Conceptual reasoning · AI evaluation · Human-AI interaction · AI governance
Research on the stages of an interaction that occur before the outcomes that evaluation usually measures, and on how institutions review decisions delegated to automated systems. The work builds evaluation targets, rubrics, and case reviews that demonstrate model behavior separately from changes in the user’s own reasoning.
The framework
Each stage occurs before the outcomes that evaluation usually measures. The research names them first and treats evaluation as the downstream question.
What becomes available for consideration.
When interpretation settles before the evidence does.
How questioning develops through an interaction.
Rubrics, protocols, the Decision Authority Index, and case reviews.
Frame Substitution through priming: supplied context silently replaces direct looking, while the appearance of observation is preserved.
The Laundering Signal: one verifiable detail transfers its certainty to the unverifiable readings around it.
Working papers
Twelve papers and companion pieces, read in order. Each is preceded by a page stating, in the same order, what it defines, what it asks, how it proceeds, and where it sits in the series.
Perceptual Availability, Premature Coherence, Inquiry Trajectory, and the Mechanisms of Frame Substitution and Laundering
A Prior Evaluation Target in Human-AI Interaction
A Proposed Evaluation Target in Human-AI Interaction
A Missing Evaluation Target in Human-AI Interaction
How One Verifiable Detail Confirms a Stack of Unverifiable Readings
A Companion Analysis
Algorithmic Preorganization and the Conditions of Recognition
A Conceptual Framework
Perceptual Availability and the Prior Ecology Behind Platform Harm
A Small Lesson in Substitution
A Pre-Theoretical Exploration of Attention, Thresholds, and the Delay of Premature Coherence
Configuration as Epistemic Object
Earlier working paper
Relational Structures Across Scales
Review work
Published through Witness Mark. The Index rates what vendors document about limits on automation, human review, and routes to correction, which is a separate question from how well their systems perform. Each sector edition is scored on a four-part rubric: decision authority, reversibility, independent evidence, accountability.
Bibliography
Sixteen published, six in production.
What We Allow to Decide: ten essays on delegated decisions, evidence, and accountability, with follow-on posts.
Contact
For evaluation protocol design, pre-deployment review, or anything arising from the papers.