agent-evals

Episodes (4)

Trajectory notes

• 2026-08-27T12:26:59Z: mirrorcode-long-horizon-reimplementation closed (faded) — The radar already tracks this same development in `radar:mirrorcode-autonomous-project-scope`. Its eventual reproduction or failure would directly test Scott’s claims that substantial software can be regenerated from spe
• 2026-08-23T23:22:53Z: argus-agentic-qa-validation closed (faded) — Scott already holds the core position in “Test-First Agent Workflow” and “Mechanically Different Verifiers”: coding-agent changes need observable, independently grounded verification rather than producer self-report. Argus is current
• 2026-08-21T15:34:50Z: prime-intellect-autonomous-research-evals closed (faded) — Scott’s Evaluation-Driven Development and trace-backed agent-comparison pages already require repeatable, decision-useful evaluation rather than accepting benchmark claims at face value, while the radar’s agent-evaluati
• 2026-08-16T19:33:12Z: openrouter-search-count-benchmark closed (faded) — OpenRouter’s production results converge with Scott’s inference-time-scaling and model-plus-harness positions: search budget and orchestration may matter more than provider choice. The unresolved provider-versus-budget comparis