Measurement, Evidence & Research Infrastructure
How can complex longitudinal human–AI systems be observed, measured, reconstructed and studied in a scientifically traceable way?
What this space studies
The measurement and research infrastructure for the other seven spaces. Concretely: kernel, runtime, provenance, datasets, metrics, and the Observatory and SELFGRAPH systems as measuring instruments.
Central questions
- How can change be observed over long periods?
- How does it stay traceable where a statement comes from?
- How is observed evidence kept apart from inferred evidence?
Research object
Observing such systems reliably without confusing instrument, interpretation and object.
More than technical infrastructure
The research object also includes how measurement instruments, data models and research infrastructure themselves influence or distort statements.
Emergence has to be operationalized before it counts as a finding
In space H, emergence is a claim whose operationalization must be tested before it is reported as a finding. The space asks which statements are directly observed, which structure is derived from them, which hypotheses stay open and which temporal and relational data are needed. Without this separation, what was measured and what was interpreted blur together.
Research programs
This space has no main program of its own. The programs below belong to other spaces and touch it.
Longitudinal Interaction Impact Evaluation
How do multi-year human–AI interactions develop: beneficially, stably or with increasing deterioration?
Program details for Longitudinal Interaction Impact Evaluation →Neuroadaptive Human–AI SystemsPlanned
How could a system adapt to the changing states of a person? Today EIL uses only interaction data and behavioral proxies for this.
Physiological signals, wearables and EEG/BCI are not implemented.
Program details for Neuroadaptive Human–AI Systems →
Core phenomena
- Observability
- Evidence
- Provenance
- Longitudinal measurement
- Validity
- Integrity
- Infrastructure
- Instruments
How this space is investigated
Observation and relation
This space is examined as a measuring instrument before it measures.
System Integrity & Runtime
Understanding when intelligent systems degrade without producing an explicit error.
Central question
Where does the actual system state quietly diverge from its previously validated state, without an error, log or alert showing it?
Memory, State and Identity
How a system, or an actor reconstructed from interaction, maintains a coherent state across sessions and years without unnoticed errors settling into it.
Central question
How does a system, or an actor reconstructed from interaction history, maintain a coherent state across sessions, restarts and years without unnoticed errors, outdated assumptions or apparent continuity permanently settling into that structure?
Methods, systems and axes
Methods (10)
- Semantic Drift AnalysisMethods and analysis procedures
- Behavioral Feature ExtractionMethods and analysis procedures
- Anomaly AnalysisMethods and analysis procedures
- UIPArchitectures, protocols and benchmarks
- LIIEArchitectures, protocols and benchmarks
Systems (12)
- ObservatoryResearch systems
- SELFGRAPHResearch systems
- MIRRORMulti-agent systems
- KernelShared technical foundation
- StateSystem state, memory, security & integrity
Cross-cutting research axes (5)
Evidence & limits
Four layers of an emergence statement
Observed are individual events with timestamp and context. Derived is a structure such as recurrence or stabilization, computed from such events over time. A hypothesis is the reading that interaction between components produces this structure. The theoretical interpretation is the concept of emergence that brings the layers together. It is not reported as a measurement result.
Required temporal and relational data
A statement about emergence needs temporal order: timestamped events, session and context identifiers, versioned states and provenance for each statement. It needs relations: who reacts to whom, which handoff carries which result, and which component holds a state. Without relational data only co-occurrence can be established, not effect. It also needs a control dataset in which the pattern should be absent.
Stability and the boundary to correlation
Stability is tested by checking whether the pattern recurs across time windows, sessions and contexts (recurrence, cross-context recurrence) and changes when conditions are deliberately varied. An emergent structure is more than a correlation: two quantities that occur together say nothing yet about how something arose. Only when the structure cannot be sufficiently explained by individual components, direct rules or mere repetition is it a candidate for emergence. It then remains a hypothesis until interventions support it.
Alternative explanations: status of testing
Alternative explanations include model or tool upgrades, changes of task, instrument drift, false continuity between sessions and confounding. Hypothesis H5 and space A name them as conditions under which a change may not count as an effect of the interaction. The existing entries contain no completed systematic test of these alternatives. They are open test questions and not excluded explanations.
Boundary of this space
Measurability is not the same as truth. Instruments, data models and metrics capture only what they can operationalize, and can themselves create systematic blind spots.
How to read status and origin
Every statement carries its status (observed, inferred, hypothesis, planned or unknown), no program is externally validated, and origin statements about systems are candidates.
Topics and publications
working formulation Formulated from the research program, not yet a fixed wording.
