Relationship between observability traces and verifiable evidence bundles

[joy7758]

Observability traces show what happened in a system.

However, traces are not typically structured as verifiable evidence objects.

We built an experimental runtime evidence profile (AEP) that packages agent runs into integrity-verifiable bundles.

Pipeline:

runtime → evidence bundle → offline verification → tamper detection

Repo:
https://github.com/joy7758/agent-evidence

Artifact DOI:
https://doi.org/10.5281/zenodo.19055948

This raises a question:

Could there be a conceptual bridge between observability traces and verifiable runtime evidence artifacts?

In other words:

traces → debugging visibility
evidence bundles → verifiable execution artifacts

[joy7758]

Quick update: the minimal external demo is now merged into main and released.

This is relevant to the traces-vs-evidence-bundles boundary discussed here.
The public demo now makes a smallest reproducible contrast between:

• plain agent completion logs / traces
• bounded, exportable execution evidence with explicit intent, policy, verification, and receipt

Current public artifacts:

• repo: https://github.com/joy7758/verifiable-agent-demo
• release: https://github.com/joy7758/verifiable-agent-demo/releases/tag/minimal-demo-f669ba7
• demo page: https://github.com/joy7758/verifiable-agent-demo/blob/main/docs/outreach/minimal-killer-demo.md

The point is not that observability should be replaced.
The point is that observability alone does not yet give a bounded, portable receipt layer.