Reasoning Efficiency Engine
Token counters tell you what you spent. CARTIE tells you whether the tokens resolved an intent. No other FinOps tool measures this — yet.
Of the $0 spent on agent traces, 0% leaked on unresolved loops.
Worst features by cost-per-resolved-intent
The features wasting most $ per useful answer.
How this metric works
For every trace, we record total cost and step count from your existing trace pipeline. A trace is resolved when an engineer (or end-user) gives a thumbs-up via the feedback chip in the Trace Viewer. Cost per Resolution = Σ(resolved trace cost) ÷ count(resolved).Loop waste % = Σ(unresolved trace cost) ÷ total cost. A 20% loop waste rate on a $50k/mo agentic workflow is $10k/mo of unproductive tokens — that's the leak this engine surfaces.
First-to-market: as of May 2026 no other FinOps tool measures step-to-resolution efficiency.