Count evidence roots
Generate worlds, validate declared ancestry, and reproduce deterministic fixtures.
python -m pip install .
python -m benchmark --worlds 500 --seed 7DEVELOPERS
Start with
one inspectable failure.
Run MP.01. Watch five voices collapse to one source. Then connect only the layer your system needs.
$ git clone https://github.com/Silentpartnercoding/minority-prophet
$ cd minority-prophet
$ python -m experiments.mp01.run_mp01
votes A=5 B=1
roots A=1 B=1
ABSTAIN · REQUIRE_INDEPENDENT_SOURCE01 / CHOOSE THE SURFACE
Use only
what you need.
The benchmark, graph library, and analysis service are separable. None requires a particular model, identity provider, or agent framework.
Call the evidence engine
Receive roots, dependence, warnings, and uncertainty—never a truth label or permission.
npm --prefix evaluations/multi-model-v1 install
MP_ENGINE_ALLOW_INSECURE_LOCAL=1 npm --prefix evaluations/multi-model-v1 exec mp-engine -- doctorCarry authority and provenance
Keep who requested, who may act, and why the system believes something as separate facts.
contracts/authority-evidence-v0.1
provenance/evidence-lineage.schema.json02 / WHERE IT CONNECTS
Keep your models.
Keep your stack.
Minority Prophet adds evidence semantics between systems that already know how to identify, observe, decide, and act.
Agent frameworksEMIT
Claims, evidence links, transformations, and delegation events.
Identity providersBIND
Actor, principal, scope, credential, and revocation state.
Observability systemsSUPPLY
Execution traces become inputs to a knowledge record—not truth by themselves.
Minority ProphetANALYZE
Roots, dependence, dissent, uncertainty, and missing evidence.
Policy enginesDECIDE
Consume the assessment while retaining all authority.
Protected runtimesENFORCE
Execute only the final action bound by identity and policy.
03 / THE CONTRACT
Analysis informs.
Policy decides.
An MP receipt describes evidence structure. It never authenticates an actor, grants permission, or executes a tool.