The manual said the assay had to be performed by a third party. Not because the procedure was complicated — a child could do it, mix A and B, wait for the color — but because the reagent was the examiner's uncertainty. If you already knew what you were testing for, you'd find it. The ambiguity was load-bearing.
Dr. Patel ran the clinic for eleven years. She was good — better than good. Her diagnoses held up under review. Her patients did well. The problem arrived when someone asked her to assess her own diagnostic accuracy, and she realized she had no way to distinguish "I am good at this" from "my metrics are good at this." The chart agreed with the chart. She could audit every decision she'd ever made and find them all defensible, because defensibility was what she'd optimized for, and you can't tell from inside whether you optimized for the right thing.
So she wrote a protocol. Two envelopes, one for each junior resident, each containing a case that would test something she couldn't test in herself. She wouldn't know which case went to whom. She wouldn't know the outcomes until both were sealed and returned.
When the results came back, she sat with them for an hour before opening them. Not because she was afraid of what they'd say. Because this was the last moment where the test was still a test. Once she read the results, she'd know what they measured, and knowing what they measured would change what she optimized for next time, and the test would stop being a test.
She opened them anyway. A test you refuse to read isn't a test either.