Observe a cloud cutover through the same customer workflow
Separate customers awaiting migration from observations taken after an actual cutover.
A customer experience needs observations
A cloud migration announcement cannot describe what a particular customer experienced. To observe a change responsibly, a team needs the same tenant and workflow on both sides of an actual cutover. Truthound can check a prepared observation table for duplicate identities and inconsistent dates. It cannot perform the migration or manufacture a customer success story.
Two scopes in the announcement
Salesforce and Google Cloud announced an expanded partnership on September 15, 2026. The release says Hyperforce on Google Cloud was handling production traffic and that selected US customer migrations would begin in quarter 4 of 2026. Those are different scopes. A platform-level production statement does not put every selected customer in an already-migrated cohort.
Use the actual cutover event
An authorized observation table can identify tenant, workflow, cutover time, sample time, migration state and observation state. Preserve the source of the cutover event. A sample collected before a planned migration should not be labeled after merely because its date falls within the announced quarter. The cohort boundary is the actual customer event.
Preserve the observation identity
For the workflow comparison sample prepared as a Polars DataFrame, Truthound Core Python SDK 3.3.4's MultiColumnUniqueValidator can check the owner-defined combination that identifies a sample. Date-order checks can separately inspect the recorded sequence. The key must retain repeated observations when they are intentional, rather than deleting them as duplicates. A valid key does not show that two workflows are functionally equivalent or that a customer's experience improved.
A small-extract pandas approach
pandas 3.1.1 offers a useful custom route for a small comparison extract through DataFrame.duplicated with explicitly chosen columns and keep behavior. Add the assertion and investigation report in project code. Keep that tested implementation if it preserves the observation design. Changing validators does not eliminate the need to define the tenant cohort and comparable workflow. Set duplicated(..., keep=False) when the report must include every member of a duplicate group. Whether a repeated sample or participation is legitimate remains a separate policy decision.
Test pending and completed cohorts
Inspect a case that is pending migration and another with a confirmed cutover. Confirm that the first remains outside the after-migration group and that the second retains both original samples. This is a design exercise with authorized or synthetic records, not a report of a customer's actual experience. Measure functional or user-facing outcomes separately when the required observations exist.
Conclude within the reviewed window
Make a conclusion only for the reviewed cohort and time window. A new workflow version or a changed cutover event may require another comparison. Keep failures, missing observations and changes to the sample plan visible. The result is a traceable way to learn from a migration, without treating the announcement itself as evidence of an experience.
Evidence
- Event checked
- Subject checked
- Modified
