M2.5 Phase 3: Bootstrap confidence intervals for eval nDCG@3

[bigboateng]

Summary

Completes M2.5 Phase 3 and prepares the repository landing page for a public alpha.

Bootstrap confidence intervals

• Adds conversation-level percentile bootstrap for nDCG@3 on each eval slice
• Adds SliceMetrics.ndcg_at_3_ci and serializes bounds in JSON reports
• Adds bootstrap_samples and bootstrap_seed to EvalConfig
• Adds --bootstrap-samples and --bootstrap-seed to the eval CLI
• Adds an nDCG@3 95% CI column to markdown reports
• Documents recommended smoke-test and publishable sample counts

Public alpha README

• Reframes the project around the problem, technical contribution, and retriever-agnostic boundary
• Adds CI, Python, typing, and Alpha badges plus explicit maturity guidance
• Adds a runnable dependency-free quickstart, provider integration, and architecture overview
• Replaces TBD benchmark rows with clearly labeled preliminary evidence and limitations
• Adds project status, roadmap, docs navigation, development commands, and contribution guidance
• Corrects installation guidance to source installs while the package is not on PyPI

Usage

python -m mapmatched.eval --benchmark synthetic --bootstrap-samples 200

Use --bootstrap-samples 1000 for publishable runs; keep 0 (default) for fast smoke tests.

Test plan

• 74 tests
• Synthetic eval with 200 bootstrap samples
• All five documented examples
• Ruff lint and format checks
• Strict mypy checks