orbit-command-deck / orbit_core.py
xxbanditxx's picture
Update orbit_core.py
9297b1b verified
Raw
History Blame Contribute Delete
18.5 kB
from __future__ import annotations
import hashlib
import json
import math
import re
import threading
from dataclasses import asdict, dataclass, field
from datetime import datetime, timezone
from pathlib import Path
from typing import Dict, List, Optional, Tuple
CLAIM_TYPES = (
"world_claim",
"personal_report",
"opinion",
"preference",
"emotion",
"intention",
"prediction",
"hypothesis",
"inference",
"instruction",
"fiction",
)
RELATIONS = ("support", "contradict")
SOURCE_TYPES = (
"direct_measurement",
"primary_document",
"firsthand_report",
"secondary_source",
"model_output",
"conversation",
"unknown",
)
DEFAULT_RELIABILITY = {
"direct_measurement": 0.95,
"primary_document": 0.85,
"firsthand_report": 0.70,
"secondary_source": 0.60,
"model_output": 0.45,
"conversation": 0.45,
"unknown": 0.35,
}
def utc_now() -> str:
return datetime.now(timezone.utc).isoformat(timespec="seconds")
def clamp(value: float, lower: float = 0.0, upper: float = 1.0) -> float:
return max(lower, min(upper, value))
def normalize_text(value: str) -> str:
return re.sub(r"\s+", " ", value.strip().lower())
def stable_id(prefix: str, *parts: str) -> str:
digest = hashlib.sha256("\x1f".join(parts).encode("utf-8")).hexdigest()[:16]
return f"{prefix}_{digest}"
@dataclass
class Evidence:
id: str
relation: str
source_type: str
source_ref: str = ""
speaker: str = ""
quote: str = ""
note: str = ""
reliability: float = 0.5
observed_at: str = ""
submitted_at: str = field(default_factory=utc_now)
def validate(self) -> None:
if self.relation not in RELATIONS:
raise ValueError(f"relation must be one of: {', '.join(RELATIONS)}")
if self.source_type not in SOURCE_TYPES:
raise ValueError(f"source_type must be one of: {', '.join(SOURCE_TYPES)}")
self.reliability = round(clamp(float(self.reliability)), 3)
@property
def dedupe_key(self) -> str:
return normalize_text(
"|".join(
[
self.relation,
self.source_type,
self.source_ref,
self.speaker,
self.quote,
self.note,
self.observed_at,
]
)
)
@dataclass
class Belief:
id: str
subject: str
predicate: str
obj: str
context: str = ""
claim_type: str = "world_claim"
evidence: List[Evidence] = field(default_factory=list)
revision_triggers: List[str] = field(default_factory=list)
instrument_limits: List[str] = field(default_factory=list)
created_at: str = field(default_factory=utc_now)
updated_at: str = field(default_factory=utc_now)
def validate(self) -> None:
if not self.subject.strip() or not self.predicate.strip() or not self.obj.strip():
raise ValueError("subject, predicate, and object are required")
if self.claim_type not in CLAIM_TYPES:
raise ValueError(f"claim_type must be one of: {', '.join(CLAIM_TYPES)}")
for item in self.evidence:
item.validate()
@property
def statement(self) -> str:
return f"{self.subject} {self.predicate} {self.obj}".strip()
@property
def normalized_statement(self) -> str:
return normalize_text(self.statement)
@property
def support_weight(self) -> float:
return round(sum(item.reliability for item in self.evidence if item.relation == "support"), 3)
@property
def contradiction_weight(self) -> float:
return round(sum(item.reliability for item in self.evidence if item.relation == "contradict"), 3)
@property
def evidence_mass(self) -> float:
total = self.support_weight + self.contradiction_weight
return round(1.0 - math.exp(-total / 2.5), 3)
@property
def confidence(self) -> float:
support = self.support_weight
contradiction = self.contradiction_weight
total = support + contradiction
if total <= 0:
return 0.0
direction = support / total
return round(clamp(direction * self.evidence_mass), 3)
@property
def pressure(self) -> float:
support = self.support_weight
contradiction = self.contradiction_weight
total = support + contradiction
if total <= 0:
return 0.0
conflict = 2.0 * min(support, contradiction) / total
uncertainty = 1.0 - self.evidence_mass
return round(clamp((0.75 * conflict) + (0.25 * uncertainty)), 3)
@property
def status(self) -> str:
support = self.support_weight
contradiction = self.contradiction_weight
total = support + contradiction
if total == 0:
return "deferred"
if support > 0 and contradiction > 0 and self.pressure >= 0.30:
return "contested"
if contradiction > support and contradiction >= 0.70:
return "contradicted"
if self.confidence >= 0.65:
return "supported"
return "provisional"
@property
def evidence_count(self) -> int:
return len(self.evidence)
@property
def unique_source_refs(self) -> int:
refs = {item.source_ref.strip() for item in self.evidence if item.source_ref.strip()}
return len(refs)
@property
def unique_speakers(self) -> int:
speakers = {item.speaker.strip() for item in self.evidence if item.speaker.strip()}
return len(speakers)
@property
def unique_source_types(self) -> int:
return len({item.source_type for item in self.evidence})
@property
def source_diversity(self) -> float:
score = (
min(self.unique_source_refs, 5) * 0.45
+ min(self.unique_speakers, 5) * 0.35
+ min(self.unique_source_types, 5) * 0.20
) / 5.0
return round(clamp(score), 3)
@property
def risk_flags(self) -> List[str]:
flags: List[str] = []
if not self.revision_triggers:
flags.append("missing_revision_trigger")
if not self.instrument_limits:
flags.append("missing_instrument_limit")
if self.confidence >= 0.70 and self.evidence_count <= 1:
flags.append("high_confidence_sparse_evidence")
if self.status in {"contested", "contradicted"}:
flags.append("under_pressure")
if self.source_diversity <= 0.20 and self.evidence_count >= 3:
flags.append("low_source_diversity")
return flags
def add_unique(self, field_name: str, value: str) -> None:
value = value.strip()
if not value:
return
target = getattr(self, field_name)
if value not in target:
target.append(value)
def has_duplicate_evidence(self, candidate: Evidence) -> bool:
candidate_key = candidate.dedupe_key
return any(item.dedupe_key == candidate_key for item in self.evidence)
def summary(self) -> dict:
return {
"id": self.id,
"statement": self.statement,
"context": self.context,
"claim_type": self.claim_type,
"status": self.status,
"support_weight": self.support_weight,
"contradiction_weight": self.contradiction_weight,
"confidence": self.confidence,
"pressure": self.pressure,
"evidence_count": self.evidence_count,
"source_diversity": self.source_diversity,
"revision_triggers": self.revision_triggers,
"instrument_limits": self.instrument_limits,
"risk_flags": self.risk_flags,
"updated_at": self.updated_at,
}
class OrbitStore:
SCHEMA_VERSION = 2
def __init__(self, path: Path):
self.path = Path(path)
self.path.parent.mkdir(parents=True, exist_ok=True)
self._lock = threading.RLock()
self.beliefs: Dict[str, Belief] = {}
self.load()
@staticmethod
def belief_id(subject: str, predicate: str, obj: str, context: str = "") -> str:
return stable_id(
"belief",
normalize_text(subject),
normalize_text(predicate),
normalize_text(obj),
normalize_text(context),
)
def load(self) -> None:
with self._lock:
if not self.path.exists():
self.beliefs = {}
return
payload = json.loads(self.path.read_text(encoding="utf-8"))
version = int(payload.get("schema_version", 1))
if version != self.SCHEMA_VERSION:
raise ValueError(
f"Unsupported Orbit data schema {version}; expected {self.SCHEMA_VERSION}."
)
loaded: Dict[str, Belief] = {}
for raw_item in payload.get("beliefs", []):
raw = dict(raw_item)
evidence = [Evidence(**item) for item in raw.pop("evidence", [])]
belief = Belief(evidence=evidence, **raw)
belief.validate()
loaded[belief.id] = belief
self.beliefs = loaded
def save(self) -> None:
with self._lock:
payload = {
"schema_version": self.SCHEMA_VERSION,
"saved_at": utc_now(),
"beliefs": [asdict(item) for item in self.beliefs.values()],
}
temp = self.path.with_suffix(self.path.suffix + ".tmp")
temp.write_text(json.dumps(payload, indent=2, ensure_ascii=False), encoding="utf-8")
temp.replace(self.path)
def seed_if_empty(self) -> None:
if self.beliefs:
return
self.record_evidence(
subject="Orbit",
predicate="governs",
obj="how conclusions are formed and revised",
context="reasoning under uncertainty",
claim_type="world_claim",
relation="support",
source_type="primary_document",
source_ref="OPERATIONAL_SPEC.md",
speaker="ORBIT specification",
quote="ORBIT is a governor that constrains how conclusions are formed, held, revised, and audited.",
reliability=0.90,
note="Seeded from the project specification.",
revision_trigger="A later specification materially changes Orbit's role.",
instrument_limit="The specification defines intended behavior, not proven effectiveness.",
)
self.record_evidence(
subject="Contradictions",
predicate="should remain",
obj="visible until resolved",
context="Orbit belief handling",
claim_type="world_claim",
relation="support",
source_type="primary_document",
source_ref="README.md",
speaker="ORBIT specification",
quote="Contradictions remain visible instead of being silently discarded.",
reliability=0.85,
note="Seed belief.",
)
def record_evidence(
self,
*,
subject: str,
predicate: str,
obj: str,
context: str = "",
claim_type: str = "world_claim",
relation: str = "support",
source_type: str = "unknown",
source_ref: str = "",
speaker: str = "",
quote: str = "",
reliability: Optional[float] = None,
note: str = "",
observed_at: str = "",
revision_trigger: str = "",
instrument_limit: str = "",
allow_duplicate: bool = False,
) -> Belief:
subject = subject.strip()
predicate = predicate.strip()
obj = obj.strip()
context = context.strip()
belief_id = self.belief_id(subject, predicate, obj, context)
reliability_value = (
DEFAULT_RELIABILITY.get(source_type, 0.35)
if reliability is None
else float(reliability)
)
evidence = Evidence(
id=stable_id(
"evidence",
belief_id,
relation,
source_type,
source_ref.strip(),
speaker.strip(),
quote.strip(),
note.strip(),
observed_at.strip(),
utc_now(),
),
relation=relation,
source_type=source_type,
source_ref=source_ref.strip(),
speaker=speaker.strip(),
quote=quote.strip(),
note=note.strip(),
reliability=reliability_value,
observed_at=observed_at.strip(),
)
evidence.validate()
with self._lock:
belief = self.beliefs.get(belief_id)
if belief is None:
belief = Belief(
id=belief_id,
subject=subject,
predicate=predicate,
obj=obj,
context=context,
claim_type=claim_type,
)
self.beliefs[belief_id] = belief
elif belief.claim_type != claim_type and belief.claim_type == "world_claim":
belief.claim_type = claim_type
if not allow_duplicate and belief.has_duplicate_evidence(evidence):
belief.add_unique("revision_triggers", revision_trigger)
belief.add_unique("instrument_limits", instrument_limit)
belief.updated_at = utc_now()
belief.validate()
self.save()
return belief
belief.evidence.append(evidence)
belief.add_unique("revision_triggers", revision_trigger)
belief.add_unique("instrument_limits", instrument_limit)
belief.updated_at = utc_now()
belief.validate()
self.save()
return belief
def get(self, belief_id: str) -> Optional[Belief]:
return self.beliefs.get(belief_id)
def all(self) -> List[Belief]:
return sorted(
self.beliefs.values(),
key=lambda belief: (
belief.pressure,
belief.confidence,
belief.evidence_mass,
belief.updated_at,
),
reverse=True,
)
def recent(self, limit: int = 25) -> List[Belief]:
return sorted(
self.beliefs.values(),
key=lambda belief: belief.updated_at,
reverse=True,
)[:limit]
def search(self, query: str) -> List[Belief]:
query_norm = normalize_text(query)
tokens = [token for token in query_norm.split(" ") if token]
if not tokens:
return self.all()
scored: List[Tuple[float, Belief]] = []
for belief in self.beliefs.values():
statement = normalize_text(belief.statement)
context = normalize_text(belief.context)
claim_type = normalize_text(belief.claim_type)
revisions = normalize_text(" ".join(belief.revision_triggers))
limits = normalize_text(" ".join(belief.instrument_limits))
score = 0.0
if query_norm == statement:
score += 8.0
elif query_norm in statement:
score += 5.0
for token in tokens:
if token in statement:
score += 2.5
if token in context:
score += 1.5
if token in claim_type:
score += 0.5
if token in revisions:
score += 0.5
if token in limits:
score += 0.5
if score > 0:
score += belief.confidence * 1.5
score += belief.evidence_mass * 1.0
score += belief.source_diversity * 0.75
scored.append((score, belief))
scored.sort(
key=lambda pair: (
pair[0],
pair[1].confidence,
pair[1].evidence_mass,
pair[1].pressure,
),
reverse=True,
)
return [belief for _, belief in scored]
def pressure_queue(self) -> List[Belief]:
return [
belief
for belief in self.all()
if belief.status in {"contested", "contradicted", "provisional"}
]
def summaries(self, limit: int = 100) -> List[dict]:
return [belief.summary() for belief in self.all()[:limit]]
def recent_summaries(self, limit: int = 25) -> List[dict]:
return [belief.summary() for belief in self.recent(limit)]
def export_snapshot(self) -> dict:
return {
"schema_version": self.SCHEMA_VERSION,
"exported_at": utc_now(),
"beliefs": [asdict(item) for item in self.all()],
}
def required_confidence(stakes: str, reversibility: str, time_pressure: str) -> float:
stakes_base = {"low": 0.30, "medium": 0.60, "high": 0.85}
reversibility_adjustment = {"high": -0.15, "medium": 0.0, "low": 0.15}
time_adjustment = {"high": -0.15, "medium": 0.0, "low": 0.10}
try:
threshold = (
stakes_base[stakes.lower()]
+ reversibility_adjustment[reversibility.lower()]
+ time_adjustment[time_pressure.lower()]
)
except KeyError as exc:
raise ValueError("stakes, reversibility, and time pressure must be low, medium, or high") from exc
threshold = clamp(threshold)
if stakes.lower() == "high" and reversibility.lower() == "low":
threshold = max(threshold, 0.85)
return round(threshold, 2)
def decision_gate(
confidence: float,
stakes: str,
reversibility: str,
time_pressure: str,
) -> dict:
confidence = round(clamp(float(confidence)), 3)
threshold = required_confidence(stakes, reversibility, time_pressure)
permitted = confidence >= threshold
return {
"confidence": confidence,
"required_confidence": threshold,
"permitted": permitted,
"recommendation": (
"bounded action permitted"
if permitted
else "prefer reversible probing or gather more signal"
),
}