Problem: Spark's robustness expectations (long product names, missing images, large variant matrices, empty collections) live in design-QA checklists, not in anything runnable. Every QA pass improvises against whatever content the dev store happens to have.
Evidence: docs/section-specs/hero-image.md:123-134 (checklist states); no fixture set or seeded catalog anywhere in the repo; .preview/ is a single gitignored file.
Done when:
- A documented fixture set exists (product and content states covering long names, missing media, variant extremes, sold-out, empty collections) with a script or guide for seeding a dev store.
- The QA checklist references fixture states by name, so two people running QA test the same things.
Problem: Spark's robustness expectations (long product names, missing images, large variant matrices, empty collections) live in design-QA checklists, not in anything runnable. Every QA pass improvises against whatever content the dev store happens to have.
Evidence:
docs/section-specs/hero-image.md:123-134(checklist states); no fixture set or seeded catalog anywhere in the repo;.preview/is a single gitignored file.Done when: