A colleague points out that your new labels are not hum...

When the premise of a technical argument changes—in this case, moving from discrete human grades to continuous dwell time—it is critical to acknowledge that the initial objection regarding scale is no longer valid. However, simply switching to regression does not automatically solve all problems. You must evaluate whether the loss function is still aligned with the desired metric. Even if the scale issue is resolved, the loss-metric mismatch may persist, and that must be addressed independently of the data type.