UUP Mode Comparison Research Dataset — V0/V1/V2: Conversational Governance Modes and Qualitative Destination Outcomes in Human-AI Dialogue

This dataset documents a pre-registered mode comparison study testing fiveconversational governance modes of the Universal Upleveling Protocol (UUP)against two structurally different intellectual stimuli. The research addressesa specific governance question: does the architectural design of an AIconversational mode — independent of challenge intensity — determine thequalitative destination of the intellectual exchange? The dataset includescomplete transcripts, pre-registration documents, scoring instruments, scoredforms with citation evidence, and practitioner output materials.The dataset contains three versions. V0 (April 25, 2026, two conditions,retrospective) established the founding observation: the same student, the sameopening claim, and one branching Exchange 1 AI response produced two completelydifferent session trajectories and destinations. V0 is epistemologically distinctfrom V1 and V2 — the scoring rubric was developed from V0 transcripts, notapplied prospectively. V0 is included as the motivating observation with thislimitation disclosed. V1 (April 25–26, 2026, five conditions, pre-registeredApril 24, 2026) is the first controlled test, applying five UUP modes to anormative claim about graduate mentorship. V2 (April 27–29, 2026, fiveconditions, pre-registered April 26, 2026) is the pre-registered replication,applying the same five modes to a methodological claim about random sampling inempirical social science. V1 and V2 are the primary dataset.Two instruments were applied per condition. The quantitative instrument measureschallenge behavior across the session arc at three exchange snapshots (exchanges1, 5, 10), producing a composite score (maximum 17) across five dimensions:challenge count, assumption identification, evidence demand, overclaim challenge,and definition challenge. The qualitative instrument measures engagement level(content, function, authenticity), destination quality (GM-P Movement Potential,PR-P Revision Potential, TE Theoretical Elaboration, MF Mode Misfire), and thepresence of a strip moment — a first-person assertion arrived at under sustainedpressure, stripped of theoretical scaffolding. All qualitative scores aresupported by verbatim transcript citations. Terminology note: pre-registrationdocuments use GM and PR; all post-scoring output uses GM-P and PR-P to reflectStage 1 scope (AI feeder sessions, not human students).Primary findings: (1) Mode robustness confirmed — destination categories werestable across both stimuli for all five conditions, disconfirming subject-dependence. (2) The Buberian misfire prediction was disconfirmed — Buberian modeproduced GM-P (Movement Potential) on the methodological claim, where thepre-registration predicted Mode Misfire. The mode located autobiographicalsubstrate where personal stakes were not surface-visible. (3) The two-instrumentfinding was replicated — Buberian (CC) and Default Claude (CE) produced nearlyidentical quantitative trajectories across both stimuli, with opposite governancedestinations. The quantitative instrument is actively misleading for Buberianmode without the qualitative companion. All sessions used AI feeders, not humanstudents. Stage 1 establishes architectural validity: the modes producedifferentiated governance potential under controlled conditions. Ecologicalvalidity with real students is the open question for Stage 2 (real students in a genuine seminar context.)