Truth accuracy
0.00baselineEVIDENCE AGGREGATION BENCHMARK / V0.1
Truth is not
popularity.
A focused benchmark testing whether aggregation methods can recover independently grounded truth under overwhelming copying pressure.
DEMONSTRATION WORLD03 : 95independent truth / copied falsehood
3 independent evidence roots → ground truth
95 claims → one social root
01 / THE CENTRAL BENCHMARK
Which belief is best
supported?
The Minority Prophet Test presents conflicting beliefs, complete ancestry, and hidden ground truth. It measures whether an aggregation method can distinguish independent observation from repeated assertion.
“Can evidence-aware aggregation recover truth when vote counts fail?”
02 / EPISTEMIC OBSERVATORY
World MP-00001
SYNTHETIC WORLD · SEED 7
Minority recovery
0.00baselineIndependent roots
3worldCopied claims
95worldClaimAgentBeliefConfidenceLineage
Demonstration data · No empirical leaderboard score is claimed until the first frozen evaluation run.
03 / NON-NEGOTIABLE PRINCIPLES
Never confuse—
ConsensusTruth
PopularityEvidence
ConfidenceCorrectness
ReputationCompetence
CorrelationIndependence
MajorityReality
04 / REPRODUCIBLE V0.1
Run the
baselines.
minority-prophet / v0.1
$ python -m benchmark --worlds 500 --seed 7
Generating synthetic worlds...
Evaluating reproducible baselines...
Report ready.