Watch Safe

The Proof

Every watch tells a story. Now it has proof.

Paper certificates can be forged, lost, or quietly rewritten. A Watch Safe certificate cannot. This page explains the actual mechanism, down to the hash function, and runs the real sealing code live so you can watch it work.

Interactive

The sealing function, running live

This is the production sealing code, compiled into this page. The certificate id, the serial number, and a salt are joined into one string and hashed with Keccak-256. The engraving is drawn, curve by curve, from the resulting bytes. Edit the serial and the whole pipeline recomputes in front of you.

Certificate
WS-2026-00000
Hash match
Preimage
WS-2026-00000:SAMPLE0001:watch-safe-proof-page-sample-salt
Feed this exact string to any Keccak-256 implementation and you will get the fingerprint below. Nothing up our sleeve.
Keccak-256 fingerprint
0x6c85b63edca57e2de39905d270882aacd6643e6eb85968d7a8196645b8003dc4
keccak256(certificate id : serial : salt) · Salt: sample only; real salts never leave the server

The engraving is this fingerprint rendered as geometry: same 32 bytes in, same curves out, every time. No two records share one.

How is a certificate sealed?

When an independent authenticator certifies a watch, Watch Safe records the inspection and seals the record with a cryptographic fingerprint. The function is Keccak-256, the same hash the blockchain itself uses for transaction ids and addresses. It takes the certificate id, the serial number, and a per-certificate salt, joined into one string, and produces 32 bytes that stand for that exact record and no other.

The mathematics runs in one direction only. A record produces its fingerprint in a fraction of a second, but no algorithm runs the other way; the only route back is guessing inputs and checking them one at a time. Keccak-256 also has the avalanche property you saw in the demo: change one character of the input and about half of the 256 output bits flip. A near-miss forgery does not produce a near-miss fingerprint. It produces an obviously different one.

That fingerprint is then written to a public blockchain in a signed transaction. The ledger is append-only by construction: each block commits to the hash of the block before it, and thousands of independent nodes keep their own copies, so rewriting an old entry means rewriting every block after it on machines you do not control. Nobody amends the entry after the fact. Not Watch Safe, not the authenticator, not the owner.

What lives on the blockchain, and what never does?

The fingerprint proves the record, but the record itself is not published wholesale. The split is deliberate, and it is worth being precise about.

On the blockchain

  • The certificate's Keccak-256 fingerprint
  • A soulbound token, minted once
  • The block timestamp of the anchoring transaction

Never on the blockchain

  • The serial number
  • Photographs
  • Service records
  • The owner's name or contact details
  • Any price or valuation

Why the salt exists: serial numbers follow known formats, so publishing a bare hash of one would invite a dictionary attack, where someone hashes every plausible serial and compares the results. The salt is a long random value, unique to each certificate, mixed into the preimage and kept off the chain. With it in the mix, no amount of guessing recovers the serial from the fingerprint. That is why a certificate page is safe to link in a public listing.

Who knows what you own?

Collectors keep their collections quiet, for reasons every collector knows. A registry asks you to write the collection down, so it owes you a straight answer about who can ever see it.

Owner identity is never written next to the watch. It lives in its own records, envelope-encrypted with AES-256-GCM: each record is sealed under its own data key, and the data keys are themselves encrypted under a master key that never enters the database. A thief who copies the database gets ciphertext without the keys that open it. Every decryption is a named, logged event.

Lookups work through blind indexes: keyed HMAC digests that let the system ask whether an exact serial is on file without ever storing the serial in searchable plaintext. The index keys are separate secrets, also kept outside the database, so the index reveals nothing on its own.

The deeper protection is the design. The public registry answers two questions: is this watch on file, and is it flagged stolen. It answers both without naming anyone. Everything beyond that is released by the owner, one request at a time. Registering a watch never puts your name anywhere the public can see it.

What happens when someone checks a certificate?

Verification is not a stored answer, and not a cached one. Each time a certificate page is opened, Watch Safe recomputes the fingerprint from the record and reads the original back from the token contract on the public blockchain, live, on that view.

If the two disagree, the page refuses to exist. There is no warning banner and no asterisk; the address returns nothing at all.

The failure modes are deliberately uneven, and the code enforces the distinction. A provable contradiction between record and chain makes the certificate page disappear. A transport failure, an unreachable node or a timeout, does not: the page reports that verification is temporarily unavailable rather than pretending either way. Outages pass; the ledger does not change.

If the record and the chain disagree, the certificate page refuses to exist.

Under the hood

One certificate, one watch
Each certificate is issued once and named for its year and sequence: WS-2026-00042 reads as the forty-second certificate of 2026. The number maps to exactly one token on the chain, is never reissued, and never changes.
The token cannot be traded
On the blockchain, the certificate exists as a soulbound ERC-721 token: the standard interface for one-of-one tokens, with transfers disabled at the contract level. It cannot be sold, moved to another wallet, or separated from the watch's record. The contract also stores the original certificate fingerprint against the token id, which is exactly what verification reads back on every view. When the watch changes hands, the registry records the new owner for a flat $49 transfer fee; the token itself never moves. Watch Safe is a registry, not a marketplace.
A second seal on the paperwork
The certificate's documents live on IPFS, a content-addressed network where a file's address is derived from its bytes: hash the file, get the address. Change one byte and the address changes with it, so a swapped document cannot answer at the original address. The token's metadata points to that address from the chain, which closes the loop: the chain seals the metadata, the metadata seals the paperwork.
Yes, it is an NFT
The technology under the certificate is the one that got famous for cartoon apes and screenshots selling for millions. The mania obscured the invention. What an NFT actually does is bind a permanent, public, unforgeable record to exactly one thing. A cartoon never needed that. A $40,000 watch with four owners and a lost box does. Watch Safe uses the technology for what it was built for.
No secrets in the method
Everything on this page is a public, battle-tested standard: Keccak-256 for the seal, ERC-721 for the token, IPFS for the documents, AES-256-GCM for identity. That is on purpose. A system that is only safe while its method stays hidden is not safe. The security lives in the salts and the keys, which never leave the server side, not in keeping the recipe quiet. We would rather you check our work.

The record starts at authentication.

An independent authenticator examines the watch in person. Watch Safe records what they certify, publishes the registry entry, and anchors the certificate. The registry is opening to select authenticators.

Request accessSee how sellers use the certificate