# Agent brief: Meet the original pipeline again

- **Lesson id:** 07-back-to-the-pipeline
- **Goal:** Rebuild the clothing classifier from the bottom up, then send the reader back to the original course's first lesson with every arrow understood as a lower-layer requirement.
- **Prerequisites:** Complete lessons 01 through 06.
- **Inputs:** Metal execution, Core AI graph and runtime, `clothing.aimodel`, tensor contract, training choices.
- **Outputs/artifacts:** Three reversed flow diagrams, one forward pipeline diagram, explicit link to `../pytorch-core-ai-xcode/01-the-pipeline.html`.
- **Concrete steps:**
  1. Diagram hardware to graph to asset.
  2. Add runtime and `NDArray` host obligations.
  3. Add UI and label mapping.
  4. Turn arrows forward while naming each requirement for the layer below.
  5. End with the original course lesson 01 link and backward reading instruction.
- **Constraints:** Use HTML `.flow`, not Mermaid; no quizzes, CUDA, or em dashes; explicitly send the reader to the prior course first lesson.
- **Acceptance checks:** Reversed series and forward series exist, all lower-layer requirements are named, the original URL is correct, and the human page links this brief.
- **Human lesson:** [07-back-to-the-pipeline.html](07-back-to-the-pipeline.html)

## Key concepts

- The backward path is complete: silicon limits, graph, `.aimodel`, runtime, thin UI, conversion, training.
- Turn the arrows forward again: Fashion-MNIST and `ClothingCNN` to `.aimodel` to `AIModel` to photo to label.
- Each forward arrow is a requirement for the layer below.
- Profiling checks whether the chip received what the stack promised.

## Takeaways

- You can hold the reversed map and the original forward pipeline at the same time.
- Open the original course's first pipeline lesson with this backward map beside it.
- A useful label only happens when Metal, runtime, model file, tensor contract, and training all met their limits.