EI-L0
None · Blind spot
Follows commands or rules, but human emotion does not change its judgment. Evidence required: Confirm that affective inputs do not change the system judgment or response.
EI Capability Ladder
This framework explains what each level can do and what evidence is required.
EI Capability Ladder
Six levels define capability boundaries and the evidence each level requires.
EI-L1
Kindergarten · Sensing
Detects very strong affective signals and gives a single fixed response. Evidence required: State which strong signals are recognized and test whether basic feedback remains stable across conditions.
EI-L2
Elementary school · Understanding
Combines modalities to understand basic current emotion and clearly expressed affective intent. Evidence required: Test each input and their combination, including what happens when signals are missing or conflict.
EI-L3
Middle school · Prediction
Develops a clearer self and uses change over time to anticipate intent and likely next actions. Evidence required: Test intent and next-state prediction across continuous interactions, reporting errors and applicable scope.
EI-L4
University · Resonance
Understands self, others and setting well enough to improve the emotional climate of an interaction. Evidence required: Evaluate whether support remains appropriate across time, context and people, with human review.
EI-L5
Beyond · Wisdom
Continues learning through plans, feedback and curiosity without ongoing external maintenance. Evidence required: Run longitudinal evaluation and record feedback, version changes, ethical limits and stop conditions.
Six things we check in an evaluation
A single input or result cannot define capability. Evaluation records sensing, understanding, memory, action, uncertainty and contextual robustness together.
01
Sensing
Input modalities, capture conditions, missing signals and noise range.
02
Understanding
State definitions, context use, conflict handling and interpretation boundaries.
03
Memory
Permission, relevance, retention, withdrawal and multi-turn behavior.
04
Action
Candidate actions, product rules, fallback paths and human intervention.
05
Uncertainty
Confidence bounds, unknown cases, distribution shifts and abstention conditions.
06
Context Robustness
Applicability boundaries across devices, environments, people and time.
How an evaluation runs
The protocol bounds the question and data before repeated evaluation, uncertainty reporting, versioning and pre-publication review.
- 01
Scope Declaration
Declare the research question, intended use, excluded uses and applicable contexts.
- 02
Dataset and Protocol
Record data sources, consent basis, splits, tasks and evaluation procedure.
- 03
Repeated Evaluation
Repeat evaluation under predeclared conditions and variations.
- 04
Uncertainty Reporting
Report errors, variation, unknown cases and failure conditions together.
- 05
Versioning
Record versions of models, data, protocols, configurations and interpretation rules.
- 06
Review
Review the method, scope, claim language and responsibility boundaries before publication.
What to check alongside a result
Home, Research and Benchmark use the same EI-L0–EI-L5 definitions. Finnox-V2.0 is an EI-L3 multimodal affective-intelligence model; the level describes its ability to use continuous information to analyze intent and likely future action sequences.
Applicability
Applicability boundaries across devices, environments, people and time. Declare the research question, intended use, excluded uses and applicable contexts.
Uncertainty
Confidence bounds, unknown cases, distribution shifts and abstention conditions. Report errors, variation, unknown cases and failure conditions together.
Versioning
Record versions of models, data, protocols, configurations and interpretation rules.
Known limits
Home, Research and Benchmark use the same EI-L0–EI-L5 definitions. Finnox-V2.0 is an EI-L3 multimodal affective-intelligence model; the level describes its ability to use continuous information to analyze intent and likely future action sequences.
Want to discuss the evaluation method?
Home, Research and Benchmark use the same EI-L0–EI-L5 definitions. Finnox-V2.0 is an EI-L3 multimodal affective-intelligence model; the level describes its ability to use continuous information to analyze intent and likely future action sequences.