A blockchain stamp can prove a record hasn't changed since it was saved, but can it prove when things happened?

This explores whether a blockchain anchor can separate 'when something happened' from 'when it was written down and committed'. The corpus says it can't: the anchor only speaks to the second. What can a blockchain anchor actually prove about records? is blunt about it. An anchor shows a record hasn't changed since it was committed. It cannot verify when the event actually occurred, whether it was captured correctly, who was authorized to anchor it, or what caused it. Tamper evidence is not truth.

Think of the anchor as a sealed, dated envelope. It proves the contents existed by the sealing date and haven't been touched since. Anything the contents say about an earlier moment, such as 'the agent approved this at 14:02', is still only a claim, now frozen in place. A false timestamp can be made just as permanent as a true one. That's why Can external anchoring detect tampering in agentic process logs? treats anchoring as a layer on top of ordinary logging, not a replacement. Organizations still need conventional logs to reconstruct what agents did and in what order, and anchoring adds detection of after-the-fact edits. Can commitments protect sensitive agent data while enabling verification? adds a wrinkle. Only a commitment goes on-chain, so the organization must keep the underlying content. The story of when things happened therefore lives in the retained records, not on the chain.

The same gap appears when anchors are pitched for oversight. Does anchored evidence actually enable regulatory compliance or just readiness? argues that temporal anchoring and artifact integrity can't substitute for ordering, capture authenticity, and causal traceability. Those are the controls a regulator would need to verify that human oversight actually occurred. Ordering is a property of sequences, which is why Can stateless checks ever catch sequence-level constraint violations? is relevant. Checks that look at one action at a time can't express constraints that depend on history, so you need something stateful. A pile of individually anchored fingerprints doesn't supply that by itself.

There's also a quieter weak point. Who decides which agent communications get anchored? points out that the architecture anchors 'selected' communications without saying who selects them or how you'd notice a critical trace was left out. Something that never gets anchored has no sealed time at all, and nothing signals the omission.

Elsewhere in the library, time is something engineers have to build in on purpose. Can retrieval systems ground answers in the right time? adds an explicit time term to retrieval scoring so answers come from the right version of a document. Can routing mask future experts to prevent knowledge leakage? goes further and bakes it into the model, masking experts trained on periods after the question's date. By contrast, an anchor gives you only the commit-time half. Knowing when the event itself happened has to come from other evidence.