Bourdon is designed to preserve recognition around the work rather than around a single vendor account. A same-machine field test on 2026-05-15 exercised that boundary.
The test changed the active vendor identity while keeping the local machine and project substrate unchanged. In the first new conversation, the agent recognized Bourdon and its purpose from bounded local context.
What the test supports
The observation is consistent with Bourdon’s architecture:
- Project recognition can be reconstructed from local, user-controlled state.
- The recognition path does not require access to another account’s remote conversation history.
- The agent could distinguish project recognition from unrelated application interface state.
This was a field observation, not an independent benchmark. It should not be read as evidence that every vendor, installation, or account transition behaves the same way.
What remained unresolved
- First-turn latency was not measured under controlled reasoning settings.
- Recognition was tested with a direct project question, not an unrelated prompt.
- The response did not provide perfect source attribution between Bourdon context and the agent’s other local context.
Those gaps are why Bourdon’s release materials describe the result as a field test. Repeatable latency, unprompted recognition, and explicit source attribution need separate measurements.
Why the boundary matters
An account-independent recognition layer gives users a practical portability property: project context can remain on infrastructure they control, in a documented format, while the active agent or vendor changes.
The test supports that narrow claim on one machine. Cross-machine equivalence and broader vendor coverage remain separate release criteria.