Service reliability and change reliability are related but different. Service metrics tell you how a system behaves. Change reliability asks how a particular change performs after entering production.

Useful signals

Consider change scope, affected dependencies, testing and verification, production behavior, code volatility, incident history, deployment outcome, and rollback or remediation.

Why diff size is not enough

A large refactor in an isolated service may have limited production impact. A five line change in a shared request path may have a much larger potential impact.

A conceptual model

There is no single universally accepted formula for software change reliability. A useful model combines scope, dependencies, verification, production behavior, history, and deployment conditions.

What makes a measure useful?

A good reliability measure should be relevant to the change, traceable to evidence, actionable, comparable over time, and clear about uncertainty. A number can summarize evidence but should not replace it.
Tomosu's Production Reliability Index (PRI) is one approach to summarizing reliability signals around a change.
Explore PRI: https://tomosu.ai/start