# Agent brief: End-to-end Android checklist

- **Lesson id:** 07
- **Goal:** Check the complete Mac → Linux LiteRT conversion → Android APK path.
- **Prerequisites:** All earlier files and environments, [human lesson](07-end-to-end-checklist.html).
- **Inputs:** `clothing.pt`, `clothing.tflite`, Android project and device.
- **Outputs/artifacts:** Passing release-check checklist and one on-device label.

## Steps

1. Re-run Mac `uv` training with MPS/CPU and check `clothing.pt`.
2. Re-run Linux `uv` conversion with `litert-torch`. Check that `clothing.tflite` is not empty.
3. Check NCHW float32 input, logits output, and fixed labels at every step.
4. Build and install the Android app with the LiteRT dependency and APK asset.
5. Run a sample, show the softmax label and confidence, and log cold and warm latency.
6. Record any optional accelerator result and the CPU fallback.

## Constraints

No CUDA, secrets, emails, unrelated edits, or older Interpreter as the default. Link the [human lesson](07-end-to-end-checklist.html).

## Key concepts

- Full path: Mac `uv` training, Linux `litert-torch` conversion, Android `CompiledModel`.
- The contract is preprocessing, layout, dtype, logits, and the ten labels at every step.
- The APK must package `assets/clothing.tflite`. Buffers are allocated once.
- This lesson is acceptance. It is not a new code path.

## Takeaways

- You can rebuild the published Android path from the earlier lessons without guessing.
- Same training idea as the Core AI course. This one ends in `.tflite`, LiteRT, and Kotlin.
- A stable CPU label plus cold and warm logs is the release bar.

## Acceptance checks

Every release-check item in the lesson is checked, and an emulator or device shows a sensible, stable label.