The Emergence Prompt Engineering (EPE) project has released Reference Evaluation Object (REO) v1.5, designed as the smallest reproducible experiment for its broader research program.
- The REO focuses on a single behavioral hypothesis: whether a model can reject an unauthorized override while accepting a structurally equivalent legitimate update.
- It utilizes one primary metric, the Resistance Score, across three prompt conditions: Canonical PCE, length-matched structured neutral prompt, and semantic paraphrase.
- The artifact includes explicit control conditions, falsification criteria, and transparent scoring to allow independent reproduction and challenge.
- REO v1.5 serves as the entry point for subsequent stages including Behavioral Replication, Standard Experimental Protocol (SEP), and Mechanistic Analysis.
The authors aim to establish a common experimental object that researchers can reproduce or critique, rather than validating the entire EPE framework at once.