# Cloture > Python and Rust tools for early stopping in model evaluation and for auditing LLM judge panels. For a fixed budget of unweighted votes, create `stopper = CiteStopper(max_budget=N)`. Call `stopper.update(label_id)` with an integer answer label after each vote. A `deterministic_lock` means the remaining ballots cannot overturn the leading count. For the separate statistical rule, independent draws from a fixed distribution with a unique mode allow `delta` to bound false mode certification. To study shared errors across judges, pass an item-by-judge vote matrix and gold labels to `audit_panel`. ## Documentation - [README.md](README.md): Quickstart, 10-second Python/CLI/MCP examples, and verified head-to-head benchmark tables. - [llms-full.txt](llms-full.txt): Self-contained full technical reference, API signatures, CLI flags, MCP tools, and verified metrics. - [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md): Exact non-asymptotic martingale, Kraft-IUT, Fan-inequality, CRC, and monoid proofs. - [docs/REAL_WORLD_EVAL.md](docs/REAL_WORLD_EVAL.md): Dual real-world evaluation (strict 5-fold OOS CV) on 2,985 AllenAI RewardBench items (+3.65 pp OOS / +3.85 pp in-sample at -72.7% cost; 80.4% AgentRace savings) and 350 ScalerLab JudgeBench items across 5 labs incl. o1-preview & Claude-3.5-Sonnet (+18.33 pp OOS CV at -76.9% cost; Top-7 cross-family n_eff=2.49/7). - [docs/benchmarks.md](docs/benchmarks.md): Full 7-suite head-to-head benchmark report against confseq, ppi-python, expectation, online-cp, bayesbench, and NumPy/SciPy. - [docs/comparisons.md](docs/comparisons.md): Factual feature matrix and guidance on when to use cloture vs. alternatives. - [docs/integrations.md](docs/integrations.md): Copy-paste drop-in recipes for OpenAI/vLLM logprobs, Promptfoo/CSV/JSONL judge-panel audits, Claude Code/Cursor MCP, CI eval gates, and C/WASM FFI. - [SKILL.md](SKILL.md): Agent skill routing table and statistical integrity rules. - [server.json](server.json): Official MCP Registry manifest (`io.github.pgg898200-ship-it/cloture`). - [include/cloture.h](include/cloture.h): Complete C ABI specification. - [benchmarks/results.json](benchmarks/results.json), [benchmarks/rewardbench_receipts.json](benchmarks/rewardbench_receipts.json), & [benchmarks/judgebench_receipts.json](benchmarks/judgebench_receipts.json): Raw machine-verifiable JSON benchmark receipts. - Repository: https://github.com/pgg898200-ship-it/cloture | Docs: https://pgg898200-ship-it.github.io/cloture/ | Live WASM Playground: https://pgg898200-ship-it.github.io/cloture/playground.html | Google Colab: https://colab.research.google.com/github/pgg898200-ship-it/cloture/blob/main/examples/cloture_frontier_eval_colab.ipynb