Continuous proof of storage

S3 has 11 nines of durability. Prova has continuous, verifiable durability. That's a stronger property.

The difference

Property	S3 / Backblaze / IPFS pin	Prova
Durability claim	Vendor-asserted	Cryptographically verified
Verification	None — you trust the vendor	Every 30 seconds, public proof
Loss detection	When you try to read	Within seconds, on-chain
Economic backing	Vendor SLA credits	Slashed prover stake
Audit trail	Internal logs	Public Ethereum L2

S3 says "we won't lose your file" and you take their word for it. Prova says "here is a fresh proof, posted to the public ledger 8 seconds ago, that the bytes you stored are still on the prover's disk."

How it works

Prova uses PDP (Provable Data Possession), specifically the variant pioneered by Filecoin and adapted for Base in ProofVerifier.sol.

Every 30 seconds:

1. Challenge. The on-chain contract emits a pseudo-random challenge — a small set of byte offsets within the piece.
2. Prove. The prover reads those bytes off disk, computes a Merkle proof against the piece-cid, and submits the proof to the contract.
3. Verify. The contract checks the proof matches the on-chain cid. Pass → release payment. Fail → don't release; if it fails too many times in a row, slash.

The key property: the prover cannot fake a proof without actually having the bytes. The Merkle structure is committed at deal creation, so any byte the prover claims to read has to hash up to the original cid.

What "every 30 seconds" buys you

• Loss detection in seconds. If a prover loses your file, the next missed proof tells you within 30 seconds. Compare to S3, where you might not know until you try to read.
• Non-repudiation. The prover cannot say "we had it, but our logs got corrupted." There is a public chain of signed proofs.
• Slashing economics. A prover who deletes your file to save disk loses more in slashed stake than they save. The math is in Earnings.
• Re-pinning. When a prover starts failing, the marketplace automatically re-lists the cid. Healthy provers pick it up. Your file is never one disk failure away from being lost.

Why 30 seconds and not, say, 1 second

Compute and gas cost. Each proof is a Base L2 transaction. At 30 second cadence:

• ~2,880 proofs per day per deal.
• ~$0.003 gas per proof on Base L2.
• ~$8.64/day in gas per deal — bounded by the deal's payment, not the protocol's.

Faster cadence is technically possible but reduces the prover's net margin. We picked 30 seconds as a balance: fast enough to detect failures within a single human attention span, slow enough that the gas cost is a small fraction of the storage fee.

What this is not

• It's not Proof of Replication. We don't prove the prover sealed the data into a unique encoded form. PoRep is heavier and the security gain is marginal for non-Sybil-resistant networks.
• It's not zero-knowledge. The prover learns nothing about the challenge ahead of time, but the proof itself is public (and trivially small). If you need privacy on top, encrypt before uploading.
• It's not time-locked. The prover proves the file is currently there, not that it was continuously there since deal start. The two are equivalent in practice (you'd notice the gap), but a paranoid client should index the on-chain proofs themselves.

Audit trail

You can verify any deal's history:

# Get all proofs for a deal
curl https://prova.network/api/files/<cid>/proofs \
  -H "authorization: Bearer pk_live_..."

Returns a stream of { blockNumber, txHash, proofData, valid } records, one per attempted proof. You can independently re-verify any of them by calling ProofVerifier.verifyProof(...) on Base.