File size: 10,341 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 | """
Provenance Tracking Module (Phase 5.0)
=======================================
W3C PROV-inspired source tracking for MnemoCore memories.
Tracks the full lifecycle of every MemoryNode:
- origin: where/how the memory was created
- lineage: ordered list of transformation events
- version: incremented on each significant mutation
This is the foundation for:
- Trust & audit trails (AI Governance)
- Contradiction resolution
- Memory-as-a-Service lineage API
- Source reliability scoring
Public API:
record = ProvenanceRecord.new(origin_type="observation", agent_id="agent-001")
record.add_event("consolidated", source_memories=["mem_a", "mem_b"])
serialized = record.to_dict()
restored = ProvenanceRecord.from_dict(serialized)
"""
from __future__ import annotations
from dataclasses import dataclass, field
from datetime import datetime, timezone
from typing import Any, Dict, List, Optional
# ------------------------------------------------------------------ #
# Origin types #
# ------------------------------------------------------------------ #
ORIGIN_TYPES = {
"observation", # Direct input from agent or user
"inference", # Derived/reasoned by LLM or engine
"dream", # Produced by SubconsciousAI dream cycle
"consolidation", # Result of SemanticConsolidation merge
"external_sync", # Fetched from external source (RSS, API, etc.)
"user_correction", # Explicit user override
"prediction", # Stored as a future prediction
}
# ------------------------------------------------------------------ #
# Lineage event #
# ------------------------------------------------------------------ #
@dataclass
class LineageEvent:
"""
A single step in a memory's transformation history.
Examples:
created – initial storage
accessed – retrieved by a query
consolidated – merged into or from a proto-memory cluster
verified – reliability confirmed externally
contradicted – flagged as contradicting another memory
updated – content or metadata modified
archived – moved to COLD tier
expired – TTL reached or evicted
"""
event: str
timestamp: str # ISO 8601
actor: Optional[str] = None # agent_id, "system", "user", etc.
source_memories: List[str] = field(default_factory=list) # for consolidation
outcome: Optional[bool] = None # for verification events
notes: Optional[str] = None
extra: Dict[str, Any] = field(default_factory=dict)
def to_dict(self) -> Dict[str, Any]:
d: Dict[str, Any] = {
"event": self.event,
"timestamp": self.timestamp,
}
if self.actor is not None:
d["actor"] = self.actor
if self.source_memories:
d["source_memories"] = self.source_memories
if self.outcome is not None:
d["outcome"] = self.outcome
if self.notes:
d["notes"] = self.notes
if self.extra:
d["extra"] = self.extra
return d
@classmethod
def from_dict(cls, d: Dict[str, Any]) -> "LineageEvent":
return cls(
event=d["event"],
timestamp=d["timestamp"],
actor=d.get("actor"),
source_memories=d.get("source_memories", []),
outcome=d.get("outcome"),
notes=d.get("notes"),
extra=d.get("extra", {}),
)
# ------------------------------------------------------------------ #
# Origin #
# ------------------------------------------------------------------ #
@dataclass
class ProvenanceOrigin:
"""Where/how a memory was first created."""
type: str # One of ORIGIN_TYPES
agent_id: Optional[str] = None
session_id: Optional[str] = None
source_url: Optional[str] = None # For external_sync
timestamp: str = field(
default_factory=lambda: datetime.now(timezone.utc).isoformat()
)
def to_dict(self) -> Dict[str, Any]:
d: Dict[str, Any] = {
"type": self.type,
"timestamp": self.timestamp,
}
if self.agent_id:
d["agent_id"] = self.agent_id
if self.session_id:
d["session_id"] = self.session_id
if self.source_url:
d["source_url"] = self.source_url
return d
@classmethod
def from_dict(cls, d: Dict[str, Any]) -> "ProvenanceOrigin":
return cls(
type=d.get("type", "observation"),
agent_id=d.get("agent_id"),
session_id=d.get("session_id"),
source_url=d.get("source_url"),
timestamp=d.get("timestamp", datetime.now(timezone.utc).isoformat()),
)
# ------------------------------------------------------------------ #
# ProvenanceRecord — the full provenance object on a MemoryNode #
# ------------------------------------------------------------------ #
@dataclass
class ProvenanceRecord:
"""
Full provenance object attached to a MemoryNode.
Designed to be serialized into node.metadata["provenance"] for
backward compatibility with existing storage layers.
"""
origin: ProvenanceOrigin
lineage: List[LineageEvent] = field(default_factory=list)
version: int = 1
confidence_source: str = "bayesian_ltp" # How the confidence score is derived
# ---- Factory methods ------------------------------------------ #
@classmethod
def new(
cls,
origin_type: str = "observation",
agent_id: Optional[str] = None,
session_id: Optional[str] = None,
source_url: Optional[str] = None,
actor: Optional[str] = None,
) -> "ProvenanceRecord":
"""Create a fresh ProvenanceRecord and log the 'created' event."""
now = datetime.now(timezone.utc).isoformat()
origin = ProvenanceOrigin(
type=origin_type if origin_type in ORIGIN_TYPES else "observation",
agent_id=agent_id,
session_id=session_id,
source_url=source_url,
timestamp=now,
)
record = cls(origin=origin)
record.add_event(
event="created",
actor=actor or agent_id or "system",
)
return record
# ---- Mutation ------------------------------------------------- #
def add_event(
self,
event: str,
actor: Optional[str] = None,
source_memories: Optional[List[str]] = None,
outcome: Optional[bool] = None,
notes: Optional[str] = None,
**extra: Any,
) -> "ProvenanceRecord":
"""Append a new lineage event and bump the version counter."""
evt = LineageEvent(
event=event,
timestamp=datetime.now(timezone.utc).isoformat(),
actor=actor,
source_memories=source_memories or [],
outcome=outcome,
notes=notes,
extra=extra,
)
self.lineage.append(evt)
self.version += 1
return self
def mark_consolidated(
self,
source_memory_ids: List[str],
actor: str = "consolidation_worker",
) -> "ProvenanceRecord":
"""Convenience wrapper for consolidation events."""
return self.add_event(
event="consolidated",
actor=actor,
source_memories=source_memory_ids,
)
def mark_verified(
self,
success: bool,
actor: str = "system",
notes: Optional[str] = None,
) -> "ProvenanceRecord":
"""Record a verification outcome."""
return self.add_event(
event="verified",
actor=actor,
outcome=success,
notes=notes,
)
def mark_contradicted(
self,
contradiction_group_id: str,
actor: str = "contradiction_detector",
) -> "ProvenanceRecord":
"""Flag this memory as contradicted."""
return self.add_event(
event="contradicted",
actor=actor,
contradiction_group_id=contradiction_group_id,
)
# ---- Serialization -------------------------------------------- #
def to_dict(self) -> Dict[str, Any]:
return {
"origin": self.origin.to_dict(),
"lineage": [e.to_dict() for e in self.lineage],
"version": self.version,
"confidence_source": self.confidence_source,
}
@classmethod
def from_dict(cls, d: Dict[str, Any]) -> "ProvenanceRecord":
return cls(
origin=ProvenanceOrigin.from_dict(d.get("origin", {"type": "observation"})),
lineage=[LineageEvent.from_dict(e) for e in d.get("lineage", [])],
version=d.get("version", 1),
confidence_source=d.get("confidence_source", "bayesian_ltp"),
)
# ---- Helpers -------------------------------------------------- #
@property
def created_at(self) -> str:
"""ISO timestamp of the creation event."""
for event in self.lineage:
if event.event == "created":
return event.timestamp
return self.origin.timestamp
@property
def last_event(self) -> Optional[LineageEvent]:
"""Most recent lineage event."""
return self.lineage[-1] if self.lineage else None
def is_contradicted(self) -> bool:
return any(e.event == "contradicted" for e in self.lineage)
def is_verified(self) -> bool:
return any(
e.event == "verified" and e.outcome is True for e in self.lineage
)
def __repr__(self) -> str:
return (
f"ProvenanceRecord(origin_type={self.origin.type!r}, "
f"version={self.version}, events={len(self.lineage)})"
)
|