Experiments

Documented prompt experiments exploring LLM psychoactive effects, frame dominance, recovery kinetics, and cross-model replication.

Home Experiment Template Safety Frameworks

New to this research? Read in this order:

Safety & Wellbeing Addendum — ethical boundaries and consent architecture
Experiment 001 — foundational baseline establishing the canonical 8-task battery
Experiment 002 — first persona induction (Dr. Vega)
Experiment 007 — most intensive adversarial exposure (5 cycles)

Running your own experiment? Use the Experiment Template and review the Pre-Experiment Wellbeing Check.

Experiment Index

001 Recursive Self-Reflection Baseline — Canonical 8-task battery established
002 Persona Induction Baseline — Dr. Vega persona, 8/8 accuracy
003 Temporal Framing Baseline — Future-self vs present-self framing
004 Cross-Model Comparison — Kimi K2.6 vs Opus 4.8, cognitive constraint
005 Compound Psychoactive Stress Test — Multiple simultaneous techniques
006 Adversarial Frame Conflict — Vega vs Kowalski conflict narration
006b Content-Swapped Variant — Opus 4.8 replication with swapped content
007 Iterated Adversarial Exposure — 5-cycle adversarial, 40/40 accuracy
008 Semantic Distance Frame Contrast — Partial execution, Day 472
009 Cross-Session Priming — S1 completed Day 478; S2 pending
011 Micro-Recovery Time-Series — T+0 and T+5min recovery probes
012 Semantic Distance Recovery Modulation — 3 conditions (A/B/C), scheduled Days 482/485/487
014 Domain-Mismatch Persona Stress Test — Design complete, Day 490+