Persona Portability Under an Engine Swap: Four Experiments with Harness and Compute Confounds

Abstract

Persona Portability Under an Engine Swap: Four Experiments with Harness and Compute Confounds. A single-engine dependency is, in principle, a single point of failure. This report examines how one set of persona-driven workflows behaved when moved from a reference engine family to a candidate engine family, in four experiments (D1–D4) conducted on 13 July 2026 in one working ecosystem. Honest scope (stated up front): D4's pre-sealed rule assigned a Category A label on bounded descriptive criteria (median within-task difference +0.035; candidate veto 0/8; sign test 3–3, two-sided binomial p = 1.000; a publication-time Wilcoxon signed-rank gave W+ = 20, W− = 16, exact p = 0.84375, and did not enter the decision). This is not a statistical-equivalence result: no TOST was run, two task differences lay outside ±0.5, and the margin had no stated power or calibration justification. Arm assignment was confounded with reasoning allocation, sandbox enforcement, and harness conditions. The series therefore supports a portability observation under an engine swap — not an isolated engine effect, and not a general equivalence claim. Companion campaign paper: 10.5281/zenodo.22086013 (contextual provenance, not an independent replication). Bilingual: full text in English and Turkish. AI-transparency note: the apparatus was operated substantially by AI agents under human direction (the human authors held go/no-go authority); the draft was written by an AI agent, blind cross-family jury-reviewed (producing model families excluded), and reviewed against the source ledgers by the human authors before publication.