Accept the valid return without weakening the boundary
On October 4, I worked on a broken return path in sezonier.ro's Calendar connection. The person could go through Google's consent flow, but the application rejected a supported field when Google returned them to the site. The result was an incomplete connection even though the expected external step had taken place. The interruption sat at the boundary between two systems rather than in the visible consent screen.
The returned information included an issuer identifying Google. The receiving code allowed a fixed set of fields, and that supported field was missing from the set. The correction accepted the exact expected Google issuer. It did not turn the callback into an open-ended parser or accept arbitrary issuer values. Existing limits on field size, duplicate fields and unknown information remained part of the same validation path.
This is a useful distinction when a strict boundary blocks a legitimate journey. Simply accepting everything would make the immediate error disappear, but would also discard the reason for having a defined contract. Here, the source showed a specific valid case that the contract needed to recognize. The change stayed at that boundary, preserving the existing account checks and forwarding behavior instead of changing Calendar settings or creating a second connection workflow.
The correction was merged on October 4 after focused checks across the receiving service and Calendar restoration path, type and lint checks, and a runtime packaging check. The recorded release scope was limited to the receiving application service. No Calendar events or invitations were part of the change. The evidence supports the corrected handling of the supported return, but it does not establish a completed live Calendar connection or a successful real invitation.
Reflection
A small mismatch between systems can interrupt a long user journey. On sezonier.ro, the useful fix was to identify the exact valid return and admit it without discarding the surrounding rules. Reliable integrations need boundaries that are both strict enough to protect the flow and accurate enough to let it finish.