This work explores a challenge I kept running into while studying how autonomous software agents interact with modern digital commerce systems. Although there has been steady progress in agent‑initiated payments and protocol design, I noticed that very little attention has been given to how these systems should record and verify the events that make up a transaction. In practice, this creates all kinds of problems — logs stored in different locations don’t line up, fraud labels can be added or changed without reliable evidence, and machine‑learning models often train on data whose origins are unclear. To address this, I designed a framework that creates a verifiable event timeline for agentic commerce. The idea is simple: every event generated by an agent or a participating system is serialized in a consistent way, batched, and committed using a Merkle‑tree structure. These commitments are then anchored on a blockchain to provide an independent, tamper‑evident reference. On top of this, I introduce a fraud‑marker mechanism that ties risk labels directly to the cryptographic evidence behind them. The final part of the work shows how these anchored events can be used to build reproducible and trustworthy datasets for training AI models, ensuring that labels and features actually originate from the events they claim to represent. The goal of the paper is not to replace existing agentic payment standards but to complement them with a transparency layer that helps developers, auditors, regulators, and researchers trust the information flowing through these systems. I also include a prototype implementation and discuss performance measurements to show that the approach is practical at scale. Overall, the work is meant to contribute toward safer and more accountable autonomous commerce ecosystems.