HAL — Holistic Agent Leaderboard

A standardized, cost-aware agent-evaluation harness and leaderboard from Arvind Narayanan's CITP and PLI groups (ICLR 2026), with the harness live since mid-2024. HAL is the "missing infrastructure" for agent evals: it runs agents across benchmarks under controlled, contamination-aware, cost-controlled conditions. It follows the group's influential critique "AI Agents That Matter" (2024), which established cost-controlled, reproducible agent-evaluation norms.