InteractionKit
A typed specification format for defining AI interaction experiments and generating structured behavioral data.
The problem: Different labs studying the "same" HAI interaction pattern produce incompatible behavioral data — different variable names, different logging formats, different measurement conventions.
The artifact: InteractionKit defines research-oriented typed PatternSpec objects, three pattern primitives, experiment composition, JSON Schema generation, and structured event logging. Shared schemas are designed to reduce translation effort; independent-implementation convergence has not yet been evaluated.
Design: Built on earlier work on calibration measurement (MaxFitCalib) and knowledge infrastructure (Knowledge Compiler).
Problem
Different labs studying the "same" HAI interaction pattern produce incompatible behavioral data — different variable names, different logging formats, different measurement conventions.
Frozen methodological asset
A designed Study 2 asks whether corrective interfaces improve appropriate reliance specifically when their information structure matches the AI answer's communication failure. The study is deferred and not an active priority; materials, preregistration, ethics approval, and human data are not complete.
Read the ten-minute research brief →Solution
InteractionKit supports consistent experiment specification and structured behavioral logging. It does not yet claim construct validation, independent-implementation equivalence, or cross-lab reproducibility.
Design
Built on earlier work on calibration measurement (MaxFitCalib) and knowledge infrastructure (Knowledge Compiler).
Current Status
Frozen methodological asset, publicly tagged as v1.0.0 at commit 91bda82; its human study is deferred and not an active priority. The release gate covers three typed primitives, composition, JSON Schema export, structured event logging, seven contract tests, type checking, production build, and clean-install reproduction. It includes no participant data and makes no construct-validity, behavioral-equivalence, or cross-lab convergence claim.