Spaces:
Sleeping
Sleeping
File size: 1,196 Bytes
26bf1c9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 | """
Scripted baseline policies for the CounterFeint FraudArena.
Each policy takes a role observation (dict from the WebSocket payload) and
returns a role-specific Pydantic action. They are deterministic, dependency-
free, and safe to use in CI or as opponents during self-play warmup.
Policies:
fraudster.ScriptedFraudster fixed 2-propose / end_turn / commit sequence
fraudster.ReactiveFraudster adapts category + slot based on verdicts seen
fraudster.GibberishFraudster low-plausibility adversary (negative control)
investigator.ScriptedInvestigator one-investigate-then-verdict heuristic
auditor.HeuristicAuditor rule-based flags + derived scores (Phase 1)
Replaced by graders/auditor_track_{a,b}.py
in Phase 2.
"""
from .auditor import HeuristicAuditor
from .fraudster import (
GibberishFraudster,
ReactiveFraudster,
ScriptedFraudster,
)
from .investigator import ScriptedInvestigator
__all__ = [
"GibberishFraudster",
"HeuristicAuditor",
"ReactiveFraudster",
"ScriptedFraudster",
"ScriptedInvestigator",
]
|