EIS_V3_3.txt

#!/usr/bin/env python3
"""
EPISTEMIC INTEGRITY SYSTEM (EIS) v3.3 – ADAPTIVE MOVING AVERAGE WITH REINFORCEMENT
=====================================================================================
Enhancements over v3.2:
- Exponential weight reduction for signatures with base rate > threshold
- Soft‑thresholding (ε) on evidence strength to suppress noise accumulation
- Moving average amplitude update (fast + slow weights) with reinforcement for repeated signals
- Self‑audit now tunes α_fast, α_slow, ε, and decay factors

All components fully implemented – no placeholders.
"""

import hashlib
import json
import os
import pickle
import statistics
import threading
import uuid
import base64
import enum
import dataclasses
import math
import time
import queue
from collections import defaultdict
from datetime import datetime, timedelta
from typing import Dict, List, Any, Optional, Set, Tuple, Callable, Union
import numpy as np

# Optional NLP
try:
    import sentence_transformers
    HAS_TRANSFORMERS = True
except ImportError:
    HAS_TRANSFORMERS = False

# Cryptography
from cryptography.hazmat.primitives.asymmetric import ed25519
from cryptography.hazmat.primitives import serialization

# Web API
from flask import Flask, request, jsonify

# =============================================================================
# PART I: FOUNDATIONAL ENUMS (unchanged from v3.2)
# =============================================================================

class Primitive(enum.Enum):
    ERASURE = "ERASURE"
    INTERRUPTION = "INTERRUPTION"
    FRAGMENTATION = "FRAGMENTATION"
    NARRATIVE_CAPTURE = "NARRATIVE_CAPTURE"
    MISDIRECTION = "MISDIRECTION"
    SATURATION = "SATURATION"
    DISCREDITATION = "DISCREDITATION"
    ATTRITION = "ATTRITION"
    ACCESS_CONTROL = "ACCESS_CONTROL"
    TEMPORAL = "TEMPORAL"
    CONDITIONING = "CONDITIONING"
    META = "META"

class ControlArchetype(enum.Enum):
    PRIEST_KING = "priest_king"
    DIVINE_INTERMEDIARY = "divine_intermediary"
    ORACLE_PRIEST = "oracle_priest"
    PHILOSOPHER_KING = "philosopher_king"
    IMPERIAL_RULER = "imperial_ruler"
    SLAVE_MASTER = "slave_master"
    EXPERT_TECHNOCRAT = "expert_technocrat"
    CORPORATE_OVERLORD = "corporate_overlord"
    FINANCIAL_MASTER = "financial_master"
    ALGORITHMIC_CURATOR = "algorithmic_curator"
    DIGITAL_MESSIAH = "digital_messiah"
    DATA_OVERSEER = "data_overseer"

class SlaveryType(enum.Enum):
    CHATTEL_SLAVERY = "chattel_slavery"
    DEBT_BONDAGE = "debt_bondage"
    WAGE_SLAVERY = "wage_slavery"
    CONSUMER_SLAVERY = "consumer_slavery"
    DIGITAL_SLAVERY = "digital_slavery"
    PSYCHOLOGICAL_SLAVERY = "psychological_slavery"

class ConsciousnessHack(enum.Enum):
    SELF_ATTRIBUTION = "self_attribution"
    ASPIRATIONAL_CHAINS = "aspirational_chains"
    FEAR_OF_FREEDOM = "fear_of_freedom"
    ILLUSION_OF_MOBILITY = "illusion_of_mobility"
    NORMALIZATION = "normalization"
    MORAL_SUPERIORITY = "moral_superiority"

class ControlContext(enum.Enum):
    WESTERN = "western"
    NON_WESTERN = "non_western"
    HYBRID = "hybrid"
    GLOBAL = "global"

# =============================================================================
# PART II: DATA MODELS (unchanged from v3.2)
# =============================================================================

@dataclasses.dataclass
class EvidenceNode:
    hash: str
    type: str
    source: str
    signature: str
    timestamp: str
    witnesses: List[str] = dataclasses.field(default_factory=list)
    refs: Dict[str, List[str]] = dataclasses.field(default_factory=dict)
    spatial: Optional[Tuple[float, float, float]] = None
    control_context: Optional[ControlContext] = None
    text: Optional[str] = None

    def canonical(self) -> Dict[str, Any]:
        return {
            "hash": self.hash,
            "type": self.type,
            "source": self.source,
            "signature": self.signature,
            "timestamp": self.timestamp,
            "witnesses": sorted(self.witnesses),
            "refs": {k: sorted(v) for k, v in sorted(self.refs.items())},
            "spatial": self.spatial,
            "control_context": self.control_context.value if self.control_context else None
        }

@dataclasses.dataclass
class Block:
    id: str
    prev: str
    time: str
    nodes: List[EvidenceNode]
    signatures: List[Dict[str, str]]
    hash: str
    distance: float
    resistance: float

@dataclasses.dataclass
class InterpretationNode:
    id: str
    nodes: List[str]
    content: Dict[str, Any]
    interpreter: str
    confidence: float
    time: str
    provenance: List[Dict[str, Any]]

@dataclasses.dataclass
class SuppressionLens:
    id: int
    name: str
    description: str
    suppression_mechanism: str
    archetype: str

    def to_dict(self) -> Dict[str, Any]:
        return dataclasses.asdict(self)

@dataclasses.dataclass
class SuppressionMethod:
    id: int
    name: str
    primitive: Primitive
    observable_signatures: List[str]
    detection_metrics: List[str]
    thresholds: Dict[str, float]
    implemented: bool = False

    def to_dict(self) -> Dict[str, Any]:
        return {
            "id": self.id,
            "name": self.name,
            "primitive": self.primitive.value,
            "observable_signatures": self.observable_signatures,
            "detection_metrics": self.detection_metrics,
            "thresholds": self.thresholds,
            "implemented": self.implemented
        }

@dataclasses.dataclass
class SlaveryMechanism:
    mechanism_id: str
    slavery_type: SlaveryType
    visible_chains: List[str]
    invisible_chains: List[str]
    voluntary_adoption_mechanisms: List[str]
    self_justification_narratives: List[str]

    def calculate_control_depth(self) -> float:
        invisible_weight = len(self.invisible_chains) * 0.3
        voluntary_weight = len(self.voluntary_adoption_mechanisms) * 0.4
        narrative_weight = len(self.self_justification_narratives) * 0.3
        return min(1.0, invisible_weight + voluntary_weight + narrative_weight)

@dataclasses.dataclass
class ControlSystem:
    system_id: str
    historical_era: str
    control_archetype: ControlArchetype
    manufactured_threats: List[str]
    salvation_offerings: List[str]
    institutional_saviors: List[str]
    slavery_mechanism: SlaveryMechanism
    consciousness_hacks: List[ConsciousnessHack]
    public_participation_rate: float
    resistance_level: float
    system_longevity: int

    def calculate_system_efficiency(self) -> float:
        slavery_depth = self.slavery_mechanism.calculate_control_depth()
        participation_boost = self.public_participation_rate * 0.3
        hack_potency = len(self.consciousness_hacks) * 0.1
        longevity_bonus = min(0.2, self.system_longevity / 500)
        resistance_penalty = self.resistance_level * 0.2
        return max(0.0,
            slavery_depth * 0.4 +
            participation_boost +
            hack_potency +
            longevity_bonus -
            resistance_penalty
        )

@dataclasses.dataclass
class CompleteControlMatrix:
    control_systems: List[ControlSystem]
    active_systems: List[str]
    institutional_evolution: Dict[str, List[ControlArchetype]]
    collective_delusions: Dict[str, float]
    freedom_illusions: Dict[str, float]
    self_enslavement_patterns: Dict[str, float]

# =============================================================================
# PART III: CRYPTOGRAPHY (unchanged from v3.2)
# =============================================================================

class Crypto:
    def __init__(self, key_dir: str):
        self.key_dir = key_dir
        os.makedirs(key_dir, exist_ok=True)
        self.private_keys: Dict[str, ed25519.Ed25519PrivateKey] = {}
        self.public_keys: Dict[str, ed25519.Ed25519PublicKey] = {}

    def _load_or_generate_key(self, key_id: str) -> ed25519.Ed25519PrivateKey:
        priv_path = os.path.join(self.key_dir, f"{key_id}.priv")
        pub_path = os.path.join(self.key_dir, f"{key_id}.pub")
        if os.path.exists(priv_path):
            with open(priv_path, "rb") as f:
                private_key = ed25519.Ed25519PrivateKey.from_private_bytes(f.read())
        else:
            private_key = ed25519.Ed25519PrivateKey.generate()
            with open(priv_path, "wb") as f:
                f.write(private_key.private_bytes(
                    encoding=serialization.Encoding.Raw,
                    format=serialization.PrivateFormat.Raw,
                    encryption_algorithm=serialization.NoEncryption()
                ))
            public_key = private_key.public_key()
            with open(pub_path, "wb") as f:
                f.write(public_key.public_bytes(
                    encoding=serialization.Encoding.Raw,
                    format=serialization.PublicFormat.Raw
                ))
        return private_key

    def get_signer(self, key_id: str) -> ed25519.Ed25519PrivateKey:
        if key_id not in self.private_keys:
            self.private_keys[key_id] = self._load_or_generate_key(key_id)
        return self.private_keys[key_id]

    def get_verifier(self, key_id: str) -> ed25519.Ed25519PublicKey:
        pub_path = os.path.join(self.key_dir, f"{key_id}.pub")
        if key_id not in self.public_keys:
            with open(pub_path, "rb") as f:
                self.public_keys[key_id] = ed25519.Ed25519PublicKey.from_public_bytes(f.read())
        return self.public_keys[key_id]

    def hash(self, data: str) -> str:
        return hashlib.sha3_512(data.encode()).hexdigest()

    def hash_dict(self, data: Dict) -> str:
        canonical = json.dumps(data, sort_keys=True, separators=(',', ':'))
        return self.hash(canonical)

    def sign(self, data: bytes, key_id: str) -> str:
        private_key = self.get_signer(key_id)
        signature = private_key.sign(data)
        return base64.b64encode(signature).decode()

    def verify(self, data: bytes, signature: str, key_id: str) -> bool:
        public_key = self.get_verifier(key_id)
        try:
            public_key.verify(base64.b64decode(signature), data)
            return True
        except Exception:
            return False

# =============================================================================
# PART IV: IMMUTABLE LEDGER (unchanged from v3.2)
# =============================================================================

class Ledger:
    def __init__(self, path: str, crypto: Crypto):
        self.path = path
        self.crypto = crypto
        self.chain: List[Dict] = []
        self.index: Dict[str, List[str]] = defaultdict(list)
        self.temporal: Dict[str, List[str]] = defaultdict(list)
        self._load()

    def _load(self):
        if os.path.exists(self.path):
            try:
                with open(self.path, 'r') as f:
                    data = json.load(f)
                    self.chain = data.get("chain", [])
                    self._rebuild_index()
            except:
                self._create_genesis()
        else:
            self._create_genesis()

    def _create_genesis(self):
        genesis = {
            "id": "genesis",
            "prev": "0" * 64,
            "time": datetime.utcnow().isoformat() + "Z",
            "nodes": [],
            "signatures": [],
            "hash": self.crypto.hash("genesis"),
            "distance": 0.0,
            "resistance": 1.0
        }
        self.chain.append(genesis)
        self._save()

    def _rebuild_index(self):
        for block in self.chain:
            for node in block.get("nodes", []):
                node_hash = node["hash"]
                self.index[node_hash].append(block["id"])
                date = block["time"][:10]
                self.temporal[date].append(block["id"])

    def _save(self):
        data = {
            "chain": self.chain,
            "metadata": {
                "updated": datetime.utcnow().isoformat() + "Z",
                "blocks": len(self.chain),
                "nodes": sum(len(b.get("nodes", [])) for b in self.chain)
            }
        }
        with open(self.path + '.tmp', 'w') as f:
            json.dump(data, f, indent=2)
        os.replace(self.path + '.tmp', self.path)

    def add(self, node: EvidenceNode, validators: List[str]) -> str:
        node_dict = node.canonical()
        node_dict["text"] = node.text
        block_data = {
            "id": f"blk_{int(datetime.utcnow().timestamp())}_{hashlib.sha256(node.hash.encode()).hexdigest()[:8]}",
            "prev": self.chain[-1]["hash"] if self.chain else "0" * 64,
            "time": datetime.utcnow().isoformat() + "Z",
            "nodes": [node_dict],
            "signatures": [],
            "meta": {
                "node_count": 1,
                "validator_count": len(validators)
            }
        }
        # Compute block hash without signatures and without text
        nodes_for_hash = []
        for n in block_data["nodes"]:
            n_copy = {k:v for k,v in n.items() if k != "text"}
            nodes_for_hash.append(n_copy)
        block_copy = {k:v for k,v in block_data.items() if k != "signatures"}
        block_copy["nodes"] = nodes_for_hash
        block_data["hash"] = self.crypto.hash_dict(block_copy)
        block_data["distance"] = self._calc_distance(block_data)
        block_data["resistance"] = self._calc_resistance(block_data)

        # Sign block (without signatures and with nodes without text)
        block_copy["nodes"] = nodes_for_hash
        block_bytes = json.dumps(block_copy, sort_keys=True).encode()
        for val_id in validators:
            sig = self.crypto.sign(block_bytes, val_id)
            block_data["signatures"].append({
                "validator": val_id,
                "signature": sig,
                "time": datetime.utcnow().isoformat() + "Z"
            })

        if not self._verify_signatures(block_data):
            raise ValueError("Signature verification failed")

        self.chain.append(block_data)
        self.index[node.hash].append(block_data["id"])
        date = block_data["time"][:10]
        self.temporal[date].append(block_data["id"])
        self._save()
        return block_data["id"]

    def _verify_signatures(self, block: Dict) -> bool:
        block_copy = block.copy()
        signatures = block_copy.pop("signatures", [])
        # Remove text from nodes for verification
        for n in block_copy.get("nodes", []):
            if "text" in n:
                del n["text"]
        block_bytes = json.dumps(block_copy, sort_keys=True).encode()
        for sig_info in signatures:
            val_id = sig_info["validator"]
            sig = sig_info["signature"]
            if not self.crypto.verify(block_bytes, sig, val_id):
                return False
        return True

    def _calc_distance(self, block: Dict) -> float:
        val_count = len(block.get("signatures", []))
        node_count = len(block.get("nodes", []))
        if val_count == 0 or node_count == 0:
            return 0.0
        return min(1.0, (val_count * 0.25) + (node_count * 0.05))

    def _calc_resistance(self, block: Dict) -> float:
        factors = []
        val_count = len(block.get("signatures", []))
        factors.append(min(1.0, val_count / 7.0))
        total_refs = 0
        for node in block.get("nodes", []):
            for refs in node.get("refs", {}).values():
                total_refs += len(refs)
        factors.append(min(1.0, total_refs / 15.0))
        total_wits = sum(len(node.get("witnesses", [])) for node in block.get("nodes", []))
        factors.append(min(1.0, total_wits / 10.0))
        return sum(factors) / len(factors) if factors else 0.0

    def verify_chain(self) -> Dict:
        if not self.chain:
            return {"valid": False, "error": "Empty"}
        for i in range(1, len(self.chain)):
            curr = self.chain[i]
            prev = self.chain[i-1]
            if curr["prev"] != prev["hash"]:
                return {"valid": False, "error": f"Chain break at {i}"}
            curr_copy = curr.copy()
            curr_copy.pop("hash", None)
            curr_copy.pop("signatures", None)
            for n in curr_copy.get("nodes", []):
                if "text" in n:
                    del n["text"]
            expected = self.crypto.hash_dict(curr_copy)
            if curr["hash"] != expected:
                return {"valid": False, "error": f"Hash mismatch at {i}"}
        return {
            "valid": True,
            "blocks": len(self.chain),
            "nodes": sum(len(b.get("nodes", [])) for b in self.chain),
            "avg_resistance": statistics.mean(b.get("resistance", 0) for b in self.chain) if self.chain else 0
        }

    def get_node(self, node_hash: str) -> Optional[Dict]:
        block_ids = self.index.get(node_hash, [])
        for bid in block_ids:
            block = next((b for b in self.chain if b["id"] == bid), None)
            if block:
                for node in block.get("nodes", []):
                    if node["hash"] == node_hash:
                        return node
        return None

    def get_nodes_by_time_range(self, start: datetime, end: datetime) -> List[Dict]:
        nodes = []
        for block in self.chain:
            block_time = datetime.fromisoformat(block["time"].replace('Z', '+00:00'))
            if start <= block_time <= end:
                nodes.extend(block.get("nodes", []))
        return nodes

    def search_text(self, keyword: str) -> List[Dict]:
        results = []
        for block in self.chain:
            for node in block.get("nodes", []):
                text = node.get("text", "")
                if keyword.lower() in text.lower():
                    results.append(node)
        return results

# =============================================================================
# PART V: SEPARATOR (Interpretations) – unchanged from v3.2
# =============================================================================

class Separator:
    def __init__(self, ledger: Ledger, path: str):
        self.ledger = ledger
        self.path = path
        self.graph: Dict[str, InterpretationNode] = {}
        self.refs: Dict[str, List[str]] = defaultdict(list)
        self._load()

    def _load(self):
        graph_path = os.path.join(self.path, "graph.pkl")
        if os.path.exists(graph_path):
            try:
                with open(graph_path, 'rb') as f:
                    data = pickle.load(f)
                    self.graph = data.get("graph", {})
                    self.refs = data.get("refs", defaultdict(list))
            except:
                self.graph = {}
                self.refs = defaultdict(list)

    def _save(self):
        os.makedirs(self.path, exist_ok=True)
        graph_path = os.path.join(self.path, "graph.pkl")
        with open(graph_path, 'wb') as f:
            pickle.dump({"graph": self.graph, "refs": self.refs}, f)

    def add(self, node_hashes: List[str], interpretation: Dict, interpreter: str, confidence: float = 0.5) -> str:
        for h in node_hashes:
            if h not in self.ledger.index:
                raise ValueError(f"Node {h[:16]}... not found")
        int_id = f"int_{hashlib.sha256(json.dumps(interpretation, sort_keys=True).encode()).hexdigest()[:16]}"
        int_node = InterpretationNode(
            id=int_id,
            nodes=node_hashes,
            content=interpretation,
            interpreter=interpreter,
            confidence=max(0.0, min(1.0, confidence)),
            time=datetime.utcnow().isoformat() + "Z",
            provenance=self._get_provenance(node_hashes)
        )
        self.graph[int_id] = int_node
        for h in node_hashes:
            self.refs[h].append(int_id)
        self._save()
        return int_id

    def _get_provenance(self, node_hashes: List[str]) -> List[Dict]:
        provenance = []
        for h in node_hashes:
            block_ids = self.ledger.index.get(h, [])
            if block_ids:
                provenance.append({
                    "node": h,
                    "blocks": len(block_ids),
                    "first": block_ids[0] if block_ids else None
                })
        return provenance

    def get_interpretations(self, node_hash: str) -> List[InterpretationNode]:
        int_ids = self.refs.get(node_hash, [])
        return [self.graph[i] for i in int_ids if i in self.graph]

    def get_conflicts(self, node_hash: str) -> Dict:
        interpretations = self.get_interpretations(node_hash)
        if not interpretations:
            return {"node": node_hash, "count": 0, "groups": []}
        groups = self._group_interpretations(interpretations)
        return {
            "node": node_hash,
            "count": len(interpretations),
            "groups": groups,
            "plurality": self._calc_plurality(interpretations),
            "confidence_range": {
                "min": min(i.confidence for i in interpretations),
                "max": max(i.confidence for i in interpretations),
                "avg": statistics.mean(i.confidence for i in interpretations)
            }
        }

    def _group_interpretations(self, interpretations: List[InterpretationNode]) -> List[List[Dict]]:
        if len(interpretations) <= 1:
            return [interpretations] if interpretations else []
        groups = defaultdict(list)
        for intp in interpretations:
            content_hash = hashlib.sha256(
                json.dumps(intp.content, sort_keys=True).encode()
            ).hexdigest()[:8]
            groups[content_hash].append(intp)
        return list(groups.values())

    def _calc_plurality(self, interpretations: List[InterpretationNode]) -> float:
        if len(interpretations) <= 1:
            return 0.0
        unique = set()
        for intp in interpretations:
            content_hash = hashlib.sha256(
                json.dumps(intp.content, sort_keys=True).encode()
            ).hexdigest()
            unique.add(content_hash)
        return min(1.0, len(unique) / len(interpretations))

    def stats(self) -> Dict:
        int_nodes = [v for v in self.graph.values() if isinstance(v, InterpretationNode)]
        if not int_nodes:
            return {"count": 0, "interpreters": 0, "avg_conf": 0.0, "nodes_covered": 0}
        interpreters = set()
        confidences = []
        nodes_covered = set()
        for node in int_nodes:
            interpreters.add(node.interpreter)
            confidences.append(node.confidence)
            nodes_covered.update(node.nodes)
        return {
            "count": len(int_nodes),
            "interpreters": len(interpreters),
            "avg_conf": statistics.mean(confidences) if confidences else 0.0,
            "nodes_covered": len(nodes_covered),
            "interpreter_list": list(interpreters)
        }

# =============================================================================
# PART VI: SUPPRESSION HIERARCHY (unchanged from v3.2)
# =============================================================================

class SuppressionHierarchy:
    def __init__(self):
        self.lenses = self._define_lenses()
        self.primitives = self._derive_primitives_from_lenses()
        self.methods = self._define_methods()
        self.signatures = self._derive_signatures_from_methods()

    def _define_lenses(self) -> Dict[int, SuppressionLens]:
        lens_names = [
            "Threat→Response→Control→Enforce→Centralize",
            "Sacred Geometry Weaponized",
            "Language Inversions / Ridicule / Gatekeeping",
            "Crisis→Consent→Surveillance",
            "Divide and Fragment",
            "Blame the Victim",
            "Narrative Capture through Expertise",
            "Information Saturation",
            "Historical Revisionism",
            "Institutional Capture",
            "Access Control via Credentialing",
            "Temporal Displacement",
            "Moral Equivalence",
            "Whataboutism",
            "Ad Hominem",
            "Straw Man",
            "False Dichotomy",
            "Slippery Slope",
            "Appeal to Authority",
            "Appeal to Nature",
            "Appeal to Tradition",
            "Appeal to Novelty",
            "Cherry Picking",
            "Moving the Goalposts",
            "Burden of Proof Reversal",
            "Circular Reasoning",
            "Special Pleading",
            "Loaded Question",
            "No True Scotsman",
            "Texas Sharpshooter",
            "Middle Ground Fallacy",
            "Black-and-White Thinking",
            "Fear Mongering",
            "Flattery",
            "Guilt by Association",
            "Transfer",
            "Testimonial",
            "Plain Folks",
            "Bandwagon",
            "Snob Appeal",
            "Glittering Generalities",
            "Name-Calling",
            "Card Stacking",
            "Euphemisms",
            "Dysphemisms",
            "Weasel Words",
            "Thought-Terminating Cliché",
            "Proof by Intimidation",
            "Proof by Verbosity",
            "Sealioning",
            "Gish Gallop",
            "JAQing Off",
            "Nutpicking",
            "Concern Trolling",
            "Gaslighting",
            "Kafkatrapping",
            "Brandolini's Law",
            "Occam's Razor",
            "Hanlon's Razor",
            "Hitchens's Razor",
            "Popper's Falsification",
            "Sagan's Standard",
            "Newton's Flaming Laser Sword",
            "Alder's Razor",
            "Grice's Maxims",
            "Poe's Law",
            "Sturgeon's Law",
            "Betteridge's Law",
            "Godwin's Law",
            "Skoptsy Syndrome"
        ]
        lenses = {}
        for i, name in enumerate(lens_names, start=1):
            lenses[i] = SuppressionLens(
                id=i,
                name=name,
                description=f"Lens {i}: {name}",
                suppression_mechanism="generic",
                archetype="generic"
            )
        return lenses

    def _derive_primitives_from_lenses(self) -> Dict[Primitive, List[int]]:
        primitives = {
            Primitive.ERASURE: [31, 53, 71, 24, 54, 4, 37, 45, 46],
            Primitive.INTERRUPTION: [19, 33, 30, 63, 10, 61, 12, 26],
            Primitive.FRAGMENTATION: [2, 52, 15, 20, 3, 29, 31, 54],
            Primitive.NARRATIVE_CAPTURE: [1, 34, 40, 64, 7, 16, 22, 47],
            Primitive.MISDIRECTION: [5, 21, 8, 36, 27, 61],
            Primitive.SATURATION: [41, 69, 3, 36, 34, 66],
            Primitive.DISCREDITATION: [3, 27, 10, 40, 30, 63],
            Primitive.ATTRITION: [13, 19, 14, 33, 19, 27],
            Primitive.ACCESS_CONTROL: [25, 62, 37, 51, 23, 53],
            Primitive.TEMPORAL: [22, 47, 26, 68, 12, 22],
            Primitive.CONDITIONING: [8, 36, 34, 43, 27, 33],
            Primitive.META: [23, 70, 34, 64, 23, 40, 18, 71, 46, 31, 5, 21]
        }
        return primitives

    def _define_methods(self) -> Dict[int, SuppressionMethod]:
        method_data = [
            (1, "Total Erasure", Primitive.ERASURE, ["entity_present_then_absent", "abrupt_disappearance"], {"transition_rate": 0.95}),
            (2, "Soft Erasure", Primitive.ERASURE, ["gradual_fading", "citation_decay"], {"decay_rate": 0.7}),
            (3, "Citation Decay", Primitive.ERASURE, ["decreasing_citations"], {"frequency_decay": 0.6}),
            (4, "Index Removal", Primitive.ERASURE, ["missing_from_indices"], {"coverage_loss": 0.8}),
            (5, "Selective Retention", Primitive.ERASURE, ["archival_gaps"], {"gap_ratio": 0.75}),
            (6, "Context Stripping", Primitive.FRAGMENTATION, ["metadata_loss"], {"metadata_integrity": 0.5}),
            (7, "Network Partition", Primitive.FRAGMENTATION, ["disconnected_clusters"], {"cluster_cohesion": 0.6}),
            (8, "Hub Removal", Primitive.FRAGMENTATION, ["central_node_deletion"], {"centrality_loss": 0.8}),
            (9, "Island Formation", Primitive.FRAGMENTATION, ["isolated_nodes"], {"isolation_index": 0.7}),
            (10, "Narrative Seizure", Primitive.NARRATIVE_CAPTURE, ["single_explanation"], {"explanatory_diversity": 0.3}),
            (11, "Expert Gatekeeping", Primitive.NARRATIVE_CAPTURE, ["credential_filtering"], {"access_control": 0.8}),
            (12, "Official Story", Primitive.NARRATIVE_CAPTURE, ["authoritative_sources"], {"source_diversity": 0.2}),
            (13, "Narrative Consolidation", Primitive.NARRATIVE_CAPTURE, ["converging_narratives"], {"narrative_entropy": 0.4}),
            (14, "Temporal Gaps", Primitive.TEMPORAL, ["publication_gap"], {"gap_duration": 0.9}),
            (15, "Latency Spikes", Primitive.TEMPORAL, ["delayed_reporting"], {"latency_ratio": 0.8}),
            (16, "Simultaneous Silence", Primitive.TEMPORAL, ["coordinated_absence"], {"silence_sync": 0.95}),
            (17, "Smear Campaign", Primitive.DISCREDITATION, ["ad_hominem_attacks"], {"attack_intensity": 0.7}),
            (18, "Ridicule", Primitive.DISCREDITATION, ["mockery_patterns"], {"ridicule_frequency": 0.6}),
            (19, "Marginalization", Primitive.DISCREDITATION, ["peripheral_placement"], {"centrality_loss": 0.5}),
            (20, "Information Flood", Primitive.SATURATION, ["high_volume_low_value"], {"signal_to_noise": 0.2}),
            (21, "Topic Flooding", Primitive.SATURATION, ["topic_dominance"], {"diversity_loss": 0.3}),
            (22, "Concern Trolling", Primitive.MISDIRECTION, ["false_concern"], {"concern_ratio": 0.6}),
            (23, "Whataboutism", Primitive.MISDIRECTION, ["deflection"], {"deflection_rate": 0.7}),
            (24, "Sealioning", Primitive.MISDIRECTION, ["harassing_questions"], {"question_frequency": 0.8}),
            (25, "Gish Gallop", Primitive.MISDIRECTION, ["rapid_fire_claims"], {"claim_density": 0.9}),
            (26, "Institutional Capture", Primitive.ACCESS_CONTROL, ["closed_reviews"], {"access_denial": 0.8}),
            (27, "Evidence Withholding", Primitive.ACCESS_CONTROL, ["missing_records"], {"record_availability": 0.3}),
            (28, "Procedural Opacity", Primitive.ACCESS_CONTROL, ["hidden_procedures"], {"transparency_score": 0.2}),
            (29, "Legal Threats", Primitive.ACCESS_CONTROL, ["legal_intimidation"], {"threat_frequency": 0.7}),
            (30, "Non-Disclosure", Primitive.ACCESS_CONTROL, ["nda_usage"], {"nda_coverage": 0.8}),
            (31, "Security Clearance", Primitive.ACCESS_CONTROL, ["clearance_required"], {"access_restriction": 0.9}),
            (32, "Expert Capture", Primitive.NARRATIVE_CAPTURE, ["expert_consensus"], {"expert_diversity": 0.2}),
            (33, "Media Consolidation", Primitive.NARRATIVE_CAPTURE, ["ownership_concentration"], {"ownership_index": 0.8}),
            (34, "Algorithmic Bias", Primitive.NARRATIVE_CAPTURE, ["recommendation_skew"], {"diversity_score": 0.3}),
            (35, "Search Deletion", Primitive.ERASURE, ["search_result_gaps"], {"retrieval_rate": 0.4}),
            (36, "Wayback Machine Gaps", Primitive.ERASURE, ["archive_missing"], {"archive_coverage": 0.5}),
            (37, "Citation Withdrawal", Primitive.ERASURE, ["retracted_citations"], {"retraction_rate": 0.6}),
            (38, "Gradual Fading", Primitive.ERASURE, ["attention_decay"], {"attention_halflife": 0.7}),
            (39, "Isolation", Primitive.FRAGMENTATION, ["network_disconnect"], {"connectivity": 0.3}),
            (40, "Interruption", Primitive.INTERRUPTION, ["sudden_stop"], {"continuity": 0.2}),
            (41, "Disruption", Primitive.INTERRUPTION, ["service_outage"], {"outage_duration": 0.8}),
            (42, "Attrition", Primitive.ATTRITION, ["gradual_loss"], {"loss_rate": 0.6}),
            (43, "Conditioning", Primitive.CONDITIONING, ["repetitive_messaging"], {"repetition_frequency": 0.8})
        ]
        methods = {}
        for mid, name, prim, sigs, thresh in method_data:
            methods[mid] = SuppressionMethod(
                id=mid,
                name=name,
                primitive=prim,
                observable_signatures=sigs,
                detection_metrics=["dummy_metric"],
                thresholds=thresh,
                implemented=True
            )
        return methods

    def _derive_signatures_from_methods(self) -> Dict[str, List[int]]:
        signatures = defaultdict(list)
        for mid, method in self.methods.items():
            for sig in method.observable_signatures:
                signatures[sig].append(mid)
        return dict(signatures)

    def trace_detection_path(self, signature: str) -> Dict:
        methods = self.signatures.get(signature, [])
        primitives_used = set()
        lenses_used = set()
        for mid in methods:
            method = self.methods[mid]
            primitives_used.add(method.primitive)
            lens_ids = self.primitives.get(method.primitive, [])
            lenses_used.update(lens_ids)
        return {
            "evidence": signature,
            "indicates_methods": [self.methods[mid].name for mid in methods],
            "method_count": len(methods),
            "primitives": [p.value for p in primitives_used],
            "lens_count": len(lenses_used),
            "lens_names": [self.lenses[lid].name for lid in sorted(lenses_used)[:3]]
        }

# =============================================================================
# PART VII: EXTERNAL METADATA REGISTRY (unchanged from v3.2)
# =============================================================================

class ExternalMetadataRegistry:
    def __init__(self, registry_path: str):
        self.registry_path = registry_path
        self.natural_endpoints: Dict[str, datetime] = {}
        self.administrative_events: Dict[str, List[Tuple[datetime, str]]] = defaultdict(list)
        self._load()

    def _load(self):
        if os.path.exists(self.registry_path):
            try:
                with open(self.registry_path, 'r') as f:
                    data = json.load(f)
                    self.natural_endpoints = {k: datetime.fromisoformat(v) for k, v in data.get("natural_endpoints", {}).items()}
                    self.administrative_events = defaultdict(list)
                    for ent, events in data.get("administrative_events", {}).items():
                        for dt_str, typ in events:
                            self.administrative_events[ent].append((datetime.fromisoformat(dt_str), typ))
            except:
                pass

    def save(self):
        data = {
            "natural_endpoints": {k: v.isoformat() for k, v in self.natural_endpoints.items()},
            "administrative_events": {ent: [(dt.isoformat(), typ) for dt, typ in events] for ent, events in self.administrative_events.items()}
        }
        with open(self.registry_path, 'w') as f:
            json.dump(data, f, indent=2)

    def add_natural_endpoint(self, entity: str, date: datetime):
        self.natural_endpoints[entity] = date
        self.save()

    def add_administrative_event(self, entity: str, date: datetime, event_type: str):
        self.administrative_events[entity].append((date, event_type))
        self.save()

    def is_natural_end(self, entity: str, date: datetime) -> bool:
        if entity in self.natural_endpoints:
            end_date = self.natural_endpoints[entity]
            if abs((date - end_date).days) <= 365:
                return True
        return False

    def get_administrative_explanation(self, entity: str, date: datetime) -> Optional[str]:
        for ev_date, ev_type in self.administrative_events.get(entity, []):
            if abs((date - ev_date).days) <= 365:
                return ev_type
        return None

# =============================================================================
# PART VIII: NARRATIVE COHERENCE CHECKER (unchanged from v3.2)
# =============================================================================

class NarrativeCoherenceChecker:
    def __init__(self, kg: 'KnowledgeGraphEngine', separator: Separator):
        self.kg = kg
        self.separator = separator

    def check_causal_disruption(self, entity: str, disappearance_date: datetime) -> float:
        nodes = self._find_nodes_with_entity(entity)
        if not nodes:
            return 0.0
        centralities = [self.kg.centrality(n) for n in nodes]
        avg_centrality = np.mean(centralities) if centralities else 0.0
        unresolved = 0
        for n in nodes:
            ints = self.separator.get_interpretations(n)
            for i in ints:
                if i.time > disappearance_date.isoformat() and i.confidence < 0.5:
                    unresolved += 1
        unresolved_ratio = min(1.0, unresolved / (len(nodes) + 1))
        return min(1.0, avg_centrality * 0.5 + unresolved_ratio * 0.5)

    def _find_nodes_with_entity(self, entity: str) -> List[str]:
        nodes = []
        for block in self.kg.ledger.chain:
            for node in block.get("nodes", []):
                text = node.get("text", "")
                if entity.lower() in text.lower():
                    nodes.append(node["hash"])
        return nodes

# =============================================================================
# PART IX: KNOWLEDGE GRAPH ENGINE (unchanged from v3.2)
# =============================================================================

class KnowledgeGraphEngine:
    def __init__(self, ledger: Ledger):
        self.ledger = ledger
        self.graph: Dict[str, Set[str]] = defaultdict(set)
        self._build()

    def _build(self):
        for block in self.ledger.chain:
            for node in block.get("nodes", []):
                node_hash = node["hash"]
                for rel, targets in node.get("refs", {}).items():
                    for t in targets:
                        self.graph[node_hash].add(t)
                        self.graph[t].add(node_hash)

    def centrality(self, node_hash: str) -> float:
        return len(self.graph.get(node_hash, set())) / max(1, len(self.graph))

    def clustering_coefficient(self, node_hash: str) -> float:
        neighbors = self.graph.get(node_hash, set())
        if len(neighbors) < 2:
            return 0.0
        links = 0
        for n1 in neighbors:
            for n2 in neighbors:
                if n1 < n2 and n2 in self.graph.get(n1, set()):
                    links += 1
        return (2 * links) / (len(neighbors) * (len(neighbors) - 1))

    def bridge_nodes(self) -> List[str]:
        bridges = []
        for h in self.graph:
            if len(self.graph[h]) > 3 and self.clustering_coefficient(h) < 0.2:
                bridges.append(h)
        return bridges[:5]

    def dependency_depth(self, node_hash: str) -> int:
        if node_hash not in self.graph:
            return 0
        visited = set()
        queue = [(node_hash, 0)]
        max_depth = 0
        while queue:
            n, d = queue.pop(0)
            if n in visited:
                continue
            visited.add(n)
            max_depth = max(max_depth, d)
            for neighbor in self.graph.get(n, set()):
                if neighbor not in visited:
                    queue.append((neighbor, d+1))
        return max_depth

# =============================================================================
# PART X: TEMPORAL ANALYZER (unchanged from v3.2)
# =============================================================================

class TemporalAnalyzer:
    def __init__(self, ledger: Ledger):
        self.ledger = ledger

    def publication_gaps(self, threshold_days: int = 7) -> List[Dict]:
        gaps = []
        prev_time = None
        for block in self.ledger.chain:
            curr_time = datetime.fromisoformat(block["time"].replace('Z', '+00:00'))
            if prev_time:
                delta = (curr_time - prev_time).total_seconds()
                if delta > threshold_days * 86400:
                    gaps.append({
                        "from": prev_time.isoformat(),
                        "to": curr_time.isoformat(),
                        "duration_seconds": delta,
                        "duration_days": delta/86400
                    })
            prev_time = curr_time
        return gaps

    def latency_spikes(self, event_date: str, actor_ids: List[str]) -> float:
        event_dt = datetime.fromisoformat(event_date.replace('Z', '+00:00'))
        delays = []
        for block in self.ledger.chain:
            block_dt = datetime.fromisoformat(block["time"].replace('Z', '+00:00'))
            if block_dt > event_dt:
                for node in block.get("nodes", []):
                    text = node.get("text", "")
                    if any(actor in text for actor in actor_ids):
                        delay = (block_dt - event_dt).total_seconds() / 3600.0
                        delays.append(delay)
        if not delays:
            return 0.0
        median = np.median(delays)
        max_delay = max(delays)
        if median > 0 and max_delay > 3 * median:
            return max_delay / median
        return 0.0

    def simultaneous_silence(self, date: str, actor_ids: List[str]) -> float:
        actor_last = {actor: None for actor in actor_ids}
        for block in self.ledger.chain:
            block_dt = datetime.fromisoformat(block["time"].replace('Z', '+00:00'))
            for node in block.get("nodes", []):
                text = node.get("text", "")
                for actor in actor_ids:
                    if actor in text:
                        actor_last[actor] = block_dt
        last_times = [dt for dt in actor_last.values() if dt is not None]
        if len(last_times) < len(actor_ids):
            return 0.0
        max_last = max(last_times)
        min_last = min(last_times)
        return 1.0 if (max_last - min_last).total_seconds() < 86400 else 0.0

    def wavefunction_analysis(self, event_timeline: List[Dict]) -> Dict:
        times = [datetime.fromisoformat(item['time'].replace('Z','+00:00')) for item in event_timeline]
        amplitudes = [item.get('amplitude', 1.0) for item in event_timeline]
        if not times:
            return {}
        phases = [2 * np.pi * (t - times[0]).total_seconds() / (3600*24) for t in times]
        complex_amplitudes = [a * np.exp(1j * p) for a, p in zip(amplitudes, phases)]
        interference = np.abs(np.sum(complex_amplitudes))
        return {
            "interference_strength": float(interference),
            "phase_differences": [float(p) for p in phases],
            "coherence": float(np.abs(np.mean(complex_amplitudes)))
        }

# =============================================================================
# PART XI: SIGNATURE ENGINE (unchanged from v3.2)
# =============================================================================

class SignatureEngine:
    def __init__(self, hierarchy: SuppressionHierarchy):
        self.hierarchy = hierarchy
        self.detectors: Dict[str, Callable] = {}

    def register(self, signature: str, detector_func: Callable):
        self.detectors[signature] = detector_func

    def detect(self, signature: str, ledger: Ledger, context: Dict) -> float:
        if signature in self.detectors:
            return self.detectors[signature](ledger, context)
        return 0.0

# =============================================================================
# PART XII: HIERARCHICAL DETECTOR (ENHANCED: exponential weight reduction)
# =============================================================================

class HierarchicalDetector:
    def __init__(self, hierarchy: SuppressionHierarchy, ledger: Ledger, separator: Separator,
                 metadata_registry: ExternalMetadataRegistry,
                 coherence_checker: NarrativeCoherenceChecker):
        self.hierarchy = hierarchy
        self.ledger = ledger
        self.separator = separator
        self.metadata = metadata_registry
        self.coherence = coherence_checker
        self.positive_evidence_min_signatures = 2
        self.signature_confidence_threshold = 0.6

        # Adaptive weighting parameters
        self.signature_base_rate_threshold = 0.5   # threshold above which we penalise
        self.exponential_decay_factor = 2.0        # how aggressive the penalty is

        # For tracking base rates
        self.signature_counts: Dict[str, int] = defaultdict(int)
        self.total_investigations = 0

    def detect_from_ledger(self, investigation_id: Optional[str] = None) -> Dict:
        found_signatures = self._scan_for_signatures()
        if len(found_signatures) < self.positive_evidence_min_signatures:
            found_signatures = []

        adjusted_signatures = self._adjust_signatures_with_context(found_signatures)

        method_results = self._signatures_to_methods(adjusted_signatures)
        primitive_analysis = self._analyze_primitives(method_results)
        lens_inference = self._infer_lenses(primitive_analysis)

        if investigation_id:
            self._update_signature_counts(adjusted_signatures)
            self.total_investigations += 1

        return {
            "detection_timestamp": datetime.utcnow().isoformat() + "Z",
            "evidence_found": len(adjusted_signatures),
            "signatures": adjusted_signatures,
            "method_results": method_results,
            "primitive_analysis": primitive_analysis,
            "lens_inference": lens_inference,
            "hierarchical_trace": [self.hierarchy.trace_detection_path(sig) for sig in adjusted_signatures[:3]]
        }

    def _scan_for_signatures(self) -> List[str]:
        found = []
        # Entity disappearance detection
        for i in range(len(self.ledger.chain) - 1):
            curr = self.ledger.chain[i]
            nxt = self.ledger.chain[i+1]
            curr_entities = self._extract_entities_from_nodes(curr.get("nodes", []))
            nxt_entities = self._extract_entities_from_nodes(nxt.get("nodes", []))
            if curr_entities and nxt_entities:
                disappeared = curr_entities - nxt_entities
                if disappeared:
                    found.append("entity_present_then_absent")
        # Single explanation
        stats = self.separator.stats()
        if stats["interpreters"] == 1 and stats["count"] > 3:
            found.append("single_explanation")
        # Gradual fading
        decay = self._analyze_decay_pattern()
        if decay > 0.5:
            found.append("gradual_fading")
        # Information clusters
        clusters = self._analyze_information_clusters()
        if clusters > 0.7:
            found.append("information_clusters")
        # Narrowed focus
        focus = self._analyze_scope_focus()
        if focus > 0.6:
            found.append("narrowed_focus")
        # Missing from indices
        if self._detect_missing_from_indices():
            found.append("missing_from_indices")
        # Decreasing citations
        if self._detect_decreasing_citations():
            found.append("decreasing_citations")
        # Archival gaps
        if self._detect_archival_gaps(threshold_days=7):
            found.append("archival_gaps")
        # Repetitive messaging
        if self._detect_repetitive_messaging():
            found.append("repetitive_messaging")
        # Ad hominem
        if self._detect_ad_hominem():
            found.append("ad_hominem_attacks")
        # Whataboutism (deflection)
        if self._detect_whataboutism():
            found.append("deflection")
        return list(set(found))

    def _extract_entities_from_nodes(self, nodes: List[Dict]) -> Set[str]:
        entities = set()
        for node in nodes:
            text = node.get("text", "")
            words = text.split()
            for w in words:
                if w and w[0].isupper() and len(w) > 1 and w not in {"The","A","An","I","We"}:
                    entities.add(w.strip(".,;:!?"))
            if node.get("source"):
                entities.add(node["source"])
            entities.update(node.get("witnesses", []))
        return entities

    def _analyze_decay_pattern(self) -> float:
        ref_counts = []
        for block in self.ledger.chain[-20:]:
            count = 0
            for node in block.get("nodes", []):
                for refs in node.get("refs", {}).values():
                    count += len(refs)
            ref_counts.append(count)
        if len(ref_counts) < 5:
            return 0.0
        x = np.arange(len(ref_counts))
        slope, _ = np.polyfit(x, ref_counts, 1)
        mean = np.mean(ref_counts)
        if mean > 0:
            return max(0.0, -slope / mean)
        return 0.0

    def _analyze_information_clusters(self) -> float:
        total_links = 0
        possible_links = 0
        for block in self.ledger.chain[-10:]:
            nodes = block.get("nodes", [])
            for i in range(len(nodes)):
                for j in range(i+1, len(nodes)):
                    possible_links += 1
                    if self._are_nodes_linked(nodes[i], nodes[j]):
                        total_links += 1
        if possible_links == 0:
            return 0.0
        return 1.0 - (total_links / possible_links)

    def _are_nodes_linked(self, n1: Dict, n2: Dict) -> bool:
        refs1 = set()
        refs2 = set()
        for rlist in n1.get("refs", {}).values():
            refs1.update(rlist)
        for rlist in n2.get("refs", {}).values():
            refs2.update(rlist)
        text1 = n1.get("text", "")
        text2 = n2.get("text", "")
        if text1 and text2:
            common = set(text1.split()) & set(text2.split())
            if len(common) > 5:
                return True
        return bool(refs1 & refs2)

    def _analyze_scope_focus(self) -> float:
        type_counts = defaultdict(int)
        total = 0
        for block in self.ledger.chain:
            for node in block.get("nodes", []):
                t = node.get("type", "unknown")
                type_counts[t] += 1
                total += 1
        if total == 0:
            return 0.0
        max_type = max(type_counts.values(), default=0)
        return max_type / total

    def _detect_missing_from_indices(self) -> bool:
        for block in self.ledger.chain:
            for node in block.get("nodes", []):
                for refs in node.get("refs", {}).values():
                    for target in refs:
                        if target not in self.ledger.index:
                            return True
        return False

    def _detect_decreasing_citations(self) -> bool:
        citation_trend = []
        for block in self.ledger.chain[-20:]:
            cites = 0
            for node in block.get("nodes", []):
                cites += sum(len(refs) for refs in node.get("refs", {}).values())
            citation_trend.append(cites)
        if len(citation_trend) < 5:
            return False
        for i in range(len(citation_trend)-1):
            if citation_trend[i+1] > citation_trend[i]:
                return False
        return True

    def _detect_archival_gaps(self, threshold_days: int = 7) -> bool:
        dates = sorted(self.ledger.temporal.keys())
        if len(dates) < 2:
            return False
        prev = datetime.fromisoformat(dates[0])
        for d in dates[1:]:
            curr = datetime.fromisoformat(d)
            if (curr - prev).days > threshold_days:
                return True
            prev = curr
        return False

    def _detect_repetitive_messaging(self) -> bool:
        texts = []
        for block in self.ledger.chain:
            for node in block.get("nodes", []):
                text = node.get("text", "")
                if text:
                    texts.append(text)
        if len(texts) < 3:
            return False
        similar = 0
        for i in range(len(texts)):
            for j in range(i+1, len(texts)):
                set_i = set(texts[i].split())
                set_j = set(texts[j].split())
                if len(set_i & set_j) / max(1, len(set_i | set_j)) > 0.8:
                    similar += 1
        return similar > len(texts) * 0.3

    def _detect_ad_hominem(self) -> bool:
        phrases = ["liar", "fraud", "stupid", "ignorant", "crank", "conspiracy theorist"]
        count = 0
        for block in self.ledger.chain:
            for node in block.get("nodes", []):
                text = node.get("text", "").lower()
                for phrase in phrases:
                    if phrase in text:
                        count += 1
                        break
        return count > 5

    def _detect_whataboutism(self) -> bool:
        patterns = ["what about", "but what about", "and what about"]
        count = 0
        for block in self.ledger.chain:
            for node in block.get("nodes", []):
                text = node.get("text", "").lower()
                for pat in patterns:
                    if pat in text:
                        count += 1
                        break
        return count > 3

    def _adjust_signatures_with_context(self, signatures: List[str]) -> List[str]:
        adjusted = []
        for sig in signatures:
            if sig == "entity_present_then_absent":
                last_block = self.ledger.chain[-1] if self.ledger.chain else None
                if last_block:
                    entities = self._extract_entities_from_nodes(last_block.get("nodes", []))
                    now = datetime.utcnow()
                    if any(self.metadata.is_natural_end(e, now) for e in entities):
                        continue
            adjusted.append(sig)
        return adjusted

    def _update_signature_counts(self, signatures: List[str]):
        for sig in signatures:
            self.signature_counts[sig] += 1

    def get_signature_base_rate(self, signature: str) -> float:
        if self.total_investigations == 0:
            return 0.0
        return self.signature_counts.get(signature, 0) / self.total_investigations

    def _signatures_to_methods(self, signatures: List[str]) -> List[Dict]:
        results = []
        for sig in signatures:
            mids = self.hierarchy.signatures.get(sig, [])
            for mid in mids:
                method = self.hierarchy.methods[mid]
                base_conf = self._calculate_method_confidence(method, sig)
                # Apply exponential weight reduction if base rate exceeds threshold
                base_rate = self.get_signature_base_rate(sig)
                if base_rate > self.signature_base_rate_threshold:
                    weight = math.exp(-self.exponential_decay_factor * (base_rate - self.signature_base_rate_threshold))
                    conf = base_conf * weight
                else:
                    conf = base_conf
                if method.implemented and conf > self.signature_confidence_threshold:
                    results.append({
                        "method_id": method.id,
                        "method_name": method.name,
                        "primitive": method.primitive.value,
                        "confidence": round(conf, 3),
                        "evidence_signature": sig,
                        "implemented": True
                    })
        return sorted(results, key=lambda x: x["confidence"], reverse=True)

    def _calculate_method_confidence(self, method: SuppressionMethod, signature: str) -> float:
        base = 0.7 if method.implemented else 0.3
        if signature in method.observable_signatures:
            base += 0.2
        if len(method.observable_signatures) > 1:
            base += 0.05
        return min(0.95, base)

    def _analyze_primitives(self, method_results: List[Dict]) -> Dict:
        counts = defaultdict(int)
        confs = defaultdict(list)
        for r in method_results:
            prim = r["primitive"]
            counts[prim] += 1
            confs[prim].append(r["confidence"])
        analysis = {}
        for prim, cnt in counts.items():
            analysis[prim] = {
                "method_count": cnt,
                "average_confidence": round(statistics.mean(confs[prim]), 3) if confs[prim] else 0.0,
                "dominant_methods": [r["method_name"] for r in method_results if r["primitive"] == prim][:2]
            }
        return analysis

    def _infer_lenses(self, primitive_analysis: Dict) -> Dict:
        active_prims = [p for p, data in primitive_analysis.items() if data["method_count"] > 0]
        active_lenses = set()
        for pstr in active_prims:
            prim = Primitive(pstr)
            lens_ids = self.hierarchy.primitives.get(prim, [])
            active_lenses.update(lens_ids)
        lens_details = []
        for lid in sorted(active_lenses)[:10]:
            lens = self.hierarchy.lenses.get(lid)
            if lens:
                lens_details.append({
                    "id": lens.id,
                    "name": lens.name,
                    "archetype": lens.archetype,
                    "mechanism": lens.suppression_mechanism
                })
        return {
            "active_lens_count": len(active_lenses),
            "active_primitives": active_prims,
            "lens_details": lens_details,
            "architecture_analysis": self._analyze_architecture(active_prims, active_lenses)
        }

    def _analyze_architecture(self, active_prims: List[str], active_lenses: Set[int]) -> str:
        analysis = []
        if len(active_prims) >= 3:
            analysis.append(f"Complex suppression architecture ({len(active_prims)} primitives)")
        elif active_prims:
            analysis.append("Basic suppression patterns detected")
        if len(active_lenses) > 20:
            analysis.append("Deep conceptual framework active")
        elif len(active_lenses) > 10:
            analysis.append("Multiple conceptual layers active")
        if Primitive.ERASURE.value in active_prims and Primitive.NARRATIVE_CAPTURE.value in active_prims:
            analysis.append("Erasure + Narrative patterns suggest coordinated suppression")
        if Primitive.META.value in active_prims:
            analysis.append("Meta-primitive active: self-referential control loops detected")
        if Primitive.ACCESS_CONTROL.value in active_prims and Primitive.DISCREDITATION.value in active_prims:
            analysis.append("Access control combined with discreditation: institutional self-protection likely")
        return "; ".join(analysis) if analysis else "No clear suppression architecture"

# =============================================================================
# PART XIII: EPISTEMIC MULTIPLEXOR (ENHANCED: moving average, fast/slow, reinforcement)
# =============================================================================

class Hypothesis:
    def __init__(self, description: str):
        self.description = description
        # Fast and slow amplitudes (initialised to a small baseline)
        self.fast_amp = 0.01
        self.slow_amp = 0.01
        self.signal_strength = 0.0           # reinforcement accumulator for repeated signals
        self.likelihood = 1.0                # for compatibility
        self.cost = 0.0
        self.history = []                    # track slow_amp over time for stability checks
        self.assumptions = []
        self.contradictions = 0
        self.ignored_evidence = 0

    def probability(self, use_slow: bool = True) -> float:
        """Return normalised probability (requires normalisation across hypotheses)."""
        return self.slow_amp if use_slow else self.fast_amp

    def record_history(self):
        self.history.append(self.slow_amp)

    def update(self, new_update: float, α_fast: float, α_slow: float):
        # Apply exponential smoothing
        self.fast_amp = (1 - α_fast) * self.fast_amp + α_fast * new_update
        self.slow_amp = (1 - α_slow) * self.slow_amp + α_slow * new_update
        # Clamp to [0,1] for safety
        self.fast_amp = max(0.0, min(1.0, self.fast_amp))
        self.slow_amp = max(0.0, min(1.0, self.slow_amp))

    def reinforce(self, β: float = 0.2, max_strength: float = 2.0):
        """Increase signal strength for repeated evidence."""
        self.signal_strength = min(max_strength, self.signal_strength + β)

class EpistemicMultiplexor:
    def __init__(self, stability_window: int = 5, collapse_threshold: float = 0.8,
                 null_hypothesis_weight: float = 0.6, positive_evidence_threshold: float = 0.3,
                 evidence_epsilon: float = 0.15,          # soft threshold for evidence strength
                 α_fast: float = 0.3, α_slow: float = 0.05,   # smoothing factors
                 reinforcement_β: float = 0.2,             # increment per repeated signal
                 reinforcement_max: float = 2.0):         # cap reinforcement
        self.hypotheses: List[Hypothesis] = []
        self.stability_window = stability_window
        self.collapse_threshold = collapse_threshold
        self.measurement_history = []
        self.null_hypothesis_weight = null_hypothesis_weight
        self.positive_evidence_threshold = positive_evidence_threshold
        self.evidence_epsilon = evidence_epsilon
        self.α_fast = α_fast
        self.α_slow = α_slow
        self.reinforcement_β = reinforcement_β
        self.reinforcement_max = reinforcement_max

    def initialize_from_evidence(self, evidence_nodes: List[EvidenceNode], base_hypotheses: List[str],
                                 include_admin_hypothesis: bool = True):
        if "Null: no suppression" not in base_hypotheses:
            base_hypotheses = ["Null: no suppression"] + base_hypotheses
        if include_admin_hypothesis and "Administrative/archival process" not in base_hypotheses:
            base_hypotheses = base_hypotheses + ["Administrative/archival process"]
        self.hypotheses = [Hypothesis(desc) for desc in base_hypotheses]
        # Normalise amplitudes so sum = 1 initially
        total = sum(h.slow_amp for h in self.hypotheses)
        if total > 0:
            for h in self.hypotheses:
                h.slow_amp /= total
                h.fast_amp /= total

    def update_amplitudes(self, evidence_nodes: List[EvidenceNode], detection_result: Dict,
                          kg_engine: KnowledgeGraphEngine, separator: Separator,
                          coherence_score: float = 0.0, refutation_evidence: Dict[str, float] = None):
        evidence_strength = self._compute_evidence_strength(detection_result)

        # Determine which signatures appeared (for reinforcement)
        signatures = detection_result.get("signatures", [])

        for h in self.hypotheses:
            # Base update factor (similar to old likelihood * adversarial)
            update_factor = self._compute_update_factor(h, evidence_strength, coherence_score,
                                                        detection_result, kg_engine, separator)
            if refutation_evidence and h.description in refutation_evidence:
                update_factor *= refutation_evidence[h.description]

            # Reinforcement: if any signature that supports this hypothesis appears
            # For simplicity, we reinforce if the hypothesis is one that would be boosted by the detection
            # (e.g., "suppression" hypothesis when suppression signatures present)
            if self._hypothesis_supported_by_signatures(h, signatures):
                h.reinforce(self.reinforcement_β, self.reinforcement_max)
                # Apply reinforcement as multiplier on update_factor
                update_factor = update_factor * (1.0 + h.signal_strength * 0.2)  # tunable

            # Apply the update
            h.update(update_factor, self.α_fast, self.α_slow)
            h.likelihood = update_factor  # for compatibility
            h.cost = self._compute_cost(h, kg_engine, separator)
            h.record_history()

        # Renormalise amplitudes so total probability sums to 1
        self._renormalise()

    def _compute_evidence_strength(self, detection_result: Dict) -> float:
        signatures = detection_result.get("signatures", [])
        raw_strength = len(signatures) / 5.0
        return max(0.0, raw_strength - self.evidence_epsilon)

    def _compute_update_factor(self, hypothesis: Hypothesis, evidence_strength: float,
                               coherence_score: float, detection_result: Dict,
                               kg_engine: KnowledgeGraphEngine, separator: Separator) -> float:
        """Compute the update factor (likelihood * adversarial) for the hypothesis."""
        # Likelihood based on hypothesis type and evidence strength
        desc = hypothesis.description.lower()
        if "null" in desc:
            likelihood = 1.0 - evidence_strength * 0.5 * (1 - coherence_score)
        elif "administrative" in desc:
            likelihood = 0.5 + evidence_strength * 0.3 * (1 - coherence_score)
        elif "suppression" in desc or "distorted" in desc:
            likelihood = evidence_strength * (coherence_score + 0.2)
        else:
            likelihood = 0.5 + evidence_strength * 0.3

        # Adversarial adjustment (penalises or boosts based on context)
        adversarial = self._adversarial_adjustment(detection_result, hypothesis, kg_engine, separator, coherence_score)

        # Product
        update = likelihood * adversarial
        # Clamp to [0,1] to keep amplitudes bounded
        return max(0.0, min(1.0, update))

    def _adversarial_adjustment(self, detection_result: Dict, hypothesis: Hypothesis,
                                kg_engine: KnowledgeGraphEngine, separator: Separator,
                                coherence_score: float) -> float:
        penalty = 1.0
        signatures = detection_result.get("signatures", [])
        evidence_strength = self._compute_evidence_strength(detection_result)

        if "entity_present_then_absent" in signatures:
            if "official" not in hypothesis.description.lower():
                penalty *= 0.7 * (1 - coherence_score)
        if "gradual_fading" in signatures:
            penalty *= 0.8
        if "single_explanation" in signatures:
            if "official" not in hypothesis.description.lower():
                penalty *= 0.5 * (1 - coherence_score)

        if evidence_strength < self.positive_evidence_threshold and coherence_score < 0.3:
            if "official" in hypothesis.description.lower():
                penalty = min(1.0, penalty * 1.2)
        if "administrative" in hypothesis.description.lower() and coherence_score < 0.3:
            penalty = min(1.0, penalty * 1.3)
        return penalty

    def _hypothesis_supported_by_signatures(self, hypothesis: Hypothesis, signatures: List[str]) -> bool:
        """Determine if the hypothesis is supported by the detected signatures (for reinforcement)."""
        desc = hypothesis.description.lower()
        if "suppression" in desc:
            # Suppression hypotheses are supported by erasure, narrative capture, etc.
            support_sigs = ["entity_present_then_absent", "single_explanation", "archival_gaps"]
            return any(sig in signatures for sig in support_sigs)
        if "official" in desc:
            support_sigs = ["authoritative_sources", "expert_consensus"]
            return any(sig in signatures for sig in support_sigs)
        if "administrative" in desc:
            support_sigs = ["archival_gaps", "missing_from_indices"]
            return any(sig in signatures for sig in support_sigs)
        return False

    def _compute_cost(self, hypothesis: Hypothesis, kg_engine: KnowledgeGraphEngine, separator: Separator) -> float:
        assumptions_cost = len(hypothesis.assumptions) * 0.1
        contradictions_cost = hypothesis.contradictions * 0.2
        ignored_cost = hypothesis.ignored_evidence * 0.05
        cost = assumptions_cost + contradictions_cost + ignored_cost
        return min(1.0, cost)

    def _renormalise(self):
        """Normalise slow and fast amplitudes so sum = 1 across hypotheses."""
        total_slow = sum(h.slow_amp for h in self.hypotheses)
        if total_slow > 0:
            for h in self.hypotheses:
                h.slow_amp /= total_slow
        total_fast = sum(h.fast_amp for h in self.hypotheses)
        if total_fast > 0:
            for h in self.hypotheses:
                h.fast_amp /= total_fast

    def get_probabilities(self, use_slow: bool = True) -> Dict[str, float]:
        """Return probability distribution over hypotheses."""
        if use_slow:
            total = sum(h.slow_amp for h in self.hypotheses)
            if total == 0:
                return {h.description: 0.0 for h in self.hypotheses}
            return {h.description: h.slow_amp / total for h in self.hypotheses}
        else:
            total = sum(h.fast_amp for h in self.hypotheses)
            if total == 0:
                return {h.description: 0.0 for h in self.hypotheses}
            return {h.description: h.fast_amp / total for h in self.hypotheses}

    def should_collapse(self) -> bool:
        """Use slow amplitudes for stability."""
        if not self.hypotheses:
            return False
        probs = self.get_probabilities(use_slow=True)
        best_desc = max(probs, key=probs.get)
        best_prob = probs[best_desc]
        if best_prob < self.collapse_threshold:
            return False
        if len(self.measurement_history) < self.stability_window:
            return False
        recent = self.measurement_history[-self.stability_window:]
        return all(desc == best_desc for desc in recent)

    def measure(self) -> Optional[Hypothesis]:
        """Collapse to the most probable hypothesis (based on slow amplitude)."""
        if not self.should_collapse():
            return None
        probs = self.get_probabilities(use_slow=True)
        best_desc = max(probs, key=probs.get)
        for h in self.hypotheses:
            if h.description == best_desc:
                return h
        return self.hypotheses[0]

    def record_measurement(self, hypothesis: Hypothesis):
        self.measurement_history.append(hypothesis.description)
        if len(self.measurement_history) > 100:
            self.measurement_history = self.measurement_history[-100:]

# =============================================================================
# PART XIV: PROBABILISTIC INFERENCE (unchanged from v3.2)
# =============================================================================

class ProbabilisticInference:
    def __init__(self):
        self.priors: Dict[str, float] = {}
        self.evidence: Dict[str, List[float]] = defaultdict(list)

    def set_prior_from_multiplexor(self, multiplexor: EpistemicMultiplexor):
        probs = multiplexor.get_probabilities(use_slow=True)
        for desc, prob in probs.items():
            self.priors[desc] = prob

    def add_evidence(self, hypothesis_id: str, likelihood: float):
        self.evidence[hypothesis_id].append(likelihood)

    def posterior(self, hypothesis_id: str) -> float:
        prior = self.priors.get(hypothesis_id, 0.5)
        likelihoods = self.evidence.get(hypothesis_id, [])
        if not likelihoods:
            return prior
        odds = prior / (1 - prior + 1e-9)
        for L in likelihoods:
            odds *= (L / (1 - L + 1e-9))
        posterior = odds / (1 + odds)
        return posterior

    def reset(self):
        self.priors.clear()
        self.evidence.clear()

    def set_prior(self, hypothesis_id: str, value: float):
        self.priors[hypothesis_id] = value

# =============================================================================
# PART XV: CONTEXT DETECTOR (unchanged from v3.2)
# =============================================================================

class ContextDetector:
    def detect(self, event_data: Dict) -> ControlContext:
        western_score = 0
        non_western_score = 0
        if event_data.get('procedure_complexity_score', 0) > 5:
            western_score += 1
        if len(event_data.get('involved_institutions', [])) > 3:
            western_score += 1
        if event_data.get('legal_technical_references', 0) > 10:
            western_score += 1
        if event_data.get('media_outlet_coverage_count', 0) > 20:
            western_score += 1
        if event_data.get('direct_state_control_score', 0) > 5:
            non_western_score += 1
        if event_data.get('special_legal_regimes', 0) > 2:
            non_western_score += 1
        if event_data.get('historical_narrative_regulation', False):
            non_western_score += 1
        if western_score > non_western_score * 1.5:
            return ControlContext.WESTERN
        elif non_western_score > western_score * 1.5:
            return ControlContext.NON_WESTERN
        elif western_score > 0 and non_western_score > 0:
            return ControlContext.HYBRID
        else:
            return ControlContext.GLOBAL

# =============================================================================
# PART XVI: META‑ANALYSIS – SAVIOR/SUFFERER MATRIX (unchanged from v3.2)
# =============================================================================

class ControlArchetypeAnalyzer:
    def __init__(self, hierarchy: SuppressionHierarchy):
        self.hierarchy = hierarchy
        self.archetype_map = {
            (Primitive.NARRATIVE_CAPTURE, Primitive.ACCESS_CONTROL): ControlArchetype.PRIEST_KING,
            (Primitive.ERASURE, Primitive.MISDIRECTION): ControlArchetype.IMPERIAL_RULER,
            (Primitive.SATURATION, Primitive.CONDITIONING): ControlArchetype.ALGORITHMIC_CURATOR,
            (Primitive.DISCREDITATION, Primitive.TEMPORAL): ControlArchetype.EXPERT_TECHNOCRAT,
            (Primitive.FRAGMENTATION, Primitive.ATTRITION): ControlArchetype.CORPORATE_OVERLORD,
        }

    def infer_archetype(self, detection_result: Dict) -> ControlArchetype:
        active_prims = set(detection_result.get("primitive_analysis", {}).keys())
        for (p1, p2), arch in self.archetype_map.items():
            if p1.value in active_prims and p2.value in active_prims:
                return arch
        return ControlArchetype.CORPORATE_OVERLORD

    def extract_slavery_mechanism(self, detection_result: Dict, kg_engine: KnowledgeGraphEngine) -> SlaveryMechanism:
        signatures = detection_result.get("signatures", [])
        visible = []
        invisible = []
        if "entity_present_then_absent" in signatures:
            visible.append("abrupt disappearance")
        if "gradual_fading" in signatures:
            invisible.append("attention decay")
        if "single_explanation" in signatures:
            invisible.append("narrative monopoly")
        bridge_nodes = kg_engine.bridge_nodes()
        if bridge_nodes:
            invisible.append("bridge node removal risk")
        return SlaveryMechanism(
            mechanism_id=f"inferred_{datetime.utcnow().isoformat()}",
            slavery_type=SlaveryType.PSYCHOLOGICAL_SLAVERY,
            visible_chains=visible,
            invisible_chains=invisible,
            voluntary_adoption_mechanisms=["aspirational identification"],
            self_justification_narratives=["I chose this"]
        )

class ConsciousnessMapper:
    def __init__(self, separator: Separator, symbolism_ai: 'SymbolismAI'):
        self.separator = separator
        self.symbolism_ai = symbolism_ai

    def analyze_consciousness(self, node_hashes: List[str]) -> Dict[str, float]:
        artifacts = []
        for h in node_hashes:
            node = self.separator.ledger.get_node(h)
            if node and node.get("text"):
                artifacts.append(node)
        if artifacts:
            scores = [self.symbolism_ai.analyze({"text": a["text"]}) for a in artifacts]
            avg_symbolism = np.mean(scores)
        else:
            avg_symbolism = 0.3
        return {
            "system_awareness": avg_symbolism * 0.8,
            "self_enslavement_awareness": avg_symbolism * 0.5,
            "manipulation_detection": avg_symbolism * 0.7,
            "liberation_desire": avg_symbolism * 0.6
        }

    def compute_freedom_illusion_index(self, control_system: ControlSystem) -> float:
        freedom_scores = list(control_system.freedom_illusions.values())
        enslavement_scores = list(control_system.self_enslavement_patterns.values())
        if not freedom_scores:
            return 0.5
        return min(1.0, np.mean(freedom_scores) * np.mean(enslavement_scores))

# =============================================================================
# PART XVII: PARADOX DETECTOR & IMMUNITY VERIFIER (unchanged from v3.2)
# =============================================================================

class RecursiveParadoxDetector:
    def __init__(self):
        self.paradox_types = {
            'self_referential_capture': "Framework conclusions used to validate framework",
            'institutional_recursion': "Institution uses framework to legitimize itself",
            'narrative_feedback_loop': "Findings reinforce narrative being analyzed",
        }

    def detect(self, framework_output: Dict, event_context: Dict) -> Dict:
        paradoxes = []
        if self._check_self_referential(framework_output):
            paradoxes.append('self_referential_capture')
        if self._check_institutional_recursion(framework_output, event_context):
            paradoxes.append('institutional_recursion')
        if self._check_narrative_feedback(framework_output):
            paradoxes.append('narrative_feedback_loop')
        return {
            "paradoxes_detected": paradoxes,
            "count": len(paradoxes),
            "resolutions": self._generate_resolutions(paradoxes)
        }

    def _check_self_referential(self, output: Dict) -> bool:
        detection = output.get("detection", {})
        if "Meta-primitive active" in detection.get("lens_inference", {}).get("architecture_analysis", ""):
            return True
        return False

    def _check_institutional_recursion(self, output: Dict, context: Dict) -> bool:
        institution = context.get("institution", "")
        if not institution:
            return False
        probabilities = output.get("multiplexor_probabilities", {})
        if probabilities.get("Official narrative is accurate", 0) > 0.7:
            return True
        return False

    def _check_narrative_feedback(self, output: Dict) -> bool:
        collapsed = output.get("collapsed_hypothesis", "")
        claim = output.get("claim", "")
        if collapsed and claim:
            if claim.lower() in collapsed.lower() or collapsed.lower() in claim.lower():
                return True
        return False

    def _generate_resolutions(self, paradoxes: List[str]) -> List[str]:
        if not paradoxes:
            return []
        res = ["Require external audit"]
        if 'self_referential_capture' in paradoxes:
            res.append("Run detection with independent validators")
        if 'institutional_recursion' in paradoxes:
            res.append("Exclude institutional sources from prior weighting")
        if 'narrative_feedback_loop' in paradoxes:
            res.append("Introduce adversarial hypothesis with opposite claim")
        return res

class ImmunityVerifier:
    def __init__(self):
        pass

    def verify(self, framework_components: Dict) -> Dict:
        tests = {
            'power_analysis_inversion': self._test_power_analysis_inversion(framework_components),
            'narrative_audit_reversal': self._test_narrative_audit_reversal(framework_components),
            'symbolic_analysis_weaponization': self._test_symbolic_analysis_weaponization(framework_components),
        }
        immune = all(tests.values())
        return {
            "immune": immune,
            "test_results": tests,
            "proof": "All inversion tests passed." if immune else "Vulnerabilities detected."
        }

    def _test_power_analysis_inversion(self, components: Dict) -> bool:
        priors = components.get("priors", {})
        if priors.get("Official narrative is accurate", 0.5) < 0.3:
            return False
        return True

    def _test_narrative_audit_reversal(self, components: Dict) -> bool:
        return True

    def _test_symbolic_analysis_weaponization(self, components: Dict) -> bool:
        return True

# =============================================================================
# PART XVIII: AI AGENTS (unchanged from v3.2)
# =============================================================================

class IngestionAI:
    def __init__(self, crypto: Crypto):
        self.crypto = crypto

    def process_document(self, text: str, source: str) -> EvidenceNode:
        node_hash = self.crypto.hash(text + source)
        node = EvidenceNode(
            hash=node_hash,
            type="document",
            source=source,
            signature="",
            timestamp=datetime.utcnow().isoformat() + "Z",
            witnesses=[],
            refs={},
            text=text
        )
        node.signature = self.crypto.sign(node_hash.encode(), "ingestion_ai")
        return node

class SymbolismAI:
    def __init__(self):
        self.model = None
        if HAS_TRANSFORMERS:
            try:
                self.model = sentence_transformers.SentenceTransformer('all-MiniLM-L6-v2')
            except:
                self.model = None

    def analyze(self, artifact: Dict) -> float:
        text = artifact.get("text", "")
        if not text:
            return 0.3 + (hash(artifact.get("id", "")) % 70) / 100.0

        if self.model is not None:
            suppressed_keywords = [
                "cover-up", "conspiracy", "truth", "hidden", "secret", "censored",
                "suppressed", "whistleblower", "classified", "exposed"
            ]
            text_embed = self.model.encode([text])[0]
            kw_embeds = self.model.encode(suppressed_keywords)
            similarities = np.dot(kw_embeds, text_embed) / (np.linalg.norm(kw_embeds, axis=1) * np.linalg.norm(text_embed))
            max_sim = np.max(similarities)
            return 0.2 + 0.7 * max_sim
        else:
            score = 0.0
            for kw in ["cover-up", "conspiracy", "truth", "hidden", "secret", "censored", "suppressed"]:
                if kw in text.lower():
                    score += 0.1
            return min(0.9, 0.3 + score)

class ReasoningAI:
    def __init__(self, inference: ProbabilisticInference, controller_ref: 'AIController'):
        self.inference = inference
        self.controller = controller_ref

    def evaluate_claim(self, claim_id: str, nodes: List[EvidenceNode], detector_result: Dict) -> Dict:
        confidence = 0.5
        if detector_result.get("evidence_found", 0) > 2:
            confidence += 0.2
        prim_analysis = detector_result.get("primitive_analysis", {})
        if prim_analysis:
            confidence *= (1 - 0.05 * len(prim_analysis))
        self.inference.set_prior(claim_id, confidence)

        # Check for failing alternative hypotheses
        if self.controller:
            probs = self.controller.multiplexor.get_probabilities(use_slow=True)
            for hyp_desc, prob in probs.items():
                if prob < 0.2 and hyp_desc not in ["Official narrative is accurate", "Evidence is suppressed or distorted"]:
                    self.controller.spawn_refutation(claim_id, hyp_desc)
                    return {"spawn_sub": True, "reason": f"Testing failing hypothesis: {hyp_desc}", "priority": "medium"}

        if confidence < 0.6:
            return {"spawn_sub": True, "reason": "low confidence", "priority": "high"}
        elif confidence < 0.75:
            return {"spawn_sub": True, "reason": "moderate confidence, need deeper analysis", "priority": "medium"}
        else:
            return {"spawn_sub": False, "reason": "sufficient evidence"}

# =============================================================================
# PART XIX: AI CONTROLLER (with refutation and self‑audit)
# =============================================================================

class AIController:
    def __init__(self, ledger: Ledger, separator: Separator, detector: HierarchicalDetector,
                 kg: KnowledgeGraphEngine, temporal: TemporalAnalyzer, inference: ProbabilisticInference,
                 ingestion_ai: IngestionAI, symbolism_ai: SymbolismAI, reasoning_ai: ReasoningAI,
                 multiplexor: EpistemicMultiplexor, context_detector: ContextDetector,
                 archetype_analyzer: ControlArchetypeAnalyzer, consciousness_mapper: ConsciousnessMapper,
                 paradox_detector: RecursiveParadoxDetector, immunity_verifier: ImmunityVerifier,
                 metadata_registry: ExternalMetadataRegistry, coherence_checker: NarrativeCoherenceChecker,
                 self_audit: 'SelfAudit'):
        self.ledger = ledger
        self.separator = separator
        self.detector = detector
        self.kg = kg
        self.temporal = temporal
        self.inference = inference
        self.ingestion_ai = ingestion_ai
        self.symbolism_ai = symbolism_ai
        self.reasoning_ai = reasoning_ai
        self.multiplexor = multiplexor
        self.context_detector = context_detector
        self.archetype_analyzer = archetype_analyzer
        self.consciousness_mapper = consciousness_mapper
        self.paradox_detector = paradox_detector
        self.immunity_verifier = immunity_verifier
        self.metadata = metadata_registry
        self.coherence = coherence_checker
        self.self_audit = self_audit
        self.contexts: Dict[str, Dict] = {}
        self._lock = threading.Lock()
        self._task_queue = queue.Queue()
        self._worker_thread = threading.Thread(target=self._process_queue, daemon=True)
        self._worker_running = True
        self._worker_thread.start()
        self._audit_timer = threading.Timer(3600, self._periodic_audit)
        self._audit_timer.daemon = True
        self._audit_timer.start()

    def _periodic_audit(self):
        self.self_audit.run_audit()
        self.self_audit.apply_suggestions()
        self._audit_timer = threading.Timer(3600, self._periodic_audit)
        self._audit_timer.start()

    def submit_claim(self, claim_text: str) -> str:
        corr_id = str(uuid.uuid4())
        context = {
            "correlation_id": corr_id,
            "parent_id": None,
            "claim": claim_text,
            "status": "pending",
            "created": datetime.utcnow().isoformat() + "Z",
            "evidence_nodes": [],
            "sub_investigations": [],
            "results": {},
            "multiplexor_state": None,
            "refutation_target": None
        }
        with self._lock:
            self.contexts[corr_id] = context
        thread = threading.Thread(target=self._investigate, args=(corr_id,))
        thread.start()
        return corr_id

    def spawn_refutation(self, parent_id: str, hypothesis_desc: str):
        sub_id = str(uuid.uuid4())
        sub_context = {
            "correlation_id": sub_id,
            "parent_id": parent_id,
            "claim": f"Refutation task for hypothesis: {hypothesis_desc}",
            "status": "pending",
            "created": datetime.utcnow().isoformat() + "Z",
            "evidence_nodes": [],
            "sub_investigations": [],
            "results": {},
            "multiplexor_state": None,
            "refutation_target": hypothesis_desc
        }
        with self._lock:
            self.contexts[sub_id] = sub_context
            if parent_id in self.contexts:
                self.contexts[parent_id]["sub_investigations"].append(sub_id)
        self._task_queue.put(sub_id)

    def _investigate(self, corr_id: str):
        with self._lock:
            context = self.contexts.get(corr_id)
            if not context:
                return
            context["status"] = "active"

        try:
            if context.get("refutation_target"):
                self._handle_refutation(corr_id)
                return

            event_data = {"description": context["claim"]}
            ctxt = self.context_detector.detect(event_data)
            context["control_context"] = ctxt.value

            detection = self.detector.detect_from_ledger(investigation_id=corr_id)
            context["detection"] = detection

            entities = self._extract_entities_from_text(context["claim"])
            coherence_score = 0.0
            if entities:
                coherence_score = self.coherence.check_causal_disruption(entities[0], datetime.utcnow())

            base_hypotheses = [
                "Official narrative is accurate",
                "Evidence is suppressed or distorted",
                "Institutional interests shaped the narrative",
                "Multiple independent sources confirm the claim",
                "The claim is part of a disinformation campaign"
            ]
            self.multiplexor.initialize_from_evidence([], base_hypotheses, include_admin_hypothesis=True)
            for _ in range(3):
                self.multiplexor.update_amplitudes([], detection, self.kg, self.separator, coherence_score)
                collapsed = self.multiplexor.measure()
                if collapsed:
                    break
            if not collapsed:
                probs = self.multiplexor.get_probabilities(use_slow=True)
                best_desc = max(probs, key=probs.get)
                collapsed = next((h for h in self.multiplexor.hypotheses if h.description == best_desc), None)
            if collapsed:
                self.multiplexor.record_measurement(collapsed)

            self.inference.set_prior_from_multiplexor(self.multiplexor)

            decision = self.reasoning_ai.evaluate_claim(corr_id, [], detection)
            if decision.get("spawn_sub") and not decision.get("reason", "").startswith("Testing failing hypothesis"):
                sub_id = str(uuid.uuid4())
                context["sub_investigations"].append(sub_id)
                sub_context = {
                    "correlation_id": sub_id,
                    "parent_id": corr_id,
                    "claim": f"Sub-investigation for {context['claim']}: {decision['reason']}",
                    "status": "pending",
                    "created": datetime.utcnow().isoformat() + "Z",
                    "evidence_nodes": [],
                    "sub_investigations": [],
                    "results": {},
                    "multiplexor_state": None,
                    "refutation_target": None
                }
                with self._lock:
                    self.contexts[sub_id] = sub_context
                self._task_queue.put(sub_id)

            archetype = self.archetype_analyzer.infer_archetype(detection)
            slavery_mech = self.archetype_analyzer.extract_slavery_mechanism(detection, self.kg)
            consciousness = self.consciousness_mapper.analyze_consciousness([])
            context["meta"] = {
                "archetype": archetype.value,
                "slavery_mechanism": slavery_mech.mechanism_id,
                "consciousness": consciousness
            }

            paradox = self.paradox_detector.detect({
                "detection": detection,
                "multiplexor_probabilities": self.multiplexor.get_probabilities(use_slow=True),
                "collapsed_hypothesis": collapsed.description if collapsed else None,
                "claim": context["claim"]
            }, event_data)
            context["paradox"] = paradox

            final_confidence = 0.6
            if paradox["count"] > 0:
                final_confidence = max(0.3, final_confidence - 0.2 * paradox["count"])
                if paradox["count"] >= 2:
                    context["requires_audit"] = True

            immunity = self.immunity_verifier.verify({"priors": self.inference.priors})
            context["immunity"] = immunity

            interpretation = {
                "narrative": f"Claim evaluated: {context['claim']}",
                "detection_summary": detection,
                "multiplexor_probabilities": self.multiplexor.get_probabilities(use_slow=True),
                "collapsed_hypothesis": collapsed.description if collapsed else None,
                "meta": context["meta"],
                "paradox": paradox,
                "immunity": immunity,
                "coherence_score": coherence_score
            }
            node_hashes = []
            int_id = self.separator.add(node_hashes, interpretation, "AI_Controller", confidence=final_confidence)
            context["results"] = {
                "confidence": final_confidence,
                "interpretation_id": int_id,
                "detection": detection,
                "collapsed_hypothesis": collapsed.description if collapsed else None,
                "meta": context["meta"],
                "paradox": paradox,
                "immunity": immunity,
                "requires_audit": context.get("requires_audit", False),
                "coherence_score": coherence_score
            }
            context["multiplexor_state"] = {
                "hypotheses": [{"description": h.description, "probability": h.slow_amp} for h in self.multiplexor.hypotheses]
            }
            context["status"] = "complete"
        except Exception as e:
            print(f"Investigation {corr_id} failed: {e}")
            with self._lock:
                if corr_id in self.contexts:
                    self.contexts[corr_id]["status"] = "failed"
                    self.contexts[corr_id]["error"] = str(e)
        finally:
            with self._lock:
                if corr_id in self.contexts:
                    self.contexts[corr_id]["status"] = context.get("status", "failed")

    def _handle_refutation(self, corr_id: str):
        with self._lock:
            context = self.contexts.get(corr_id)
            if not context:
                return
            hypothesis = context["refutation_target"]
            parent_id = context["parent_id"]

        support_score = 0.0
        if "administrative" in hypothesis.lower():
            keywords = ["classified", "archived", "sealed", "FOIA", "retention", "declassification"]
            count = 0
            for block in self.ledger.chain:
                for node in block.get("nodes", []):
                    text = node.get("text", "").lower()
                    for kw in keywords:
                        if kw in text:
                            count += 1
                            break
            support_score = min(1.0, count / 5.0)
        elif "natural lifecycle" in hypothesis.lower():
            parent_claim = self.contexts[parent_id]["claim"]
            entities = self._extract_entities_from_text(parent_claim)
            if entities:
                now = datetime.utcnow()
                if any(self.metadata.is_natural_end(e, now) for e in entities):
                    support_score = 0.8
                else:
                    support_score = 0.2
        elif "information noise" in hypothesis.lower():
            total_nodes = sum(len(block.get("nodes", [])) for block in self.ledger.chain)
            unique_sources = set()
            for block in self.ledger.chain:
                for node in block.get("nodes", []):
                    if node.get("source"):
                        unique_sources.add(node["source"])
            if total_nodes > 100 and len(unique_sources) > 20:
                support_score = 0.6
            else:
                support_score = 0.3
        else:
            support_score = 0.5

        with self._lock:
            parent = self.contexts.get(parent_id)
            if parent:
                self.inference.add_evidence(hypothesis, support_score)
                parent["results"]["refutation_evidence"] = parent["results"].get("refutation_evidence", {})
                parent["results"]["refutation_evidence"][hypothesis] = support_score
                parent["status"] = "updated_by_refutation"

        interpretation = {
            "refutation_target": hypothesis,
            "support_score": support_score,
            "method": "keyword_search"
        }
        int_id = self.separator.add([], interpretation, "RefutationAI", confidence=support_score)
        context["results"] = {"interpretation_id": int_id, "support_score": support_score}
        context["status"] = "complete"

    def _extract_entities_from_text(self, text: str) -> List[str]:
        words = text.split()
        entities = []
        for w in words:
            if w and w[0].isupper() and len(w) > 1 and w not in {"The","A","An","I","We"}:
                entities.append(w.strip(".,;:!?"))
        return entities

    def _process_queue(self):
        while self._worker_running:
            try:
                corr_id = self._task_queue.get(timeout=1)
                self._investigate(corr_id)
            except queue.Empty:
                continue

    def get_status(self, corr_id: str) -> Dict:
        with self._lock:
            return self.contexts.get(corr_id, {"error": "not found"})

    def shutdown(self):
        self._worker_running = False
        self._worker_thread.join(timeout=2)
        self._audit_timer.cancel()

# =============================================================================
# PART XX: SELF‑AUDIT MODULE (ENHANCED: tunes α and ε)
# =============================================================================

class SelfAudit:
    def __init__(self, detector: HierarchicalDetector, multiplexor: EpistemicMultiplexor,
                 metadata_registry: ExternalMetadataRegistry):
        self.detector = detector
        self.multiplexor = multiplexor
        self.metadata = metadata_registry
        self.audit_log: List[Dict] = []

    def run_audit(self) -> Dict:
        suggestions = []
        # Signature base rates
        for sig in self.detector.signature_counts:
            rate = self.detector.get_signature_base_rate(sig)
            if rate > 0.5:
                suggestions.append({
                    "signature": sig,
                    "base_rate": rate,
                    "suggestion": f"Increase threshold for {sig}, appears too often"
                })
        # Collapse rates
        if self.multiplexor.measurement_history:
            collapse_counts = defaultdict(int)
            for desc in self.multiplexor.measurement_history:
                collapse_counts[desc] += 1
            total_collapses = len(self.multiplexor.measurement_history)
            for desc, cnt in collapse_counts.items():
                rate = cnt / total_collapses
                if "suppression" in desc.lower() and rate > 0.7:
                    suggestions.append({
                        "hypothesis": desc,
                        "collapse_rate": rate,
                        "suggestion": "Too many suppression conclusions; consider raising positive_evidence_threshold or evidence_epsilon"
                    })
        # Also suggest adjustments to α_fast, α_slow based on oscillation? (optional)
        # For simplicity, we only adjust evidence_epsilon and positive_evidence_threshold here.
        audit_report = {
            "timestamp": datetime.utcnow().isoformat() + "Z",
            "suggestions": suggestions,
            "signature_counts": dict(self.detector.signature_counts),
            "total_investigations": self.detector.total_investigations,
            "multiplexor_params": {
                "evidence_epsilon": self.multiplexor.evidence_epsilon,
                "α_fast": self.multiplexor.α_fast,
                "α_slow": self.multiplexor.α_slow,
                "positive_evidence_threshold": self.multiplexor.positive_evidence_threshold
            }
        }
        self.audit_log.append(audit_report)
        return audit_report

    def apply_suggestions(self):
        # Tune detector thresholds
        for suggestion in self.run_audit().get("suggestions", []):
            if "increase threshold" in suggestion["suggestion"]:
                if suggestion["base_rate"] > 0.6:
                    self.detector.positive_evidence_min_signatures = max(2, self.detector.positive_evidence_min_signatures + 1)
            if "raise positive_evidence_threshold" in suggestion["suggestion"]:
                self.multiplexor.positive_evidence_threshold = min(0.8, self.multiplexor.positive_evidence_threshold + 0.05)
            if "evidence_epsilon" in suggestion["suggestion"]:
                self.multiplexor.evidence_epsilon = min(0.5, self.multiplexor.evidence_epsilon + 0.05)

# =============================================================================
# PART XXI: API LAYER (Flask) – unchanged from v3.2
# =============================================================================

app = Flask(__name__)
controller: Optional[AIController] = None

@app.route('/api/v1/submit_claim', methods=['POST'])
def submit_claim():
    data = request.get_json()
    claim = data.get('claim')
    if not claim:
        return jsonify({"error": "Missing claim"}), 400
    corr_id = controller.submit_claim(claim)
    return jsonify({"investigation_id": corr_id})

@app.route('/api/v1/investigation/<corr_id>', methods=['GET'])
def get_investigation(corr_id):
    status = controller.get_status(corr_id)
    return jsonify(status)

@app.route('/api/v1/node/<node_hash>', methods=['GET'])
def get_node(node_hash):
    node = controller.ledger.get_node(node_hash)
    if node:
        return jsonify(node)
    return jsonify({"error": "Node not found"}), 404

@app.route('/api/v1/interpretations/<node_hash>', methods=['GET'])
def get_interpretations(node_hash):
    ints = controller.separator.get_interpretations(node_hash)
    return jsonify([i.__dict__ for i in ints])

@app.route('/api/v1/detect', methods=['GET'])
def run_detection():
    result = controller.detector.detect_from_ledger()
    return jsonify(result)

@app.route('/api/v1/verify_chain', methods=['GET'])
def verify_chain():
    result = controller.ledger.verify_chain()
    return jsonify(result)

@app.route('/api/v1/multiplexor/state', methods=['GET'])
def get_multiplexor_state():
    if not controller:
        return jsonify({"error": "Controller not initialized"}), 500
    with controller._lock:
        state = {
            "hypotheses": [{"description": h.description, "fast_amp": h.fast_amp, "slow_amp": h.slow_amp,
                            "cost": h.cost, "likelihood": h.likelihood} for h in controller.multiplexor.hypotheses],
            "stability_window": controller.multiplexor.stability_window,
            "collapse_threshold": controller.multiplexor.collapse_threshold,
            "measurement_history": controller.multiplexor.measurement_history,
            "evidence_epsilon": controller.multiplexor.evidence_epsilon,
            "α_fast": controller.multiplexor.α_fast,
            "α_slow": controller.multiplexor.α_slow
        }
    return jsonify(state)

@app.route('/api/v1/search', methods=['GET'])
def search_text():
    keyword = request.args.get('q', '')
    if not keyword:
        return jsonify({"error": "Missing query parameter 'q'"}), 400
    results = controller.ledger.search_text(keyword)
    return jsonify(results)

@app.route('/api/v1/temporal/gaps', methods=['GET'])
def get_gaps():
    gaps = controller.temporal.publication_gaps()
    return jsonify(gaps)

@app.route('/api/v1/shutdown', methods=['POST'])
def shutdown():
    controller.shutdown()
    return jsonify({"message": "Shutting down"})

# =============================================================================
# PART XXII: MAIN – Initialization and Startup
# =============================================================================

def main():
    crypto = Crypto("./keys")
    ledger = Ledger("./ledger.json", crypto)
    separator = Separator(ledger, "./separator")
    hierarchy = SuppressionHierarchy()
    metadata_registry = ExternalMetadataRegistry("./metadata.json")
    kg = KnowledgeGraphEngine(ledger)
    coherence_checker = NarrativeCoherenceChecker(kg, separator)
    detector = HierarchicalDetector(hierarchy, ledger, separator, metadata_registry, coherence_checker)
    temporal = TemporalAnalyzer(ledger)
    inference = ProbabilisticInference()
    multiplexor = EpistemicMultiplexor(stability_window=5, collapse_threshold=0.8,
                                       null_hypothesis_weight=0.6, positive_evidence_threshold=0.3,
                                       evidence_epsilon=0.15, α_fast=0.3, α_slow=0.05,
                                       reinforcement_β=0.2, reinforcement_max=2.0)
    context_detector = ContextDetector()
    ingestion_ai = IngestionAI(crypto)
    symbolism_ai = SymbolismAI()
    reasoning_ai = ReasoningAI(inference, None)  # will set controller later
    archetype_analyzer = ControlArchetypeAnalyzer(hierarchy)
    consciousness_mapper = ConsciousnessMapper(separator, symbolism_ai)
    paradox_detector = RecursiveParadoxDetector()
    immunity_verifier = ImmunityVerifier()
    self_audit = SelfAudit(detector, multiplexor, metadata_registry)

    global controller
    controller = AIController(
        ledger=ledger,
        separator=separator,
        detector=detector,
        kg=kg,
        temporal=temporal,
        inference=inference,
        ingestion_ai=ingestion_ai,
        symbolism_ai=symbolism_ai,
        reasoning_ai=reasoning_ai,
        multiplexor=multiplexor,
        context_detector=context_detector,
        archetype_analyzer=archetype_analyzer,
        consciousness_mapper=consciousness_mapper,
        paradox_detector=paradox_detector,
        immunity_verifier=immunity_verifier,
        metadata_registry=metadata_registry,
        coherence_checker=coherence_checker,
        self_audit=self_audit
    )
    # Set controller reference in reasoning_ai now that controller exists
    reasoning_ai.controller = controller

    print("Epistemic Integrity System v3.3 (Adaptive Moving Average with Reinforcement) starting...")
    print("API available at http://localhost:5000")
    app.run(debug=True, port=5000)

if __name__ == "__main__":
    main()