Our post “A signature proves who and what. Never when.“ ended on a sentence the report prints next to every attested date:
What the token does not prove: anything beyond RSA-4096, the scheme that signs it.
That is honest, and it is awkward. A tool that spends forty pages explaining that RSA expires cannot be content with dating its own conclusions with RSA. As a reminder: RSA is the most widespread public-key algorithm, and a large enough quantum computer would break it. The day that happens has a name, Q-Day, which our post-quantum Observatory (in French) explains step by step. We had split the trust between three authorities, in three jurisdictions. But splitting trust is not doing without it.
Then the question came in a more direct form: why would my signature be worth more than a public reference? Can’t we date against something everybody can check?
We can. But the question holds two questions, and they have to be pulled apart before answering.
Who, and when: two questions, two tools
A signature answers who. Ours is doubled, by the way, with a post-quantum signature, ML-DSA (in French). Our declaration says “this data does not need to stay confidential”, and the whole verdict of the report depends on it. What matters is that an identifiable person commits to it. A public reference cannot do that: anybody can anchor any file, which proves that it existed, not that somebody stands behind it.
A date answers when. And there the question is right: nothing requires it to be carried by somebody you have to believe. A timestamping authority signs a date; you can also write a digest into a public ledger, where nobody can move it any more.
So we keep the signature for who, and we change tools for when.
OpenTimestamps, in one sentence
OpenTimestamps, an open protocol published by Peter Todd in 2016, links a digest to a block of the Bitcoin chain through a sequence of hash operations that anybody can redo.
The mechanism, without the jargon:
- You send the digest of the findings, thirty-two bytes, to public servers called calendars. As with a timestamping authority, the document never leaves. The client even mixes in sixteen random bytes before sending, so that a calendar cannot tie a published report to whoever submitted it.
- Each calendar puts the digests it receives into a Merkle tree: digests are hashed two by two, then the results two by two, until a single value is left, the root.
- Every few hours, the calendar writes that root into one Bitcoin transaction. One transaction for thousands of digests: that is why the service is free.
- Once that transaction is in a block, the proof becomes complete. It is the list of hashes leading from your digest to the Merkle root of that block.
Checking it means redoing those hashes and comparing the result with the block header. There is no signature, no certificate and no authority in the loop. What you get has the same shape as in October: this digest existed no later than when that block was mined.
Why it is stronger than three authorities
There is no signature left to believe. An RFC 3161 token says “the authority asserts”. An OpenTimestamps proof says “redo the computation”. The difference is not about how much you trust the operator: there is no operator left whose word matters. A calendar that went down tomorrow, or wanted to lie, could take nothing away from a proof that is already complete.
Forging costs a block. To backdate a document, its digest would have to appear in an old block. That block is copied by tens of thousands of nodes, and the chain of work that follows it amounts to years of electricity. To show a fake header to a verifier that checks the proof of work, you would have to mine a block at the real difficulty. At today’s prices that is in the order of 2.5 to 3 GWh of electricity, that is €120,000 to €320,000 depending on the machines’ efficiency and the price of a kWh, hardware not included, to lie to a single person. (Computed from block 976,238’s difficulty: 5.8 × 10²³ hashes per block, machines at 15 to 20 J/TH, a kWh at 5 to 10 cents.) A verifier like ours accepts a difficulty up to four times lower than its latest checkpoint, so as not to refuse the honest blocks of an easier period: the floor price to fool it is therefore rather €30,000 to €80,000.
And above all, it is only hashing. This is the point we care about, since it is what this whole site is about. A token signed with RSA-4096 falls the day a quantum computer runs Shor’s algorithm at that size: tokens can then be made at any date. An OpenTimestamps proof is made of SHA-256 alone (and sometimes RIPEMD-160). Against a hash function, a quantum computer has Grover’s algorithm, which halves the security margin instead of erasing it. Finding another document that follows the same hash path would still take in the order of 2¹²⁸ operations, even with Grover. The proof we produce today has every reason to still be readable in 2040.
One precision, so as not to claim more than is true: the security of Bitcoin as a currency also rests on ECDSA, which is quantum vulnerable. But a proof of date does not depend on it. It needs the block header to exist and to have cost work, not anybody to still be able to spend their bitcoins.
“You just swapped an authority for a blockchain”
This is the serious objection, and it needs a precise answer. There is one place left where you trust somebody: whoever gives you the block header.
To check, you need the Merkle root of block number N. If you run your own Bitcoin node, you have it, and you believe nobody. Otherwise you ask somebody, usually a public explorer, and that is where trust comes back.
Here is what we did about it in Sablier, in this order:
- Your own node, if one is configured. No trust needed.
- Otherwise, two independent explorers (blockstream.info and mempool.space), which must return the same header. If they disagree, verification fails loudly.
- And in every case, the header itself is checked: its double SHA-256 must be below the difficulty target it declares, and that difficulty must be above a floor. An explorer that wanted to lie cannot invent a header: it would have to mine one.
That is the difference with an authority: we do not believe the explorer, we check what it gives. It can refuse to answer; it cannot lie cheaply.
Offline, in --airgap mode, the header can be supplied by hand. It then goes through the same checks.
What it does not prove
This is half of the subject, and we print it in the report rather than in a footnote.
- Who. An OpenTimestamps proof says nothing about the author. That is the signature’s job, and it stays so.
- A precise time. A Bitcoin block’s timestamp is set by the miner, within a margin: it must be later than the median of the previous eleven blocks and no more than two hours ahead (the exact rules). The date is therefore right to within a few hours, not to the second. To say “inventoried before 31 December 2026”, that is plenty; to separate two filings ten minutes apart, it is not.
- Right away. The proof is only complete once it is in a block, within two hours. Measured on this site’s inventory: submitted at 05:46 UTC, in the block of 06:11, complete proof available before 07:41, because the calendars wait for six confirmations before handing it out. In the meantime you have a promise from the calendars, not a proof.
- That the document did not exist before. Like an RFC 3161 token, it is an upper bound. Proving the opposite (“this report was not made in advance”) would require including an unpredictable public value of the day, such as a draw from the drand beacon. That is another tool, and we have not built it.
- Legal standing. A qualified electronic timestamp under eIDAS enjoys a presumption that its date is accurate (Regulation (EU) No 910/2014, Article 41). An OpenTimestamps proof has none: a court may admit it, nothing obliges it to.
Both together, and why
Hence our choice: we keep the RFC 3161 tokens, and we add the anchoring. They do not serve the same horizon.
| RFC 3161 token | OpenTimestamps proof | |
|---|---|---|
| what you must believe | the authority, and the RSA that signs it | nothing, if you run a node; otherwise a header whose work is checked |
| precision | the second | a few hours |
| available | immediately | after a block, within hours |
| legal recognition | a presumption, if the timestamp is eIDAS qualified | none |
| against a quantum computer | falls with RSA | rests on hashing alone |
The token is for a dispute in the next three years. The proof is for the question we will be asked in 2040, when nobody will be able to check a 2026 RSA signature without wondering who made it yesterday.
Doing it yourself, today, without us
As in October, none of this needs our tool. The official client is enough.
Replayed in an empty container on 17 November 2026:
$ pip install opentimestamps-client
$ ots stamp artefact.json
Submitting to remote calendar https://a.pool.opentimestamps.org
Submitting to remote calendar https://b.pool.opentimestamps.org
Submitting to remote calendar https://a.pool.eternitywall.com
Submitting to remote calendar https://ots.btc.catallaxy.com
$ ots upgrade artefact.json.ots
Calendar https://alice.btc.calendar.opentimestamps.org: Pending confirmation in Bitcoin blockchain
Failed! Timestamp not complete
Right after submission, nothing is in a block yet. A few hours later, ots upgrade completes the file. Then comes verification, and this is where the official client is most honest: without a Bitcoin node, it checks nothing by itself. On this site’s inventory proof:
$ ots verify -d 8ecb829ccc278ef7ab3138f56e07ee13319cbb72ce1dabbfc58ee7b4d6b670b5 report.html.ots
Could not connect to Bitcoin node: Cookie file unusable ([Errno 2] No such file or directory: '/root/.bitcoin/.cookie') …
$ ots --no-bitcoin verify -d 8ecb829ccc278ef7ab3138f56e07ee13319cbb72ce1dabbfc58ee7b4d6b670b5 report.html.ots
Not checking Bitcoin attestation; Bitcoin disabled
To verify manually, check that Bitcoin block 976238 has merkleroot 962c81d774938b388799d7307f6f46a54e0c51b2f47a89f46b45aa5ea1c6a0b9
We did that manual check against mempool.space and blockstream.info: both give 962c81d774938b38… for block 976,238. That is exactly where trust comes back, and what Sablier does for you, adding the proof-of-work check the client does not make without a node.
Keep the .ots next to the artefact, like the .tsr. Your reader needs neither you nor a home-made format.
In Sablier
All of this fits in one flag, next to --timestamp:
sablier scan . --sign=sablier.key \
--timestamp=http://time.certum.pl \
--anchor=opentimestamps
Unlike --timestamp, there is a default here: opentimestamps names the project’s public calendars. For authorities we refused any default, because choosing who attests your dates is a decision about trust. A calendar cannot lie: at worst it does not anchor, and the proof checks without it.
The rules are the same as in October:
- a proof that already anchors these findings is kept, because the oldest one is the only one that counts;
- an unreachable calendar is reported loudly, without failing the analysis;
--airgapmode refuses anchoring, since it goes over the network.
sablier verify fetches the complete proof when it is still pending, then checks it.
This site’s inventory has been anchored since 17 November 2026 at 06:11 UTC, in block 976,238. The proof is online next to the report, with its three tokens:
$ curl -sO https://mesa.black/audit/report.html.ots
$ sablier anchor report.html.ots --lang=en
report.html.ots anchors the digest 8ecb829ccc278ef7ab3138f56e07ee13319cbb72ce1dabbfc58ee7b4d6b670b5
anchored in the Bitcoin chain at block 976238, 17/11/2026 06:11 UTC — header from blockstream.info + mempool.space, merkle root and proof of work checked here.
And sablier anchor <file> anchors any file: a SBOM, a signed declaration, minutes. Every release of Sablier now publishes the anchoring of its own archive.
Good practice
- Separate who and when. A signature commits a person. A date does not need to commit anybody; it needs to be checkable.
- Anchor a digest, not a file. Thirty-two bytes leave, nothing else.
- Keep both proofs. The token for the coming years and for legal recognition, the anchoring for the horizon where RSA is worth nothing.
- Complete the proof, then keep it. A pending proof still depends on the calendars. A complete proof depends on nobody, but it does not complete itself.
- Check the header, do not believe it. Without a node, compare at least two sources and check the proof of work.
Watch out for
- An OpenTimestamps proof bounds a date from above, to within a few hours. Presenting it as an exact time is the kind of approximation that backfires in front of a careful reader.
- A public explorer is still a third party. Without a node and without checking the proof of work, you have simply moved the trust once more.
- A pending proof is not a proof. A report that said “anchored” before the block would be saying more than it knows.
- Anchoring carries no legal standing. If you need a presumption before a court, you need an eIDAS qualified timestamp as well.
- Anchoring says nothing about the author. A digest anchored without a signature proves that a file existed, not that anybody answers for it.
References
- OpenTimestamps: the project’s site, the official client, Peter Todd’s announcement in 2016.
- Timestamping by an authority: RFC 3161; eIDAS Regulation (EU) No 910/2014, Article 41.
- Bitcoin: block timestamp rules, proof of work.
- Cryptography: RSA, Shor’s algorithm, Grover’s algorithm, SHA-2, Merkle trees.
- Public randomness: drand.
- Our post-quantum Observatory (in French): Q-Day, post-quantum cryptography, “harvest now, decrypt later“, ML-DSA, RSA, hash function, timestamping.
- Here: “A signature proves who and what. Never when.“, Sablier, and this site’s inventory.