What I'm envisioning is something along the lines of:

Some attribute
---
Version 1.0: Introduced
Version 1.1: Calculation of property now includes cached tokens where previously it didn't. Conforms to changes in OTEL spec as of v3.x.x

For example, the gen_ai.cost.input_tokens attribute  previously excluded cached input tokens and now includes them.

When the OTel spec was updated last month to reflect this, our conventions were changed to follow, but the Sentry SDK versions that are older would still be calculating this value with the previous definition, which can lead to confusion for users who visit the docs to better understand the metrics generated in Sentry.