How to Predict Translator Performance

We in the translation business know that it's virtually impossible to accurately predict the performance of a translator on a given job. Even the best translators cannot be experts in all fields, and we all have bad days from time to time. Yet translation buyers must select translators all the time, and it's not always possible to test the translator with the job at hand or a portion thereof.

Translator buyers therefore use translators' credentials to predict their performance. These credentials may include:

1. a college degree in translation
2. a college degree in the subject-matter field
3. a professional certification
4. a sample translation (not necessarily in the field in question)
5. client references
6. professional experience in a non-linguistic field

Which of these (or other) credentials are the best predictors of a translator's performance? Which ones can be faked, lied about, or dismissed as irrelevant?