Trellison AI Assessment System — cognitive profiling, niche-routed benchmarking, and continuous capability evaluation for production AI systems.
TAAS is the empirical layer underneath every Trellison Institute claim about AI capability. It scores models on per-niche tasks (coding, analysis, summarization, classification, agentic), maintains a multi-armed-bandit feedback loop, and routes production traffic toward whichever provider currently leads each niche.
Niche-routed benchmarks. Rather than one monolithic leaderboard, TAAS evaluates models per task type. Different niches surface different leaders — and we route accordingly.
Continuous feedback. Each production execution records caller, niche, latency, error class, and downstream outcome. The bandit updates rankings over time without any manual leaderboard maintenance.
Methodology audits. TAAS doesn't grade conclusions — it grades methodology. Was the prompt structured correctly? Did the response satisfy the contract schema? Did the routing decision use evidence that's still current?
TAAS is operational and routes the majority of internal LLM calls at DaedArch and Trellison Institute. Public-facing API and detailed methodology documentation are in development.
Trellison Institute · empirical research, independent evaluation