docs: narrow focus to ODS L1 + AI Disclosure per community review

[shenxianpeng]

Summary

Implements key recommendations from detailed community review. The core change: narrow ODS positioning from "AI era delivery governance standard" to "a lightweight, machine-readable standard for AI-aware pull request and delivery metadata."

Key Changes

1. Repositioning (README, docs/index)

Before	After
"Machine-readable delivery governance evidence for the AI era"	"A lightweight, machine-readable standard for AI-aware pull request and delivery metadata"
Full 9-module map as primary surface	ODS L1 + AI Disclosure as the wedge; 04-09 as experimental
"AI-native" design principle	"AI-aware, not AI-obsessed"
No disclaimer about scope	"ODS does not prove the code is correct. It proves the delivery process contains the minimum structured evidence needed for humans and machines to review the change responsibly."

2. AI Change Review (spec/04) — rewritten

Removed AI% as a compliance metric and replaced with qualitative fields:

• ai_assistance_level: none | assisted | generated | agentic
• ai_scope: docs | tests | implementation | refactor | security-sensitive | release
• risk_level: low | medium | high
• Review levels (L1/L2/L3) driven by assistance level, not line-count percentage
• ai_contribution_percentage retained as optional signal, not a hard gate
• Status: Candidate → Experimental

Includes explicit rationale: "Why Not AI Percentage?" section documenting 5 reasons it's unreliable.

3. New: PR Template (examples/ods-pr-template.md)

Copy-paste template for .github/PULL_REQUEST_TEMPLATE.md with:

• Summary, Type, AI Disclosure, Changes, Testing, Risk Assessment, Checklist
• Before/after comparison table
• CLI and validation commands

4. New: Before & After Case Study (docs/case-study.md)

Concrete comparison with reviewer experience metrics:

• Before: "fix stuff", empty body, 3-5 clarification questions, hours to review
• After: structured metadata, 0 clarification questions, minutes to review
• Honest "What ODS Proves / Doesn't Prove" table
• Surgery/aviation checklist analogy

5. Branch Naming (spec/01)

AI prefix (feature/ai-*) remains optional convention — JSON metadata is the primary attribution mechanism.

6. ROADMAP restructured

• Added Experimental status category
• Modules 04-09: Candidate → Experimental
• M2 renamed to "AI-Native Tooling" (tools built, modules stay experimental pending adoption signal)
• M3 renamed to "L1 Stable + Early Evidence"
• M4 renamed to "Governance Model"
• New M2 goal: "Adoption signal: 2+ teams using ODS L1 with positive feedback"

Design Philosophy

The wedge is: make AI-generated PRs easier to review, not to prove a full governance framework. Checklists don't perform surgery — they ensure the team has the information they need before they start.

Type

• Documentation