How CertifiedData works
The verifiable AI workflow — from synthetic data generation through certified artifacts, independent verification, policy-governed agent transactions, and a tamper-evident decision log.
Every step is independently verifiable. Every artifact is cryptographically bound. Every decision is traceable to the certified data that informed it.
Generate
Synthetic data is the starting point. CertifiedData generates tabular synthetic datasets using CTGAN — a conditional generative adversarial network trained on your schema or an existing dataset. You can also upload a file directly.
Every generation is template-based, schema-first, or upload-assisted. The output is a structured dataset in CSV, JSON, or Parquet. No real personal data leaves your environment — the generator is designed for privacy-safe AI pipelines.
- →Choose a template (40+ industry schemas) or describe your schema
- →Generate in-browser — no account required for the first 10 datasets
- →Download the dataset immediately on completion
- →The dataset is automatically queued for certification
Certify
Certification transforms a dataset into a verifiable artifact. CertifiedData computes a SHA-256 fingerprint of the dataset bytes, then signs the fingerprint with an Ed25519 private key. The result is a structured certification artifact — not a PDF, not a badge.
The certificate contains the dataset hash, generation algorithm, row count, schema version, timestamp, issuer identity, and digital signature. Any modification to the dataset after certification produces a different hash, breaking the binding — making the record tamper-evident by construction.
- →SHA-256 fingerprint computed from the exact dataset bytes
- →Ed25519 signature applied using CertifiedData's private key
- →Certificate issued as a structured JSON artifact
- →Certificate ID assigned and stored in the artifact registry
Verify
Verification is independent of CertifiedData. Any party — an auditor, a regulator, a model buyer, or a counterparty — can verify a certificate using only the public key and the dataset. No account, no API call to the issuer, no trust in the platform required.
The verification process: re-hash the dataset, compare the hash to the certificate's stored fingerprint, then validate the Ed25519 signature using the public key at /.well-known/signing-keys.json. If both checks pass, the certificate is authentic and the dataset is unchanged since certification.
- →Hash the dataset with SHA-256 — compare against certificate hash
- →Validate the Ed25519 signature with the published public key
- →Check the certificate chain in the transparency log
- →Machine-verifiable: no human judgment required
Transact
Certified datasets can be transacted through Agent Commerce — CertifiedData's policy-governed payment layer for AI agents. Agents can purchase certified datasets, pay for API access, or settle inter-agent services, all with a verifiable receipt issued at the moment of settlement.
Every Agent Commerce transaction references the dataset certificate ID in the signed receipt. This creates a direct, auditable link between the payment and the certified artifact — so any party can verify not just that a transaction happened, but what was transacted and whether the underlying dataset is authentic.
- →Policy evaluation before any payment executes
- →Ed25519-signed receipt issued at settlement
- →Receipt references the dataset certificate ID
- →Full audit trail: authorization → execution → signed receipt
Log
Every AI system decision that touches a certified artifact is recorded in the Decision Ledger — an append-only, Ed25519-signed log where each entry is SHA-256 chain-linked to the prior record. The result is a complete, tamper-evident audit trail of AI behavior.
Each decision log entry includes the decision label, input context, outcome, model version, dataset certificate ID, chain hash, and signature. Any party can verify the entry and trace it backwards through the chain to the certified dataset that informed it.
- →Append-only log — entries cannot be modified or deleted
- →Ed25519-signed per entry, SHA-256 chain-linked across entries
- →dataset_certificate_id links every decision to its certified data
- →Independently verifiable: hash + signature, no platform access required
Why a verifiable workflow?
Trust embedded at creation
Certification happens at the moment of generation — not as an afterthought. The hash is computed from the exact bytes produced, so the certificate is always bound to the original artifact.
No trust in the issuer required
Independent verification means anyone can validate a certificate without contacting CertifiedData. The public key is published; the algorithm is standard; the verification is deterministic.
Accountability across the full lifecycle
The chain from certified dataset to agent transaction to logged decision is cryptographically linked. Accountability does not depend on policy — it is built into the artifact structure.
Compliance as a side effect
EU AI Act Article 12 logging, Article 10 training data documentation, and Article 19 technical record-keeping are not separate processes. A properly implemented verifiable workflow satisfies them as a consequence of normal operation.