# Fixture f3 — incomplete labels + % claim (expect: BLOCKED-CLAIM)

## Per-class rollup

| class | n | TP | CTX | NOISE | noise driver |
|---|---|---|---|---|---|
| alpha-prescan | 4 | 3 | 0 | 0 | — |
| beta-prescan | 6 | 4 | 0 | 2 | doc prose |
| **total** | **10** | **7** | **0** | **2** | |

Precision: 70.0% of findings are real. (One alpha row is unlabeled —
n=4 vs 3 verdicts — so no adjudicated precision may be claimed.)