init: Python Performance ADRs repo structure

Three-layer model:
- papers/ - Domain exploration (full context)
- RULEBOOK.md - Tight rules for agents (~200 tokens)
- evals/ - Machine-readable rules (future)

Source: python-numbers-everyone-should-know benchmarks

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

README.md

@@ -0,0 +1,21 @@
+# Python Performance ADRs
+
+Architecture Decision Records for Python performance patterns, distilled from benchmarks.
+
+## For Coding Agents
+
+Load `RULEBOOK.md` into context for coding tasks. It's tight (~200 tokens).
+
+For deeper understanding, load the relevant paper from `papers/`.
+
+## For Review Agents
+
+Use `evals/rules.yaml` for automated code review checks.
+
+## Source
+
+Distilled from [python-numbers-everyone-should-know](../references/python-numbers-everyone-should-know/) benchmarks.
+
+---
+
+See `PLAN.md` for the full methodology.