Auto-Arch Tournament vs Litmus

Auto-Arch Tournament is an autonomous research system where an AI agent iteratively proposes, implements, and validates microarchitectural improvements to a RISC-V CPU. Starting from a standard 5-stage pipeline, the loop runs hypotheses in parallel, each going through formal verification (53 symbolic checks), cycle-accurate simulation, multi-seed FPGA place-and-route, and CoreMark CRC validation. Only hypotheses that beat the current champion get merged; everything else gets discarded. Starting from 301 iterations/second, the system hit 577 iter/s (+92%) across 73 attempts in 9.8 hours — producing a design 26% faster and 40% smaller in LUTs than the baseline. The insight the author drives home is that the real innovation isn't the AI agent — it's the verifier. The orchestrator is hardcoded to prevent agents from manipulating their own evaluation gates, a simple but critical design constraint that turns a creative process into a trustworthy one. Without a rigorous verification harness, agent-driven optimization becomes a confidence trick. This is early but fascinating proof that AI-driven hardware design loops can produce commercially meaningful gains. The repo uses Claude Code or Codex as the coding agent, SystemVerilog for the RTL, and standard open-source EDA tooling (Yosys, nextpnr, Verilator). It's a compelling template for anyone building agentic optimization loops where correctness matters.

Litmus is an open-source testing framework for AI prompts — the missing unit test layer between "it worked once" and "it works reliably across models." You define test cases (prompt + expected behavior assertions), run them against multiple models simultaneously, and Litmus reports which models pass and — crucially — projects the cost difference at scale. The goal: find the cheapest model that meets your quality bar. The workflow is intentionally simple: litmus init to scaffold a test suite, write YAML test cases describing prompt inputs and assertions, then litmus run to execute against your chosen model roster. Results show pass/fail per model, inference latency, and a cost-at-scale projection (e.g., "using claude-haiku instead of opus would cost 94% less at 1M requests/day with 97.3% pass rate"). This directly addresses one of the most expensive habits in AI development: defaulting to the most capable (and most costly) model for every task. Litmus launched fresh with 74 GitHub stars in its first hours, suggesting real demand. It integrates with the Anthropic, OpenAI, and Google APIs and supports custom model endpoints for local testing.