Expected and baseline
PASS"line_items": [
{"description":"Masks","amount":"40.00"},
{"description":"Sanitizer","amount":"36.00"}
]
"total": "82.08"Zero-install demo
See the valid-but-wrong failure in one minute.
This guided view is derived from the same bundled invoice fixture used by the Rust product. It demonstrates the workflow but does not run your data or claim production evidence.
Workflow demonstration: 12 cases are not enough independent evidence for a production release decision.
Regression gate
NOT ENOUGH EVIDENCE
The candidate repaired two missing currency fields, but introduced three valid-but-wrong cases. The configured gate requires 100 independent cases.
Schema valid10/12 → 12/12+16.7 pp
Semantically correct9/12 → 9/120.0 pp
Valid but wrong1/12 → 3/12+16.7 pp
6Both pass
3Regressions
3Improvements
0Both fail
Paired transitions
Every discordant case stays inspectable.
| Case | Transition | What changed |
|---|---|---|
| invoice-002 | Improvement | Missing currency repaired |
| invoice-005 | Regression | Tax and total changed incorrectly |
| invoice-008 | Improvement | Vendor name repaired |
| invoice-009 | Regression | Total changed to 315.00 |
| invoice-010 | Improvement | Missing currency repaired |
| invoice-011 | Regression | Sanitizer line item removed |
Case invoice-011
Schema-valid. Semantically wrong.
Candidate
REGRESSION"line_items": [
{"description":"Masks","amount":"40.00"}
]
"total": "82.08"/line_items: required keyed item Sanitizer is missing.