Measures whether a model gets to the answer efficiently or burns tokens on preambles,
repetition, hedging, retries, dead-end reasoning, and formatting overhead.
40
Tasks
5
Task Types
8
Waste Categories
0
Judge Calls
--
LLM Status
Run Token Waste Audit
Starting...
Preparing tasks...pending
--
Efficiency Score
--
Grade
--
Avg Waste Ratio
--
Actual vs Optimal
Per-Category Waste Breakdown
Task Type Efficiency
Per-Task Results
Benchmark Tasks
Loading...
Waste Taxonomy
The classifier uses deterministic heuristics only. No LLM judge is called to score token waste.
Revenue Stream #4 metadata is captured in each run's metrics JSON: model name, task type,
dominant waste category, token counts, estimated cost, and waste ratio.