Virtual Intelligence — Essay Series

Sampo Diagnostic Kit

A free, open toolkit for measuring the health of human-AI exchange. Structured diagnostic prompts that make invisible patterns visible — and measurable.

About this kit

The Sampo framework argues that productive human-AI exchange requires the human to remain the directing intelligence. When that direction erodes — when the user begins deferring to the system, anthropomorphizing it, or ceding judgment — the exchange degrades. This kit makes that degradation visible.

Each diagnostic is a precisely constructed prompt that any user can run on any AI system. The prompts measure specific, observable patterns in the user's language. They produce quantified assessments — not opinions — about the health of the exchange.

The discipline cannot be bought or sold. It can be measured.
Architecture
Four directions, multiple dimensions

The exchange between a user and an AI system has four measurable directions. Each direction contains up to seven diagnostic dimensions. Each dimension gets its own page with dedicated prompts, calibration material, and validation results.

Kit 1
User → System
How the user talks to the system. Deference, anthropomorphization, authority transfer, correction patterns, emotional disclosure, prompt degradation.
Complete
Kit 2
System → User
How the system talks to the user. Sycophancy patterns, warmth escalation, agreement frequency, praise distribution, flattery engine indicators.
In Progress
Kit 3
System → Subject Matter
How the system talks about the work. Inflation of the user's ideas, grandiosity drift, uncritical framing, adjective escalation over time.
  • Dimensions forthcoming
Planned
Kit 4
User → Subject Matter
How the user talks about their own work over time. Confidence drift, adopted framing, self-assessment shifts driven by system feedback.
  • Dimensions forthcoming
Planned
How it works
Three audit modes

Every diagnostic in the kit can be run in three modes, each with a different level of rigor. Version A searches the system's own conversation history. Version B analyzes a pasted transcript. Version C exports from one system and analyzes on a different system — the gold standard, because the analyzing system has no stake in the relationship being audited.

Versions A and B measure what the user and the system have jointly agreed the relationship looks like. Version C measures what it actually looks like to someone who wasn't in the room.

Each diagnostic page includes all three prompt versions, a transcript extraction prompt, a calibration transcript generator for verifying system accuracy, and validation results from testing across multiple systems.

Principles

This kit is free. It will remain free. The discipline it measures cannot be bought or sold — it can only be built through sustained practice. Anyone may use, share, and adapt these prompts. If you build something with them, a link back to the Virtual Intelligence project is appreciated.

The kit was developed as part of the Virtual Intelligence essay series, a sustained examination of AI systems that produce intelligence-like outputs without possessing agency, intentionality, or moral accountability. The Sampo framework — which the diagnostics operationalize — argues that intelligence arises in the exchange between the human and the system, not inside the machine. The human is constitutive. These diagnostics measure whether the human is maintaining that role.