Update README and docs for HN launch

[waleedkadous]

Closes #556

Summary

• Hero blockquote: Replaced R1 comparison scores (92-95 vs 12-15 out of 100) with real-world production data (106 PRs in 14 days, 57min median feature) and R4 controlled comparison results
• VIBE vs SPIR comparison: Updated the expandable analysis from R1 data (where VIBE scored 0% functionality) to R4 data (where both approaches produced working code — methodologically stronger). Shows consistent +1.2 quality advantage across 4 rounds on a 1-10 scale scored by 3 independent AI agents
• Eating Our Own Dog Food: Added production metrics table from the Feb 2026 development analysis (106 PRs, 801 commits, 85% autonomous builders, 20 pre-merge bugs caught, $1.59 consultation cost/PR)
• VIBE description: Fixed inaccurate claim of "working code" — R1 VIBE produced boilerplate only with 0% functionality
• Table of contents: Fixed duplicate Quick Start entry, added Real-World Performance link

Linked docs reviewed (no changes needed)

All 5 linked docs were reviewed for currency:

• docs/faq.md — current, accurate comparison of Codev concepts vs Claude Code concepts
• docs/tips.md — current, all CLI examples and tips are accurate
• docs/why.md — historical article about Codev's origin; model references (Opus 4.1) are accurate for when the experiment was run
• codev/resources/cheatsheet.md — current, protocol/role/tool tables accurate
• codev/resources/commands/overview.md — current, all command summaries accurate

Key decisions

1. R4 over R1 for primary comparison: R1 was dramatic (VIBE scored 0-15 on everything) but R4 is more credible for an HN audience since both approaches produced working code. R4's +1.2 delta is more defensible than R1's catastrophic VIBE failure.

2. Production data in hero: The development analysis numbers (106 PRs, 57min median) are more impressive to a technical audience than controlled experiment scores. Led with these.

3. Kept linked docs unchanged: All linked docs are current. docs/why.md is a historical article — updating model names would misrepresent the original experiment.

Test plan

• Verify all internal links resolve (TOC anchors, doc links, resource file links)
• Verify external links to GitHub repos are accessible
• Confirm the expandable comparison section renders correctly on GitHub
• Review production metrics table against source data in codev/resources/development-analysis-2026-02-17.md

[waleedkadous]

Architect Review

Low risk — README only, +43/-35. Clean update:

• Results banner now leads with real production data (106 PRs/14 days) instead of R1-only scores
• R4 comparison table is much more credible than the old 0-vs-100 scoring
• Production metrics section with hard numbers is compelling for HN
• Good trim of filler content

Approved.

---

Architect review