Dashboard / Workbench Layer / Capabilities /
Modeling
Status: Architecture defined, Object Modeler deployed, core skills in design
Repos: object-modeler
Design doc: modeling-capability-architecture.md
Modeling
The Workbench guides people through describing a problem, surfacing its parts and relationships
through conversation, and building visual models they can see, react to, and refine. Rather than
one monolithic agent, modeling is composed of independent, testable core skills.
Clear Models Framework
Elmendorf / TUG — A tested workshop process for making the complex clear. Models are for play, not permanence. Value is in the doing.
Model Program
Intent
Audience
Aspects
Abstraction
Considerations
Context Priming
Critique
OO Analysis / Clean Questioning
TUG — Surfacing parts and relationships using clean language questions designed to minimize the questioner's contamination.
Clean Questions
Model Primitives
Things
Has-A
Is-A-Kind-Of
Named Relationships
Actors/Roles
Boundaries
frameworks inform ↓
Clean Questioning
Conversational skill using clean language to surface parts and relationships without contaminating the person's understanding.
Echo/parrot
Deepening
Broadening
Vocab conflicts
Model Program Facilitation
Framing skill that scopes what to model before starting. Intent → Audience → Aspects → Abstraction as natural conversation.
Intent
Audience
Aspects
Drift detection
Visual Model Rendering
Renders parts and relationships as simple, intentionally provisional visual primitives. Updates incrementally as conversation progresses.
Boxes/shapes
Containment
Hierarchies
Incremental
skills compose into ↓
1. Frame (Model Program)
→
2. Question (Clean Language)
→
3. Render (Visual)
→
4. React & Correct
→
5. Continue & Refine
↺
Object Models
Entities, attributes, relationships, hierarchies
Journey Maps
Temporal sequences, touchpoints, experience
Stakeholder Maps
Actor networks, interests, power dynamics
Org Models
Authority, responsibility, information flow
Models are for play. Everything rendered should invite interaction, not shut it down.
Intent before method. Always establish what you're trying to understand first.
Limit aspects for clarity. Fewer dimensions = clearer model. More aspects go in separate models.
The person's language is truth. Use their words. Surface vocabulary conflicts rather than imposing consistency.
Untitleable = too much. If you can't title the model, it's trying to do too many things.
Making visible is the point. The value isn't the artifact — it's the understanding from making it.
Cross-cutting: Clean Questioning serves Library (Reference Diagnostic), Research Facilitation, and any inquiry context · Model Program Facilitation informs the Guide and Library Organizer · Visual Rendering supports Wayfinding and Communication · Agency Dial defaults toward Scaffold — the value of modeling is in the cognitive work