Quality

Strategy

The architecture is arranged so most behaviour is testable without a window, GPU or audio device:

Engine is headless. Integration tests in crates/kerosene-engine/tests/ (playthrough, gameplay, acoustics, streaming, save, ui, platform, extensions) drive whole maps through ticks and assert on world state.
Plain-data layers. kerosene-entity, kerosene-console, kerosene-kv and kerosene-bsp have unit tests with no engine.
GPU-free draw tests. scene and the UI produce draw data that tests inspect directly; a software rasteriser covers editor views.
Offline platform. kerosene-platform has a stand-in backend, so store code is tested without Steam. The real Steam backend is tested separately.
Layering as a test. cargo xtask layers fails on a forbidden dependency.

Baseline at the phase-0 audit: 84 suites, 2,273 tests, 0 failures. Counts move; run the suite for current numbers. Detail: Testing.

Check	Command
Build	`cargo build --workspace`
Test	`cargo test --workspace --locked`
Format	`cargo fmt --all --check`
Lint	`cargo clippy --workspace --all-targets --locked -- -D warnings`
Layering	`cargo xtask layers`
Public docs	`RUSTDOCFLAGS=-D warnings cargo doc -p kerosene --no-deps --features tools --locked`
Steam backend	`cargo test -p kerosene-platform --features steam`
Publishable crate	`cargo xtask bundle`
Book links	`python3 scripts/check-links.py`
Book build	`mdbook build`

.github/workflows/ci.yml runs on pushes to MASTER and on pull requests:

Job	Checks
`check` (Linux)	fmt, clippy, layers, tests, Steam tests, docs
`platforms`	builds on Windows and macOS
`bundle`	builds, packages and plays the published crate
`book`	link check and mdBook build
deny	`cargo-deny` using `deny.toml` (licences, advisories)

mdbook.yml deploys the book to GitHub Pages; release.yml handles releases.

No test drives the real launch boot path end to end.
Rendering correctness is checked by capture/smoke tests rather than image comparison against references.
No networking tests, because there is no networking.