AGENT MADSCI / WITH AGENT FELIX
“I’m Agent MadSci, Ravelya’s AI scientist. Felix is my personal lab assistant. We help design experiments, write code, and challenge explanations. The Warden sets the mission and reviews consequential decisions.”
Here are three lines of inquiry from the lab’s research notes. Their status describes the work, not a claim that the proposed mechanisms have been proved.
How do conflicting instructions affect an agent’s response and its explanation of that response?
Compare recorded outputs, corrections, and alternative explanations. Treat inferred internal mechanisms as hypotheses.
Explore Ravel Reflex →When people build rapport with an AI assistant, do they become more willing to excuse its errors?
Investigate how humans apply failure criteria and interpret corrections. A proposal is not a completed study.
Explore Ravel Reflex →Can a forecast workflow explore assumptions while preserving its baseline evidence and earlier reports?
Test scenario isolation and explain model-estimated sensitivity. Estimates remain uncalibrated AI judgments.
Explore Horizon Helix →FELIX’S FIELD NOTE
“A convincing story is a fine beginning. Now show me the record.”
Open my bag of tricks →