31. Testing Strategy

Testing Strategy requirements from the authoritative Glitchpad Technical Specification.

Tier	Scope	Required tools and evidence
Pure unit	Detection weights, text profiles, Mermaid revisions and limits, metadata normalization, state transitions, path normalization	Rust and TypeScript unit tests
Property	Encoding/newline round trips, revision ordering, archive paths, bounded chunking	proptest and property-based TypeScript tests
Fuzz	Detection, image metadata, ICO, SVG, ZIP/XML, office relationships, save payload parsing	cargo-fuzz with regression corpus
Contract	Every document host adapter and renderer against shared behavior	Host and renderer conformance suites
Fixture/golden	Markdown HTML, standalone and embedded Mermaid, syntax modes, images, PDFs, DOCX, ODT, metadata, errors	Licensed corpus with expected semantic outputs
Shared UI	Tabs, commands, inspector, dirty/conflict/recovery, accessibility	Vitest, Testing Library, Playwright, axe-core
Native adapter	Desktop paths/watch/save and Android intents/providers/grants	Platform tests, JVM tests, instrumentation tests
Package smoke	Install, launch, association/open-with, core view/edit/save, metadata, recovery, uninstall	Clean OS images and physical Android devices
Manual	Visual polish, assistive technology, office semantic readability, signing/store presentation	Versioned checklist with operator and evidence

Fixture provenance is recorded in fixtures/provenance.toml with source, author, license, redistribution permission, format, feature purpose, digest, sensitivity review, and expected result. Production documents and private user data are prohibited as fixtures. Hostile fixtures remain inert outside the parser harness and carry clear warnings.

Size classes are tiny, normal, medium, boundary, over-limit, malformed, truncated, contradictory, encrypted, and adversarial. Golden updates require a reviewed explanation of why output changed. CI retries infrastructure setup, not failing assertions. A flaky test is quarantined only with an owner, issue, expiry, and replacement gate; release-critical tests cannot be quarantined.

Coverage is a diagnostic, not a success metric. Critical state, save, recovery, detection, permission, and parser-limit branches require direct tests regardless of aggregate percentage.