# From real Sessions to controlled promotion

> A six-chapter guide from real failure evidence to one comparable Candidate regression and an external deployment handoff.

---

LLMS index: [llms.txt](/harbor-self-evolving/llms.txt)

---

This guide follows one honest evolution loop. It separates diagnosis from promotion and keeps three questions visible in every chapter: **what you do, what Harbor records, and what the result does not prove**.

## Contents {#contents}

- [1 Why self-editing is not progress](/harbor-self-evolving/book/01-why/)
- [2 Define the four concepts](/harbor-self-evolving/book/02-four-concepts/)
- [3 Diagnose recent Sessions](/harbor-self-evolving/book/03-diagnose/)
- [4 Turn badcases into a Dataset](/harbor-self-evolving/book/04-curate/)
- [5 Run one controlled regression](/harbor-self-evolving/book/05-regress/)
- [6 Read the Gate and hand off](/harbor-self-evolving/book/06-gate/)

---

Section pages:

- [Why self-editing is not progress](/harbor-self-evolving/book/01-why/): Improvement requires a fixed task, valid evidence, one controlled change and a policy outside the Optimizer.
- [Define the four concepts](/harbor-self-evolving/book/02-four-concepts/): Understand Dataset, Generator, Evaluator, Optimizer, train/validation/test splits and meta-evaluation in business language.
- [Diagnose recent Sessions](/harbor-self-evolving/book/03-diagnose/): Use a disclosed Historical Job to find recurring failures before building a promotion Dataset.
- [Turn badcases into a Dataset](/harbor-self-evolving/book/04-curate/): Convert recurring failure patterns into reviewable Tasks without leaking raw private history.
- [Run one controlled regression](/harbor-self-evolving/book/05-regress/): Freeze identities, establish a comparable baseline, change one surface and inspect Trial-level evidence.
- [Read the Gate and hand off](/harbor-self-evolving/book/06-gate/): Interpret PROMOTE or REJECT as a policy recommendation and hand deployment authority to external CI/CD.