Token Efficiency Ratio (TER)
TERInformation Density per Token for Inference-Efficient Processing
Proposed hypothesis — not yet testedpublished
TER measures information density per token for inference efficiency.
By Marco Patrone
tertoken_efficiencyinference_optimizationrepresentation_primitive
Definition
TER measures how efficiently asset representations convey information per token. Higher TER indicates more compact, inference-efficient representation.
TER assesses how efficiently asset representations use tokens to convey information. Higher TER indicates more compact, inference-efficient processing.
Conceptual Formula
TER(r) = information_units / token_count, normalized 0-1.Methodology
Type
index construction
Data Sources
synthetic
Confidence Level
medium
Description
TER(r) = information_units / token_count, normalized 0-1.
Limitations
- Information unit definition is heuristic
- Tokenization varies by model
Key Takeaways
Key Points
- TER scales 0-1
- Higher is more efficient
- Reduces inference cost
Target Audience
infrastructure designersai systems
Relevance Tags
tertoken_efficiencyinference_optimizationrepresentation_quality
Source Paper
Citation
For the Token Efficiency Ratio (TER), see HomeSelf Research (2026), The Zero-Click Economy.