In arxcafe, a “book” is not one giant page. It is a network (graph) of atomic concept pages.
Each concept is linkable, searchable, and reusable across multiple layers.
Core Claim
Machine learning is the art of minimizing the wrong objective in a controlled way.
Layer 0 gives you the language to explain why models work when they work, and why they break when they break.
What This Layer Enables
- Reason about generalization (why training success doesn’t guarantee production success)
- Explain overfitting without buzzwords
- Choose losses/metrics intentionally (success is defined, not discovered)
- Predict failures under distribution shift and data drift
What Breaks If Skipped
- “Model selection” becomes trial-and-error rather than reasoning
- Monitoring and retraining feel optional instead of inevitable
- Offline metrics get mistaken for business outcomes
arxcafe Implementation Plan (Summary)
- Layer (Book) → folder namespace (e.g., /ml-engineering/layer-0/)
- Part (Section) → section index page
- Concept (Topic) → atomic HTML page using a consistent template
- No monolithic book.html; the “book” is the table-of-contents plus interlinked pages