CertifiedData

CertifiedData — Tamper-evident AI-agent transaction audit corpus: for every agent-initiated spend, four linked records — an authorization record (policy decision allow / block / needs_review, policy ID, evaluated rules, written before any payment rail is called), an execution event (timestamp, rail, counterparty, outcome, or a block record), an Ed25519-signed receipt binding amount, currency, agent ID, policy decision, artifact hash and timestamp, and a public verification record. Transaction lineage links quote → policy evaluation → authorization → execution → receipt by transaction ID, and …

CertifiedData offers (Alternative, Reference, Sentiment) — Structurally dense, currently modest: each transaction yields four linked records plus a signature, so row count understates information content. Est. low thousands to tens of thousands of attested transactions to date. Separately, a fully licensable reference asset exists independent of volume: the published certificate/dataset schema registry, signing-key infrastructure, and public transparency logs for certificates, datasets, decisions and keys.

Training and evaluating AI-agent spend-governance and policy engines (the allow/block/needs_review set with rule attribution is near-unique supervision signal); fraud and abuse models for agentic payments; RL environments where agent purchase decisions carry real monetary consequence and verifiable outcomes; audit/compliance model building for AI-initiated transactions; agent reliability and failure-mode taxonomies derived from block records; content- and artifact-provenance studies linking an AI output to the transaction it caused; benchmark suites for agent task completion with cryptographic proof

The data is with 1 years of history.

Coverage spans US; Application Software, Transaction & Payment Processing Services; alternative, reference, sentiment; equities.

The supplier's own access route is a conversation with the supplier. Request a sample through Brickroad to start.