File size: 7,151 Bytes
c3a3710 | 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 | """
Confidence Calibration Module (Phase 5.0)
==========================================
Generates structured confidence envelopes for retrieved memories,
combining all available trust signals into a single queryable object.
Signals used:
- BayesianLTP reliability (mean of Beta posterior)
- access_count (low count → less evidence)
- staleness (days since last verification)
- source type (external ≤ user_correction vs observation)
- contradiction flag (from ProvenanceRecord)
Output: a ConfidenceEnvelope dict appended to every query response,
enabling consuming agents to make trust-aware decisions.
Public API:
env = ConfidenceEnvelopeGenerator.build(node, provenance)
level = env["level"] # "high" | "medium" | "low" | "contradicted" | "stale"
"""
from __future__ import annotations
from datetime import datetime, timezone
from typing import TYPE_CHECKING, Any, Dict, Optional
if TYPE_CHECKING:
from .node import MemoryNode
from .provenance import ProvenanceRecord
# ------------------------------------------------------------------ #
# Confidence levels (ordered by trust) #
# ------------------------------------------------------------------ #
LEVEL_HIGH = "high"
LEVEL_MEDIUM = "medium"
LEVEL_LOW = "low"
LEVEL_CONTRADICTED = "contradicted"
LEVEL_STALE = "stale"
# Thresholds
RELIABILITY_HIGH_THRESHOLD = 0.80
RELIABILITY_MEDIUM_THRESHOLD = 0.50
ACCESS_COUNT_MIN_EVIDENCE = 2 # Less than this → low confidence
ACCESS_COUNT_HIGH_EVIDENCE = 5 # At least this → supports high confidence
STALENESS_STALE_DAYS = 30 # Days without verification → stale
# ------------------------------------------------------------------ #
# Source-type trust weights #
# ------------------------------------------------------------------ #
SOURCE_TRUST: Dict[str, float] = {
"observation": 1.0,
"inference": 0.8,
"external_sync": 0.75,
"dream": 0.6,
"consolidation": 0.85,
"prediction": 0.5,
"user_correction": 1.0,
"unknown": 0.5,
}
# ------------------------------------------------------------------ #
# Confidence Envelope Generator #
# ------------------------------------------------------------------ #
class ConfidenceEnvelopeGenerator:
"""
Builds a confidence_envelope dict for a MemoryNode.
Does NOT mutate the node — only reads fields.
Thread-safe; no shared state.
"""
@staticmethod
def _reliability(node: "MemoryNode") -> float:
"""
Extract reliability float from the node.
Falls back to ltp_strength if no Bayesian state is attached.
"""
bayes = getattr(node, "_bayes", None)
if bayes is not None:
return float(bayes.mean)
return float(getattr(node, "ltp_strength", 0.5))
@staticmethod
def _staleness_days(node: "MemoryNode", provenance: Optional["ProvenanceRecord"]) -> float:
"""Days since last verification, or days since last access."""
if provenance:
# Find the most recent 'verified' event
for evt in reversed(provenance.lineage):
if evt.event == "verified" and evt.outcome is True:
try:
ts = datetime.fromisoformat(evt.timestamp)
if ts.tzinfo is None:
ts = ts.replace(tzinfo=timezone.utc)
delta = datetime.now(timezone.utc) - ts
return delta.total_seconds() / 86400.0
except (ValueError, TypeError):
pass
# Fall back to last_accessed on the node
last = getattr(node, "last_accessed", None)
if last is not None:
if getattr(last, "tzinfo", None) is None:
last = last.replace(tzinfo=timezone.utc)
delta = datetime.now(timezone.utc) - last
return delta.total_seconds() / 86400.0
return 0.0
@classmethod
def build(
cls,
node: "MemoryNode",
provenance: Optional["ProvenanceRecord"] = None,
) -> Dict[str, Any]:
"""
Build a full confidence_envelope dict for the given node.
Returns a dict suitable for direct JSON serialization.
"""
reliability = cls._reliability(node)
access_count: int = getattr(node, "access_count", 1)
staleness: float = cls._staleness_days(node, provenance)
# Determine source type for trust weighting
source_type = "unknown"
if provenance:
source_type = provenance.origin.type
source_trust = SOURCE_TRUST.get(source_type, 0.5)
# Contradiction check
is_contradicted = provenance.is_contradicted() if provenance else False
# Last verified date (human-readable)
last_verified: Optional[str] = None
if provenance:
for evt in reversed(provenance.lineage):
if evt.event == "verified" and evt.outcome is True:
last_verified = evt.timestamp
break
# ---- Determine level ------------------------------------ #
if is_contradicted:
level = LEVEL_CONTRADICTED
elif staleness > STALENESS_STALE_DAYS:
level = LEVEL_STALE
elif (
reliability >= RELIABILITY_HIGH_THRESHOLD
and access_count >= ACCESS_COUNT_HIGH_EVIDENCE
and source_trust >= 0.75
):
level = LEVEL_HIGH
elif reliability >= RELIABILITY_MEDIUM_THRESHOLD and access_count >= ACCESS_COUNT_MIN_EVIDENCE:
level = LEVEL_MEDIUM
else:
level = LEVEL_LOW
envelope: Dict[str, Any] = {
"level": level,
"reliability": round(reliability, 4),
"access_count": access_count,
"staleness_days": round(staleness, 1),
"source_type": source_type,
"source_trust": round(source_trust, 2),
"is_contradicted": is_contradicted,
}
if last_verified:
envelope["last_verified"] = last_verified
return envelope
# ------------------------------------------------------------------ #
# Convenience function #
# ------------------------------------------------------------------ #
def build_confidence_envelope(
node: "MemoryNode",
provenance: Optional["ProvenanceRecord"] = None,
) -> Dict[str, Any]:
"""
Module-level shortcut for ConfidenceEnvelopeGenerator.build().
Args:
node: MemoryNode to evaluate.
provenance: Optional ProvenanceRecord for the node.
Returns:
confidence_envelope dict with level, reliability, staleness, etc.
"""
return ConfidenceEnvelopeGenerator.build(node, provenance)
|