---
name: llm-lora-writing-style lesson 08
description: >-
  Agent recipe for lesson 08: verify private writing, Mac training, merge,
  tokenizer, Core AI export, on-device chat, and writing-style eval as one
  on-device delivery.
---
# End-to-end checklist

**Lesson id:** `08`  
**One-line goal:** Verify the baked writing-style path from local writing to a measured and evaluated on-device chat.

## On-device job

The delivery is a baked model, tokenizer, rebuild notes, and a device measurement. Not a notebook. Not a runtime adapter.

## Prerequisites

Lessons 01 through 07 completed on one project.

## Inputs

Your writing, adapter, merged model, export, tokenizer, Xcode app, target device or intended simulator.

## Outputs and artifacts

A confirmed checklist note with checksums, review prompt, latency, platform, and `runtime_lora_swap: false`.

## Executable steps

1. Recreate or confirm the `uv` environment and `xcodebuild -version`.
2. Confirm local cleaned writing, secret scan, split, and held-out review prompts.
3. Confirm MPS or CPU training, manifest, and sample comparison.
4. Confirm merge-before-export, remaining adapter, tokenizer sidecar.
5. Confirm `coreai.llm.export` or `torch.export` plus `coreai-torch` produced the `.aimodel`.
6. Confirm the app prepared once and replied from the baked resource only.
7. Confirm lesson 07 traces, open codes, axial buckets, and binary voice checks exist.

## Constraints

No CUDA, quizzes, Mermaid, or em dashes. No hot-swap claim. Link the four prior courses and the diffusion sibling.

## Failure modes

Adapter treated as the product, missing secret scan or device time, runtime swap language, lost adapter or tokenizer.

## Key concepts

- The package is baked model, tokenizer, rebuild notes, device measurement, and lesson 07 eval.
- Data must be local, licensed, cleaned, and secret-scanned.
- Ship merge, export, checksums, Xcode 27 prepare-once, and the platform preset you tested.
- Keep the adapter and tokenizer for rebuild.

## Takeaways

- The adapter alone is not the product.
- Missing secret scan or device latency is an untested claim.
- A new engineer should rebuild without guessing export flags.

## Done when

Writing, adapter, merge, export, tokenizer, device measurement, and lesson 07 eval artifacts exist. A new engineer can rebuild from the notes.

**Human lesson:** [End-to-end checklist](08-checklist.html)