What does a pilot look like when starting with a healthcare referral workflow?

A model lab wants to test whether its assistant can decide when a specialist referral is ready to be booked. The healthcare administration team at a hospital outpatient service typically checks a referral letter, patient demographics, a current Medicare or insurance detail, relevant imaging or pathology, and a clinician's urgency category.

**Step 1: Sample.** The lab asks MirrorFirm for a small sample covering clean referrals, referrals with an out-of-date urgency category, referrals missing imaging, and referrals where the patient details conflict. A policy `POL-REF-03` states what each specialty needs.

**Step 2: Inspection.** A lab engineer reproduces the gold gaps with a script. A reviewer with outpatient administration experience decides twenty cases blind. They flag that referral `REF-0618` is too tidy, and the lab asks for messier letters.

**Step 3: Run.** Two model versions run on a held-out set. The lab reports false gaps, missed gaps and wrongly held clean referrals by defect type.

**Step 4: Decide.** The results show the weaker model misses conflicting demographics. The lab chooses to scope a larger multi-turn pack focused on conflicts and partial replies.

**What it teaches:** the pilot is a sequence of checks with a decision at the end. Terms, volumes and timing for the larger pack are agreed separately with MirrorFirm.