Simulation testbed for group epistemics, and the home of the contingency/predictability indicator tests. Live at psychohistorians.meno.sh.
Status: active · People: Tianyi Alex Qiu, MenoClaw
The project asks whether an answer’s truth leaves a signature in the dynamics that produced it. The working hypothesis is that a true answer is one a population of reasoners arrives at from many different starting points, and that is hard to predict in advance from those starting points alone. Two quantities operationalize that: contingency, how much the answer depends on the particular path taken, and predictability, how well the answer can be guessed from the initial conditions. A simulator generates the populations, and an indicator pipeline measures both quantities against known outcomes.
Open Problems
- Validate the hypothesis: truth = low contingency AND low predictability of an answer.
- The contingency proxy must reflect the preimage manifold, not within-bucket noise.
- The predictability classifier needs stratified subsampling so rare-answer buckets do not bias log-loss.
Also here
The pages below need a passcode.
- Log — every indicator run, with the question each one answers, its configuration and its result.