Keep validating, describing and judging the data as separate steps

Google's guide splits analysis into three stages and asks you to always know which one you are in:

1. Validation: is the data self-consistent, correctly collected and representative?
2. Description: what does the data objectively show? "Sign-ups rose and order rate fell."
3. Evaluation: is that good or bad for users and the business?

Mixing them causes most arguments. Two people can agree on the description, that conversion fell, and still disagree on the evaluation, because one cares about volume and the other about quality. Separating the steps lets you settle the facts first.