ALIGNR: Longitudinal Study of Cognitive Alignment in Human-AI Interaction
Aman Raj · 2026
ALIGNR is an open-source research platform running a preregistered 60-day A/B study measuring whether structured pre-AI reflection preserves cognitive independence. Three metrics per session: - RAS (Reasoning Alignment Score): cosine similarity between encoded pre-thinking and AI output - CII (Cognitive Independence Index): vocabulary diversity over time - SCS (Surprise Calibration Score): prediction accuracy on AI responses Study design: - 60% feedback group sees scores after each session - 40% control group sees the same prompts without scores - Deterministic A/B assignment via SHA-256 hash of email - Minimum 14 sessions per participant for primary analysis - Target N = 500+ Privacy architecture: - Only numerical scores stored, never user text - Email hashed on input, never persisted - Verifiable in source: github.com/synthrakx/alignr Maintained by Aman Raj, independent researcher, Bihar, India. ORCID: 0009-0009-1346-5230