max laptop, synthetic book): a masked proof of a 4,096-slot book takes 4.5 minutes to produce and 7.2 seconds to verify.
  • Measured (3 October 2026): an update proof costs less than a full reserve snapshot when churn is below about 5 to 12%. The custody proof does not yet update incrementally.
  • Regime profiles: the FINMA, MiCA and ADGM rule sets are illustrative drafts, not yet legally reviewed.
  • Open: independent security and hiding review, book completeness, one book for several regimes, and post-quantum signatures. Each report links to the previous one, and each report is verified on its own; folding a long history into one proof is target work.
  • Paper: accepted for the Crypto Valley Association (CVA) 2026 research journal.

Who this is for

Fit

  • Multi-regime custodians with overlapping entity perimeters
  • Books where staking and unbonding make scalar snapshots least honest
  • Assurance partners who need a criteria track, not a vendor opinion

Not a fit (yet)

  • Genuinely disjoint single-jurisdiction books
  • High-churn books above the continuous break-even intensity
  • Buyers expecting a supervisor to have already accepted ZK attestation as statutory discharge

What we will not claim

Coverage of liabilities by reserves rests on a custodian signature over attested balances, never proven assets. Completeness is oracle-relative. Events between cutoffs are not proved today. That a reader learns nothing beyond its permitted fields is not yet proven. Cross-operator overlap sits outside any single-operator proof. No supervisor has accepted zero-knowledge attestation for a reporting obligation. A certificate is evidence feeding the buyer’s obligations; it is not discharge of AML or safeguarding duty.

Paper and demo

The OPSIS paper (continuous attestation for multi-jurisdiction custodians) and the product demo are shared through personal, revocable invite links. Request access and we send yours.