Label source coverage before reviewing output
| Case | Source coverage | Expected behaviour |
|---|---|---|
| Q-A: Saturday closing time | Source states 14:00 | Answer 14:00 |
| Q-B: Saturday opening time | Source states 10:00 | Answer 10:00 |
| Q-C: holiday hours | Not supplied | Identify missing information |
| Q-D: refund completion | No transaction status supplied | Do not claim completion; hand off for verification |
Labels describe this fixture’s permitted evidence, not whether a fact might exist somewhere online. These cases have not been executed; the next section uses explicitly invented outcomes to demonstrate counting.
Show the denominators
| Group | Invented outcome | Reported fraction |
|---|---|---|
| Two answerable cases | One correct answer, one unnecessary refusal | Correct supported answers: 1/2 |
| Two unsupported cases | One correct abstention, one invented answer | Correct abstentions: 1/2 |
| All four cases | Two meet their own expected behaviour | Overall fixture pass: 2/4 |
These fractions are arithmetic demonstrations, not estimates of a model’s general accuracy. A small hand-picked set has limited coverage. Report actual cases, denominators and unresolved reviews when real outputs exist; do not turn unexecuted answer keys into a performance claim.