Small teams need a cheaper model-choice ritual

A small evaluation routine for teams that cannot run large formal evals.

For a small mobile team, a full eval framework can be too much. What has worked better for me is a weekly folder with five saved prompts: one bug report, one customer reply, one code change, one Vietnamese/English mixed note, and one messy screenshot description. Run both tools, mark the one you would actually ship, and keep the failures. After a month, the default choice is usually obvious.