Evidence before infrastructure
The current priority is a small, falsifiable empirical study. Longer-term infrastructure is presented as a research direction—not as an already validated system.
Can a fixed review protocol improve the ability of human and AI reviewers to detect consequential errors in management research?
Research object
Short management-research manuscripts paired with controlled error-injected versions.
Treatment discipline
Every injected error must change at least one correct answer in the structured review form.
Comparison
Human and AI reviewers receive the same manuscript, questions, instructions, and evaluation budget.
Primary outcomes
Error-detection sensitivity and false-positive rate, followed by time, agreement, and explanation quality.
The two conceptual papers argue that independent verification may become the binding constraint in AI-accelerated science. This project tests a narrower prerequisite: whether a structured evaluation interface changes reviewers’ ability to notice errors that matter for scientific judgment.
A shortened source manuscript and a controlled error-injected counterpart have been assembled.
A domain expert must confirm the clean baseline and the mapping from each injected error to the affected review response.
A small 3–5 participant study will test task duration, comprehension, scoring, and implementation before any larger comparison.
A larger study will be considered only if the materials and measurement protocol survive the preceding gates.
Current collaboration need
I am seeking domain experts who can independently assess whether a short manuscript is a defensible clean baseline and whether the controlled errors truly change the warranted structured-review responses.
Future research directions
A modular system for claim-, evidence-, code-, data-, assumption-, and provenance-aware scientific checking.
A versioned state that preserves verified findings, disagreements, dependence, scope, and revision history.
A system that selects research actions according to expected epistemic value rather than manuscript production alone.