Android consumer of a generated Crossa Release AAR. The app compares Crossa, Retrofit+OkHttp, and Ktor against:

GET https://jsonplaceholder.typicode.com/posts

The default Crossa artifact is app/libs/crossa-generated-release.aar. Debug generation remains available for native troubleshooting and is not the benchmark artifact.

./scripts/generate-build.sh
./gradlew verifyDemo
./gradlew :app:assembleRelease

generate-build.sh runs crossa generate-build android and Gradle assembleRelease, then copies library-release.aar. Set CROSSA_CLI to the CLI built from the Crossa commit being validated; the script records CLI path, CLI SHA-256, Crossa source commit, runtime ABI, and AAR SHA-256 in app/libs/crossa-generated-release.manifest.json. Optional Debug Crossa artifacts:

CROSSA_AAR_VARIANT=debug ./scripts/generate-build.sh

The in-app harness is an observation tool, not a product performance claim.

• Warm runs create Crossa, Retrofit, and Ktor clients once and reuse them.
• Warmups are excluded from statistics.
• Measured rounds rotate client order.
• Timing uses SystemClock.elapsedRealtimeNanos().
• Failed samples are counted separately and are not stored as 0 ms successes.
• Crossa reports native-ready list availability and later field materialization separately.
• Remote JSONPlaceholder latency includes the network and is not SDK-only overhead.

Use a physical ARM64 device and the app Release build before treating numbers as production evidence.

The checked-in AAR is a generated consumer artifact, not a universal SDK. The current CLI and source commit must agree with its manifest before release.