Empirical Evidence & Telemetry
Verdict Benchmark Hub
Verifiable receipts, reproducible test suites, and empirical benchmarks. No verdict without evidence.
Coding Agent Benchmarks
Matched-model and native product comparisons. Verdict Code vs Aider, Devin CLI, Codex CLI, and Claude Code across frozen regression repairs and boundary rules.
Project Benchmark (10-Story)
10 progressive real-world stories on project_commerce with 33 test assertions. Empirical comparisons across Claude Sonnet 5.5, Haiku 5.5, Tool Terse v5, Qwen, and GPT-OSS.
Blueprint Planning Benchmark
Evaluating deterministic PRD decomposition into atomic tasks and strict dependency graphs across 10 real-world specifications without LLM hallucination.
Hardware Concurrency: ON vs OFF
Empirical telemetry measuring multi-slot concurrency (C=1 to C=4) across 6 resident local models on Apple Silicon (M4 Max), eliminating queue latency.
Context Scaling: 66k vs 262k
Stress testing multi-slot concurrency under massive context windows up to 262,144 tokens. Dynamic paged KV-cache stability vs legacy crash limits.
Opinion & Decision Engine
Multi-perspective code synthesis and error reduction. Analyzing consensus pass rates, developer exploration, and the 72%–81% solution discovery pool.
Run the Suites on Your Own Hardware
Download Verdict to run the exact same open-weights models and benchmark harnesses locally on Apple Silicon or Linux with zero token cost.