KAIROS-ORBIT
A research framework for measuring operator fluency with AI systems
I wanted a rigorous, data-agnostic way to look at how my collaboration with AI systems actually changes over time, not a vibe.
KAIROS-ORBIT is the most research-shaped thing in the lab: a framework for analyzing how a person collaborates with AI systems over time, across planning, execution, verification, repair, and outcome integration. I built it because I spend most of my days operating agents, and I wanted a structured longitudinal view of that collaboration rather than a feeling about whether it was getting better. KAIROS is the fluency model — Knowledge grounding, Agency design, Instrumented execution, Reflexive calibration, Outcome integration, and Social/affective stance — and ORBIT is the submodel for relational bearing and interaction tone, scoped carefully to observable language patterns, not diagnoses of mood or personality.
The framework works at two levels. Lite runs on transcript-only data — ordinary chat exports — while Full is workflow-aware and joins transcripts to tool calls, artifacts, verification actions, and outcomes when that telemetry is available. The repo ships as deterministic reference tooling: install, build, and score a synthetic example, or generate a Markdown report from it, with a collector that turns local transcript exports into the Lite shape.
It is intentionally data-agnostic and public. The repository contains only documentation, JSON schemas, synthetic examples, and the deterministic scorer — no private transcripts, no personal datasets, no infrastructure details. That separation is the design: the framework is the contribution, and the private data it could run against stays private. It’s early and experimental, but it’s the lens I want pointed at my own logs once the Full telemetry join is wired up against the platform’s run records.