What is verifiable data?

Verifiable data is data you can prove things about, its exactness, its deletion, an aggregate over it, with a portable certificate anyone can check without seeing your data or running our software. AT-1 issues six such proofs, and verifying any of them is free and needs no account.

How can you prove data without showing it?

You issue a proof: a small certificate that commits to your data cryptographically. The recipient runs one free command that returns yes or no, reading none of your rows and running none of our proprietary software. Non-visibility is a mathematical property of the proof, not a policy promise.

What can you prove about your data?

Six things: proof-of-exactness (an archive decodes to exactly the data you committed to), proof-of-deletion, a provable clean room (exact aggregates where the operator cannot see raw rows), verifiable unlearning (an AI forgot a specific subject), verifiable analytics (a SUM/COUNT/AVG with an O(log N) proof), and proof-of-inference (a committed model produced a claimed output).

Is verifying a proof free?

Yes. Verifying is always free and needs no account. Issuing proofs is metered per artifact, and your first 1,000 proofs are free as a one-time credit.

How much do verifiable-data proofs cost?

Proofs are billed per artifact, not per byte, because a proof is a compliance deliverable. After the free first 1,000, monthly volume rates run from $0.010 down to $0.001 per proof. Your compression bill is unaffected, this is a separate, additive line.

Verifiable Data

Don’t just store your data. Prove it.

Verifiable Data issues portable proofsabout your data that anyone can check — without your data, and without our software.Turn “trust us” into “verify it yourself.”

Start proving See how verification works

6 proofs: exactness, deletion, clean-room, unlearning, analytics, inference
free: verifying is always free, no account needed
portable: a small certificate anyone can check in seconds
per-proof: billed per artifact, not per byte, additive line

Every data-trust claim today ends in “trust us.” Trust that the archive still decodes. Trust that the deleted record is really gone. Trust that the clean room didn’t peek. Trust that the model forgot.Verifiable Data replaces the promise with a proof — four of them, each a portable certificate a third party verifies in seconds. Issuing is metered; verifying is always free and needs no account.

1 · Proof-of-Exactness

Prove a compressed archive decompresses to exactly the data you committed to, checked by re-decoding a handful of random blocks against a signed root, never the whole dataset.

For: archival integrity, data escrow, “is this the file we agreed on?”

2 · Proof-of-Deletion

Keep a WORM-immutable, tamper-evident store and honour GDPR/CCPA erasure. The bytes never change; the ability to read a subject’s data is provably destroyed, and the destruction is itself an immutable, certified event.

For: right-to-erasure on immutable / audited storage.

3 · Provable Clean Room

Run exact aggregates (SUM, COUNT, averages) across parties where the operator provably cannot see any raw row, non-visibility is a mathematical property, not a policy promise.

For: cross-party analytics, data collaborations, privacy-first measurement.

Explore

4 · Verifiable Unlearning

Prove your AI forgot a specific subject, that the updated model is exactly what training-without-them yields, and that they influenced nothing else.

For: ML compliance, “prove the model no longer contains my data.”

5 · Verifiable Analytics

Get a SUM/COUNT/AVG over a sealed column plus an O(log N) proof a third party verifies in milliseconds, without re-scanning the data. Tamper the answer and the check fails, every time.

For: audited reporting, counterparty figures, “trust the total without the rows.”

6 · Proof-of-Inference

Confirm a committed model produced a claimed output, verified at ~5–31× fewer FLOPs than re-running it, with forged outputs caught with overwhelming odds. Honest: lightweight verification, not zero-knowledge / zkML.

For: AI output trust, model-serving audits, agent pipelines.

How verification works

at1 prove seal   data.at1  -o proof.at1proof    # issue a proof (metered)
# hand the recipient ONLY proof.at1proof -- never your data
at1 prove check  proof.at1proof                 # anyone verifies -- FREE, no account
-> VERIFIED: decompresses to exactly the committed data

at1 prove range  seal sales.csv amount          # NEW: O(log N) SUM/COUNT/AVG proof over a sealed column
at1 prove infer  verify root x y proof          # NEW: verify a model produced an output, cheaper than re-running

You issue a proof with one command and hand the recipient only the certificate. They run one free command and get a yes/no — reading none of your data and running none of our proprietary software. That’s the whole product: proofs that travel, checkable by anyone.

Pricing

Billed per proof issued— a proof is a compliance deliverable, so it’s priced per artifact, not per byte. Your first 1,000 proofs are free — a one-time credit— and verifying is always free. Volume rates below are per month. Your compression bill is unaffected; this is a separate, additive line.

Proofs issued	Price per proof
First 1,000 (one-time)	Free
Next 10,000	$0.010
Up to 100,000	$0.005
Up to 1,000,000	$0.002
1,000,000+	$0.001

Questions, answered

What is verifiable data?: Verifiable data is data you can prove things about, its exactness, its deletion, an aggregate over it, with a portable certificate anyone can check without seeing your data or running our software. AT-1 issues six such proofs, and verifying any of them is free and needs no account.
How can you prove data without showing it?: You issue a proof: a small certificate that commits to your data cryptographically. The recipient runs one free command that returns yes or no, reading none of your rows and running none of our proprietary software. Non-visibility is a mathematical property of the proof, not a policy promise.
What can you prove about your data?: Six things: proof-of-exactness (an archive decodes to exactly the data you committed to), proof-of-deletion, a provable clean room (exact aggregates where the operator cannot see raw rows), verifiable unlearning (an AI forgot a specific subject), verifiable analytics (a SUM/COUNT/AVG with an O(log N) proof), and proof-of-inference (a committed model produced a claimed output).
Is verifying a proof free?: Yes. Verifying is always free and needs no account. Issuing proofs is metered per artifact, and your first 1,000 proofs are free as a one-time credit.
How much do verifiable-data proofs cost?: Proofs are billed per artifact, not per byte, because a proof is a compliance deliverable. After the free first 1,000, monthly volume rates run from $0.010 down to $0.001 per proof. Your compression bill is unaffected, this is a separate, additive line.

Start proving Talk to us about volume

Receipts, prove a specific claim

All of these issue the same artefact: a small, portable, re-checkable receipt that someone who does not trust you, and does not have your data, can verify independently.

Answers with Receipts

Ask a question in plain English, get an answer, and a signed receipt of the exact SQL that produced it: tables, columns, row count, result hash.

Complete

Completeness proofs certify ABSENCE.

Custody Receipts

Every file that enters your pipeline becomes one hash-chained append in a per-site ledger, source, content hash, row count, timestamp.

DATABOM

Seal a signed derivation certificate, input hashes, program hash, output hash, environment fingerprint…

Docs & commands →

Document Extraction

Learn a template from a few sample statements or reports, then extract each cell bound to its page and byte offset in the source…

Proofs-as-a-Service

The hosted Verified-Data API.

Reconcile

Join a bank statement to a POS tape under a rule set and get a signed certificate: these are the matches, these are the residuals.

Selective Disclosure

A redacted export of a sealed archive that still verifies against the ORIGINAL seal.

Provenance, prove where it came from

Same primitive, pointed at origin rather than content: what produced this file, what is in it, how much of it is real, and what it is worth.

Appraise

An MDL valuation standard for data.

Corpus Governance

Assemble an AI training corpus that proves its own composition and lineage, % real vs synthetic, per-record provenance…

Dataset Registry

Publish a dataset and get a content-addressed, tamper-evident, versioned, citeable artifact.

Origin

At1 origin is a data-origin classifier: from a file's compression signature it estimates how it was produced…

Reality

The provenance & authenticity layer for the synthetic-data era.

AI and model governance

The EU AI Act asks what went into a model and what it decided. These answer both, with evidence that survives someone checking it later.

AI Evidence Capsule

A notarized, reproducible receipt for a single AI inference.

AI Evidence Vault

EU AI Act record-keeping in one file: hash-chain a model's checkpoints, its training data and its inference logs into a tamper-evident vault…

Artifact

Recover the invariants and distributions a compressor has to build, c = a + b, monotone counters, category mix…

Model integrity

Seal a model in AT-1 and every weight tensor gets a fingerprint.

Verified Model Registry

Store a whole family of models and checkpoints once: the base is kept a single time, every version as a verified delta…