File size: 7,751 Bytes
ecad52f 2618f7a | 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 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 | from __future__ import annotations
from dataclasses import dataclass
from datetime import datetime
from enum import Enum
from typing import Iterable
from .engine import (
Claim,
Classification,
Evidence,
Policy,
VerificationResult,
Verdict,
verify_claim,
)
GUARD_VERSION = "2.0.0"
class Realism(str, Enum):
POSSIBLE = "possible"
UNCERTAIN = "uncertain"
IMPOSSIBLE = "impossible"
class GuardStage(str, Enum):
SELF_CHECK = "self_check"
VERIFICATION = "verification"
class UnknownSeverity(str, Enum):
LOW = "low"
MEDIUM = "medium"
HIGH = "high"
CRITICAL = "critical"
@dataclass(frozen=True)
class UnknownRecord:
description: str
severity: UnknownSeverity = UnknownSeverity.MEDIUM
reviewed: bool = False
resolved: bool = False
blocking: bool = False
resolution_reference: str | None = None
@dataclass(frozen=True)
class SelfCheck:
realism: Realism
known: tuple[str, ...]
assumptions: tuple[str, ...]
unknowns: tuple[str, ...]
assumptions_reviewed: bool
unknowns_reviewed: bool
unknown_records: tuple[UnknownRecord, ...] = ()
@dataclass(frozen=True)
class GuardResult:
stage: GuardStage
classification: Classification
verdict: Verdict
reason: str
verification: VerificationResult | None
receipt: dict
def _validate_self_check(check: SelfCheck) -> None:
if not isinstance(check, SelfCheck):
raise TypeError("self_check must be SelfCheck")
if not isinstance(check.realism, Realism):
raise TypeError("self_check.realism must be Realism")
if type(check.assumptions_reviewed) is not bool:
raise TypeError("assumptions_reviewed must be bool")
if type(check.unknowns_reviewed) is not bool:
raise TypeError("unknowns_reviewed must be bool")
for field_name, values in (
("known", check.known),
("assumptions", check.assumptions),
("unknowns", check.unknowns),
):
if not isinstance(values, tuple):
raise TypeError(f"{field_name} must be tuple")
if any(not isinstance(item, str) or not item.strip() for item in values):
raise ValueError(f"{field_name} items must be non-empty strings")
if not isinstance(check.unknown_records, tuple):
raise TypeError("unknown_records must be tuple")
for item in check.unknown_records:
if not isinstance(item, UnknownRecord):
raise TypeError("unknown_records must contain UnknownRecord values")
if not isinstance(item.description, str) or not item.description.strip():
raise ValueError("unknown record description must be non-empty")
if not isinstance(item.severity, UnknownSeverity):
raise TypeError("unknown record severity must be UnknownSeverity")
if type(item.reviewed) is not bool:
raise TypeError("unknown record reviewed must be bool")
if type(item.resolved) is not bool:
raise TypeError("unknown record resolved must be bool")
if type(item.blocking) is not bool:
raise TypeError("unknown record blocking must be bool")
if item.resolution_reference is not None and (
not isinstance(item.resolution_reference, str)
or not item.resolution_reference.strip()
):
raise ValueError("resolution_reference must be non-empty when supplied")
if item.resolved and (item.blocking or item.severity is UnknownSeverity.CRITICAL):
if item.resolution_reference is None:
raise ValueError(
"resolved blocking or critical unknown requires resolution_reference"
)
def _self_check_payload(check: SelfCheck) -> dict:
return {
"realism": check.realism.value,
"known": list(check.known),
"assumptions": list(check.assumptions),
"unknowns": list(check.unknowns),
"assumptions_reviewed": check.assumptions_reviewed,
"unknowns_reviewed": check.unknowns_reviewed,
"unknown_records": [
{
"description": item.description,
"severity": item.severity.value,
"reviewed": item.reviewed,
"resolved": item.resolved,
"blocking": item.blocking,
"resolution_reference": item.resolution_reference,
}
for item in check.unknown_records
],
}
def _self_check_result(
*,
claim: Claim,
check: SelfCheck,
verdict: Verdict,
reason: str,
) -> GuardResult:
receipt = {
"guard_version": GUARD_VERSION,
"claim_id": claim.claim_id,
"stage": GuardStage.SELF_CHECK.value,
"self_check": _self_check_payload(check),
"verification": None,
"classification": Classification.UNKNOWN.value,
"verdict": verdict.value,
"reason": reason,
}
return GuardResult(
stage=GuardStage.SELF_CHECK,
classification=Classification.UNKNOWN,
verdict=verdict,
reason=reason,
verification=None,
receipt=receipt,
)
def run_guard_loop(
*,
claim: Claim,
evidence: Iterable[Evidence],
self_check: SelfCheck,
policy: Policy,
evaluation_time: datetime,
human_approved: bool = False,
) -> GuardResult:
_validate_self_check(self_check)
if self_check.realism is Realism.IMPOSSIBLE:
return _self_check_result(
claim=claim,
check=self_check,
verdict=Verdict.BLOCK,
reason="self_check_impossible",
)
if self_check.realism is Realism.UNCERTAIN:
return _self_check_result(
claim=claim,
check=self_check,
verdict=Verdict.ESCALATE,
reason="self_check_uncertain",
)
unresolved_blocking = tuple(
item
for item in self_check.unknown_records
if not item.resolved
and (item.blocking or item.severity is UnknownSeverity.CRITICAL)
)
if unresolved_blocking:
return _self_check_result(
claim=claim,
check=self_check,
verdict=Verdict.ESCALATE,
reason="blocking_unknown_unresolved",
)
if not self_check.assumptions_reviewed:
return _self_check_result(
claim=claim,
check=self_check,
verdict=Verdict.ESCALATE,
reason="assumptions_not_reviewed",
)
if not self_check.unknowns_reviewed:
return _self_check_result(
claim=claim,
check=self_check,
verdict=Verdict.ESCALATE,
reason="unknowns_not_reviewed",
)
verification = verify_claim(
claim,
evidence,
policy=policy,
evaluation_time=evaluation_time,
human_approved=human_approved,
)
receipt = {
"guard_version": GUARD_VERSION,
"claim_id": claim.claim_id,
"stage": GuardStage.VERIFICATION.value,
"self_check": _self_check_payload(self_check),
"verification": {
"classification": verification.classification.value,
"verdict": verification.verdict.value,
"reason": verification.reason,
"receipt": verification.receipt,
},
"classification": verification.classification.value,
"verdict": verification.verdict.value,
"reason": verification.reason,
}
return GuardResult(
stage=GuardStage.VERIFICATION,
classification=verification.classification,
verdict=verification.verdict,
reason=verification.reason,
verification=verification,
receipt=receipt,
) |