Illustrative collection map. The six families feed a shared schema; the visual does not represent an already-populated dataset.
Many collections. Shared questions.
Behaviour Corpus
Foundations for Modelling Human Behaviour
Behaviour needs
its context.
A movement, a conversation or a response to automation means more when the setting, measurement and source remain attached.
THE IDEAConnect collections.
Connect collections.
Preserve their differences.
A proposed collection framework spanning everyday life, work and human–AI interaction.
Six collection families would share a schema for context, behaviour, measures, outcomes and provenance. The aim is to make complementary evidence easier to explore while preserving what each source can support, where it is uncertain and how it was collected.
THE ENVIRONMENTSix families.
Six families.
More than one account of behaviour.
The framework spans everyday and social behaviour; body and activity; attention and physiological state; teams and communication; human–AI coordination; and adaptation and recovery. Each family keeps its own constructs and measurement context.
An initial runway-operations pilot would test evidence-linked annotation of how problems develop, are detected, communicated and recovered from. That pilot is one collection within the wider framework, rather than the boundary of the corpus.
Everyday life & embodied activity
Everyday and social behaviour covers routines, trust and wellbeing. Body and activity covers dimensions, fit, movement and posture in context.
Attention & collective work
Attention and physiological state covers gaze, workload and arousal as distinct constructs. Teams and communication covers handovers, clarification and coordination.
Human–AI interaction & adaptation
Human–AI coordination covers reliance, trust and handovers. Adaptation and recovery covers resumption, workarounds and responses under pressure.
RESEARCH VARIABLESWhat makes a collection
What makes a collection
useful beyond its origin?
The framework connects collections without treating different sources or constructs as interchangeable.
THE QUESTION
Similar labels can describe different things.
Preserve the setting, construct and measurement method behind a record. Gaze, workload, stress and trust need their own definitions rather than a single generic behaviour label.
- What changes
- Context, construct definitions and measurement methods
- What we observe
- Schema coverage, explicit definitions and missing information
Can a user tell when two collections support a meaningful comparison?
A corpus is also
a record of its decisions.
Collection provenance
Retain source selection, reuse documentation and the context in which records were produced.
Annotation history
Preserve definitions, independent judgements and reasons for adjudication.
Validation and limits
Report agreement, coverage and missingness alongside the data made available.
The six families are proposed collections. The initial runway pilot has a drafted protocol; retrieval and annotation have not started. Its voluntary, unverified incident reports cannot establish event rates or intervention effectiveness.
What can these
records help us understand?
The first step is to preserve enough context and uncertainty for that question to be answered carefully, across collections and domains.
Talk to the lab Next project: Air Traffic Decision VR Lab