Skip to content
PlutusShield
Live on Cardano Preview testnet

Risks & disclosures

What exists, what doesn't, and what you would be trusting. This page is meant to be blunt.

← Back to site
DocsRisks & disclosures

Project status#

Status
ItemStatus
AuditNot audited.
Cardano deploymentLive on the Preview testnet only (test ADA and mock tUSDCx, no real value). Not deployed to mainnet, and no mainnet deploy happens before an external audit
Midnight deploymentpolicy-cover live on Midnight Preprod (testnet), with the v2 committee registry as the current registry: Cardano Preview policies mirrored, cover proofs, and claims decided by 2-of-3 committee voteClaim with demo evidence. Not on mainnet
Buying coverOpen on the Preview testnet: any CIP-30 wallet can buy on /cover (test value only)
UnderwritingOpen on the Preview testnet: deposits and withdrawals on /pool (test value only)
Real fundsNone. The pools hold testnet value only (test ADA and mock tUSDCx)
TestsCardano: 131 Aiken checks and a 36-step emulator run passing. Midnight: 60 simulation tests passing, full ZK compile OK. Off-chain: 226 SDK, oracle and committee tests passing

Passing tests show the code does what the tests check, nothing more. They don't replace an external audit. Mainnet capital is planned only after an external audit of both the Cardano and Midnight contracts. The scope we would hand that auditor, with every invariant mapped to its tests, is on Security & audit scope.

Smart-contract risk#

  • The validator or circuits may contain bugs that lose or lock funds.
  • No admin key and no pause. All parameters are fixed when the script is deployed. That removes the risk of a malicious admin, but it also means a bug or a bad parameter can't be patched in place. The fix is a new deployment plus LPs moving their capital.
  • The toolchains are pinned (Aiken v1.1.24 with stdlib v4.0.0, compactc 0.31.1, compact-runtime 0.16.0). Compiler bugs are possible, and Midnight tooling is young.

Oracle trust#

Parametric settlement is only as good as its oracles. The validator authenticates feeds (an allowlisted token under a fixed policy) and requires a quorum. It does not check where the price came from.

  • Whoever controls the oracle minting policy and the feed tokens decides what data exists. If quorum feeds collude or are compromised, they can trigger payouts that drain the pool. If they go offline or refuse to publish, valid claims can't settle.
  • Feed operators compute the time-weighted price off-chain. Each feed attests its own window, and the validator doesn't require the windows to match.
  • There is no on-chain dispute or challenge period for oracle data.
  • The same feeds gate sales. A Buy needs a fresh healthy-peg reading from every allowlisted feed, so if any feed goes stale, offline, or reports a depeg, new cover can't be sold (existing policies are unaffected). A feed that lags a real depeg could still let someone buy, but the waiting period means the depeg must last past the policy's start and a full trigger window after it.
  • Each oracle reading is its own UTxO, so an older healthy reading stays usable for a Buy until it is past the max price age (2h by default), even after a newer reading reports a depeg. The reference client always uses the newest reading per feed; closing this on-chain needs a state-thread oracle with one live UTxO per feed.
  • The oracle relay is live on Preview (median of venues, integer TWAP with a data-gap guard, depeg-window search, every datum re-checked against the on-chain rules). A fail-closed publisher reads CoinGecko, the Minswap pools on-chain and Kraken, and refreshes the Preview peg every 20 minutes, publishing nothing when fewer than two venues answer. On Preview one operator signs all three feeds, so a quorum there proves little yet; no mainnet feed set or quorum has been chosen, and independent feed operators are still to come.

Cross-chain & roles#

  • Cardano and Midnight are not trustlessly linked. The issuer mirrors policies into Midnight (registerPolicy) and expires them there (expirePolicy). It can't register a holder key or amount the buyer didn't commit to in the Cardano datum, but if the issuer is slow, offline, or wrong about expiry, the two chains can disagree. Lose your Midnight policy key and you can't prove or claim on Midnight; the Cardano claim token is unaffected.
  • The live Midnight registry (v2) decides assessed claims by a 2-of-3 assessor committee: three role commitments fixed at deployment, one vote per member per claim round, votes public and attributable. Assessed claims depend on that committee's honesty and availability — two colluding or compromised members can decide a claim either way, and if two are offline a claim can't be decided. Governance rotation is planned. (The v1 registry's single assessor key is kept as history.)
  • An approved Midnight claim doesn't move Cardano capital by itself. For exploit cover the payout path is live on Preview: after the Midnight decision, 2 of the 3 Cardano pool committee assessors co-sign a Settle that names the Midnight decision tx, and the validator pays exactly the coverage. Depeg policies still settle only on an oracle quorum, so an approved Midnight claim on a depeg policy isn't paid on Cardano. See Assessed claims.

Underwriter risk#

  • LP capital pays claims. A severe or correlated event can wipe out most of the pool. Live cover is backed 1:1 (capital ≥ active cover), but that backing is LP money.
  • There's no withdrawal delay, so better-informed LPs can withdraw free capital ahead of an unfolding event. Only the 90% utilization capital lock limits this.
  • Each currency tranche is its own risk pool. ADA LPs back only ADA policies, and USDC LPs back only USDC policies. With one product per deployment, each tranche's risk is concentrated in that product.
  • USDC tranche LPs and buyers also carry the risk of USDCx itself: Circle's USDC-backed native asset issued through xReserve. On Preview, the USDC tranche uses a mock tUSDCx that the deployer key can mint at will, so it has test value only.

Buyer risk#

  • Parametric cover pays on the trigger, not on your actual loss. You can lose money without a payout (for example, a dip to 0.96, or a depeg that lasted under 24h), and the reverse can also happen.
  • Claims must land before expiry + claim_grace_ms, and the trigger window must sit inside your cover period.
  • The user token is a bearer asset. Whoever holds it can claim. If you lose the wallet, you lose the claim right.
  • Cover doesn't start the moment you buy. It starts after the waiting period (24h by default, 60 min on Preview), and an event that begins before start isn't covered.
  • If the policy expires unclaimed, the ~2.5 ADA reference deposit is returned to the address you set as refund_to at purchase, even if you later transfer the user token.

Model & privacy limits#

  • Base rates, risk tiers, and the utilization curve are starting assumptions, not actuarial data. They may misprice risk in either direction. See Pricing.
  • LP projections in the pool simulator are illustrative, not forecasts.
  • Coverage amounts are public on Cardano today, even though they're committed on Midnight. See Privacy limits.