File size: 53,903 Bytes
824bf31 d568e3b 824bf31 3c90182 824bf31 c78f128 3c90182 824bf31 c78f128 824bf31 d568e3b c78f128 d568e3b 824bf31 c78f128 824bf31 d568e3b 824bf31 d568e3b 824bf31 d568e3b 824bf31 d568e3b 824bf31 d568e3b 824bf31 3c90182 824bf31 d568e3b 824bf31 d568e3b 3c90182 824bf31 d568e3b 824bf31 3c90182 824bf31 d568e3b 824bf31 c78f128 824bf31 3c90182 824bf31 d568e3b 824bf31 3c90182 824bf31 3c90182 824bf31 d568e3b 824bf31 3c90182 824bf31 d568e3b 824bf31 d568e3b 824bf31 3c90182 824bf31 3c90182 824bf31 c78f128 824bf31 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 |
"""
Module: agents.zumbi
Codinome: Zumbi - Investigador de Padrões
Description: Agent specialized in detecting anomalies and suspicious patterns in government data
Author: Anderson H. Silva
Date: 2025-01-24
License: Proprietary - All rights reserved
"""
import asyncio
from datetime import datetime, timedelta
from typing import Any, Dict, List, Optional, Tuple
from dataclasses import dataclass
import numpy as np
import pandas as pd
from pydantic import BaseModel, Field as PydanticField
from src.agents.deodoro import BaseAgent, AgentContext, AgentMessage, AgentResponse
from src.core import get_logger, AgentStatus
from src.core.exceptions import AgentExecutionError, DataAnalysisError
from src.core.monitoring import (
INVESTIGATIONS_TOTAL, ANOMALIES_DETECTED, INVESTIGATION_DURATION,
DATA_RECORDS_PROCESSED, TRANSPARENCY_API_DATA_FETCHED
)
from src.tools.transparency_api import TransparencyAPIClient, TransparencyAPIFilter
from src.tools.models_client import ModelsClient, get_models_client
from src.ml.spectral_analyzer import SpectralAnalyzer, SpectralAnomaly
from src.tools.dados_gov_tool import DadosGovTool
import time
@dataclass
class AnomalyResult:
"""Result of anomaly detection analysis."""
anomaly_type: str
severity: float # 0.0 to 1.0
confidence: float # 0.0 to 1.0
description: str
explanation: str
evidence: Dict[str, Any]
recommendations: List[str]
affected_entities: List[Dict[str, Any]]
financial_impact: Optional[float] = None
class InvestigationRequest(BaseModel):
"""Request for investigation with specific parameters."""
query: str = PydanticField(description="Natural language investigation query")
organization_codes: Optional[List[str]] = PydanticField(default=None, description="Specific organization codes to investigate")
date_range: Optional[Tuple[str, str]] = PydanticField(default=None, description="Date range (start, end) in DD/MM/YYYY format")
value_threshold: Optional[float] = PydanticField(default=None, description="Minimum value threshold for contracts")
anomaly_types: Optional[List[str]] = PydanticField(default=None, description="Specific types of anomalies to look for")
max_records: int = PydanticField(default=100, description="Maximum records to analyze")
enable_open_data_enrichment: bool = PydanticField(default=True, description="Enable enrichment with dados.gov.br open data")
class InvestigatorAgent(BaseAgent):
"""
Agent specialized in detecting anomalies and suspicious patterns in government data.
Capabilities:
- Price anomaly detection (overpriced contracts)
- Temporal pattern analysis (suspicious timing)
- Vendor concentration analysis (monopolization)
- Duplicate contract detection
- Unusual payment patterns
- Explainable AI for transparency
"""
def __init__(
self,
price_anomaly_threshold: float = 2.5, # Standard deviations
concentration_threshold: float = 0.7, # 70% concentration trigger
duplicate_similarity_threshold: float = 0.85, # 85% similarity
):
"""
Initialize the Investigator Agent.
Args:
price_anomaly_threshold: Number of standard deviations for price anomalies
concentration_threshold: Threshold for vendor concentration (0-1)
duplicate_similarity_threshold: Threshold for duplicate detection (0-1)
"""
super().__init__(
name="Zumbi",
description="Zumbi dos Palmares - Agent specialized in detecting anomalies and suspicious patterns in government data",
capabilities=[
"price_anomaly_detection",
"temporal_pattern_analysis",
"vendor_concentration_analysis",
"duplicate_contract_detection",
"payment_pattern_analysis",
"spectral_analysis",
"explainable_ai",
"open_data_enrichment"
],
max_retries=3,
timeout=60
)
self.price_threshold = price_anomaly_threshold
self.concentration_threshold = concentration_threshold
self.duplicate_threshold = duplicate_similarity_threshold
# Initialize models client for ML inference (only if enabled)
from src.core import settings
if settings.models_api_enabled:
self.models_client = get_models_client()
else:
self.models_client = None
self.logger.info("Models API disabled, using only local ML")
# Initialize spectral analyzer for frequency-domain analysis (fallback)
self.spectral_analyzer = SpectralAnalyzer()
# Initialize dados.gov.br tool for accessing open data
self.dados_gov_tool = DadosGovTool()
# Anomaly detection methods registry
self.anomaly_detectors = {
"price_anomaly": self._detect_price_anomalies,
"vendor_concentration": self._detect_vendor_concentration,
"temporal_patterns": self._detect_temporal_anomalies,
"spectral_patterns": self._detect_spectral_anomalies,
"duplicate_contracts": self._detect_duplicate_contracts,
"payment_patterns": self._detect_payment_anomalies,
}
self.logger.info(
"zumbi_initialized",
agent_name=self.name,
price_threshold=price_anomaly_threshold,
concentration_threshold=concentration_threshold,
)
async def initialize(self) -> None:
"""Initialize agent resources."""
self.logger.info(f"{self.name} agent initialized")
async def shutdown(self) -> None:
"""Cleanup agent resources."""
self.logger.info(f"{self.name} agent shutting down")
async def process(
self,
message: AgentMessage,
context: AgentContext
) -> AgentResponse:
"""
Process investigation request and return anomaly detection results.
Args:
message: Investigation request message
context: Agent execution context
Returns:
AgentResponse with detected anomalies
"""
investigation_start_time = time.time()
try:
self.logger.info(
"investigation_started",
investigation_id=context.investigation_id,
agent_name=self.name,
action=message.action,
)
# Parse investigation request
if message.action == "investigate":
request = InvestigationRequest(**message.payload)
# Record investigation start
INVESTIGATIONS_TOTAL.labels(
agent_type="zumbi",
investigation_type=request.anomaly_types[0] if request.anomaly_types else "general",
status="started"
).inc()
else:
raise AgentExecutionError(
f"Unsupported action: {message.action}",
agent_id=self.name
)
# Fetch data for investigation
contracts_data = await self._fetch_investigation_data(request, context)
# Record data processed
DATA_RECORDS_PROCESSED.labels(
data_source="transparency_api",
agent="zumbi",
operation="fetch"
).inc(len(contracts_data) if contracts_data else 0)
if not contracts_data:
return AgentResponse(
agent_name=self.name,
status=AgentStatus.COMPLETED,
result={
"status": "no_data",
"message": "No data found for the specified criteria",
"anomalies": [],
"summary": {"total_records": 0, "anomalies_found": 0}
},
metadata={"investigation_id": context.investigation_id}
)
# Enrich data with open data information if available
if request.enable_open_data_enrichment:
contracts_data = await self._enrich_with_open_data(
contracts_data,
context
)
# Run anomaly detection
anomalies = await self._run_anomaly_detection(
contracts_data,
request,
context
)
# Record anomalies detected
for anomaly in anomalies:
ANOMALIES_DETECTED.labels(
anomaly_type=anomaly.anomaly_type,
severity="high" if anomaly.severity > 0.7 else "medium" if anomaly.severity > 0.4 else "low",
agent="zumbi"
).inc()
# Generate investigation summary
summary = self._generate_investigation_summary(contracts_data, anomalies)
# Create result message
result = {
"status": "completed",
"query": request.query,
"anomalies": [self._anomaly_to_dict(a) for a in anomalies],
"summary": summary,
"metadata": {
"investigation_id": context.investigation_id,
"timestamp": datetime.utcnow().isoformat(),
"agent_name": self.name,
"records_analyzed": len(contracts_data),
"anomalies_detected": len(anomalies),
}
}
# Record investigation completion and duration
investigation_duration = time.time() - investigation_start_time
INVESTIGATION_DURATION.labels(
agent_type="zumbi",
investigation_type=request.anomaly_types[0] if request.anomaly_types else "general"
).observe(investigation_duration)
INVESTIGATIONS_TOTAL.labels(
agent_type="zumbi",
investigation_type=request.anomaly_types[0] if request.anomaly_types else "general",
status="completed"
).inc()
self.logger.info(
"investigation_completed",
investigation_id=context.investigation_id,
records_analyzed=len(contracts_data),
anomalies_found=len(anomalies),
duration_seconds=investigation_duration,
)
return AgentResponse(
agent_name=self.name,
status=AgentStatus.COMPLETED,
result=result,
metadata={"investigation_id": context.investigation_id}
)
except Exception as e:
# Record investigation failure
INVESTIGATIONS_TOTAL.labels(
agent_type="zumbi",
investigation_type="general", # Fallback for failed investigations
status="failed"
).inc()
self.logger.error(
"investigation_failed",
investigation_id=context.investigation_id,
error=str(e),
agent_name=self.name,
)
return AgentResponse(
agent_name=self.name,
status=AgentStatus.ERROR,
error=str(e),
result={
"status": "error",
"error": str(e),
"investigation_id": context.investigation_id,
},
metadata={"investigation_id": context.investigation_id}
)
async def _fetch_investigation_data(
self,
request: InvestigationRequest,
context: AgentContext
) -> List[Dict[str, Any]]:
"""
Fetch data from Portal da Transparência for investigation.
Args:
request: Investigation parameters
context: Agent context
Returns:
List of contract records for analysis
"""
all_contracts = []
# Default organization codes if not specified
org_codes = request.organization_codes or ["26000", "20000", "25000"] # Health, Presidency, Education
async with TransparencyAPIClient() as client:
for org_code in org_codes:
try:
# Create filters for this organization
filters = TransparencyAPIFilter(
codigo_orgao=org_code,
ano=2024, # Current year
pagina=1,
tamanho_pagina=min(request.max_records // len(org_codes), 50)
)
# Add date range if specified
if request.date_range:
filters.data_inicio = request.date_range[0]
filters.data_fim = request.date_range[1]
# Add value threshold if specified
if request.value_threshold:
filters.valor_inicial = request.value_threshold
# Fetch contracts
response = await client.get_contracts(filters)
# Record API data fetched
TRANSPARENCY_API_DATA_FETCHED.labels(
endpoint="contracts",
organization=org_code,
status="success"
).inc(len(response.data))
# Add organization code to each contract
for contract in response.data:
contract["_org_code"] = org_code
all_contracts.extend(response.data)
self.logger.info(
"data_fetched",
org_code=org_code,
records=len(response.data),
investigation_id=context.investigation_id,
)
except Exception as e:
# Record API fetch failure
TRANSPARENCY_API_DATA_FETCHED.labels(
endpoint="contracts",
organization=org_code,
status="failed"
).inc()
self.logger.warning(
"data_fetch_failed",
org_code=org_code,
error=str(e),
investigation_id=context.investigation_id,
)
continue
return all_contracts[:request.max_records]
async def _enrich_with_open_data(
self,
contracts_data: List[Dict[str, Any]],
context: AgentContext
) -> List[Dict[str, Any]]:
"""
Enrich contract data with information from dados.gov.br.
Args:
contracts_data: Contract records
context: Agent context
Returns:
Enriched contract data
"""
# Extract unique organizations from contracts
organizations = set()
for contract in contracts_data:
org_name = contract.get("orgao", {}).get("nome", "")
if org_name:
organizations.add(org_name)
# Search for related datasets for each organization
related_datasets = {}
for org_name in organizations:
try:
# Search for datasets from this organization
result = await self.dados_gov_tool._execute(
query=f"{org_name}, licitações, contratos",
action="search",
limit=5
)
if result.success and result.data:
related_datasets[org_name] = result.data.get("datasets", [])
self.logger.info(
"open_data_found",
organization=org_name,
datasets_count=len(related_datasets[org_name]),
investigation_id=context.investigation_id,
)
except Exception as e:
self.logger.warning(
"open_data_search_failed",
organization=org_name,
error=str(e),
investigation_id=context.investigation_id,
)
# Enrich contracts with open data references
for contract in contracts_data:
org_name = contract.get("orgao", {}).get("nome", "")
if org_name in related_datasets:
contract["_open_data_available"] = True
contract["_related_datasets"] = related_datasets[org_name]
else:
contract["_open_data_available"] = False
contract["_related_datasets"] = []
return contracts_data
async def _run_anomaly_detection(
self,
contracts_data: List[Dict[str, Any]],
request: InvestigationRequest,
context: AgentContext
) -> List[AnomalyResult]:
"""
Run all anomaly detection algorithms on the contract data.
Args:
contracts_data: Contract records to analyze
request: Investigation parameters
context: Agent context
Returns:
List of detected anomalies
"""
all_anomalies = []
# Determine which anomaly types to run
types_to_run = request.anomaly_types or list(self.anomaly_detectors.keys())
for anomaly_type in types_to_run:
if anomaly_type in self.anomaly_detectors:
try:
detector = self.anomaly_detectors[anomaly_type]
anomalies = await detector(contracts_data, context)
all_anomalies.extend(anomalies)
self.logger.info(
"anomaly_detection_completed",
type=anomaly_type,
anomalies_found=len(anomalies),
investigation_id=context.investigation_id,
)
except Exception as e:
self.logger.error(
"anomaly_detection_failed",
type=anomaly_type,
error=str(e),
investigation_id=context.investigation_id,
)
# Sort anomalies by severity (descending)
all_anomalies.sort(key=lambda x: x.severity, reverse=True)
return all_anomalies
async def _detect_price_anomalies(
self,
contracts_data: List[Dict[str, Any]],
context: AgentContext
) -> List[AnomalyResult]:
"""
Detect contracts with anomalous pricing.
Args:
contracts_data: Contract records
context: Agent context
Returns:
List of price anomalies
"""
anomalies = []
# Extract contract values
values = []
valid_contracts = []
for contract in contracts_data:
valor = contract.get("valorInicial") or contract.get("valorGlobal")
if valor and isinstance(valor, (int, float)) and valor > 0:
values.append(float(valor))
valid_contracts.append(contract)
if len(values) < 10: # Need minimum samples for statistical analysis
return anomalies
# Calculate statistical measures
values_array = np.array(values)
mean_value = np.mean(values_array)
std_value = np.std(values_array)
# Detect outliers using z-score
z_scores = np.abs((values_array - mean_value) / std_value)
for i, (contract, value, z_score) in enumerate(zip(valid_contracts, values, z_scores)):
if z_score > self.price_threshold:
severity = min(z_score / 5.0, 1.0) # Normalize to 0-1
confidence = min(z_score / 3.0, 1.0)
anomaly = AnomalyResult(
anomaly_type="price_anomaly",
severity=severity,
confidence=confidence,
description=f"Contrato com valor suspeito: R$ {value:,.2f}",
explanation=(
f"O valor deste contrato está {z_score:.1f} desvios padrão acima da média "
f"(R$ {mean_value:,.2f}). Valores muito acima do padrão podem indicar "
f"superfaturamento ou irregularidades no processo licitatório."
),
evidence={
"contract_value": value,
"mean_value": mean_value,
"std_deviation": std_value,
"z_score": z_score,
"percentile": np.percentile(values_array, 95),
},
recommendations=[
"Investigar justificativas para o valor elevado",
"Comparar com contratos similares de outros órgãos",
"Verificar processo licitatório e documentação",
"Analisar histórico do fornecedor",
],
affected_entities=[{
"contract_id": contract.get("id"),
"object": contract.get("objeto", "")[:100],
"supplier": contract.get("fornecedor", {}).get("nome", "N/A"),
"organization": contract.get("_org_code"),
}],
financial_impact=value - mean_value,
)
anomalies.append(anomaly)
return anomalies
async def _detect_vendor_concentration(
self,
contracts_data: List[Dict[str, Any]],
context: AgentContext
) -> List[AnomalyResult]:
"""
Detect excessive vendor concentration (potential monopolization).
Args:
contracts_data: Contract records
context: Agent context
Returns:
List of vendor concentration anomalies
"""
anomalies = []
# Group contracts by vendor
vendor_stats = {}
total_value = 0
for contract in contracts_data:
supplier = contract.get("fornecedor", {})
vendor_name = supplier.get("nome", "Unknown")
vendor_cnpj = supplier.get("cnpj", "Unknown")
vendor_key = f"{vendor_name}|{vendor_cnpj}"
valor = contract.get("valorInicial") or contract.get("valorGlobal") or 0
if isinstance(valor, (int, float)):
valor = float(valor)
total_value += valor
if vendor_key not in vendor_stats:
vendor_stats[vendor_key] = {
"name": vendor_name,
"cnpj": vendor_cnpj,
"contracts": [],
"total_value": 0,
"contract_count": 0,
}
vendor_stats[vendor_key]["contracts"].append(contract)
vendor_stats[vendor_key]["total_value"] += valor
vendor_stats[vendor_key]["contract_count"] += 1
if total_value == 0:
return anomalies
# Check for concentration anomalies
for vendor_key, stats in vendor_stats.items():
concentration = stats["total_value"] / total_value
if concentration > self.concentration_threshold:
severity = min(concentration * 1.5, 1.0)
confidence = concentration
anomaly = AnomalyResult(
anomaly_type="vendor_concentration",
severity=severity,
confidence=confidence,
description=f"Concentração excessiva de contratos: {stats['name']}",
explanation=(
f"O fornecedor {stats['name']} concentra {concentration:.1%} do valor total "
f"dos contratos analisados ({stats['contract_count']} contratos). "
f"Alta concentração pode indicar direcionamento de licitações ou "
f"falta de competitividade no processo."
),
evidence={
"vendor_name": stats["name"],
"vendor_cnpj": stats["cnpj"],
"concentration_percentage": concentration * 100,
"total_value": stats["total_value"],
"contract_count": stats["contract_count"],
"market_share": concentration,
},
recommendations=[
"Verificar se houve direcionamento nas licitações",
"Analisar competitividade do mercado",
"Investigar relacionamento entre órgão e fornecedor",
"Revisar critérios de seleção de fornecedores",
],
affected_entities=[{
"vendor_name": stats["name"],
"vendor_cnpj": stats["cnpj"],
"contract_count": stats["contract_count"],
"total_value": stats["total_value"],
}],
financial_impact=stats["total_value"],
)
anomalies.append(anomaly)
return anomalies
async def _detect_temporal_anomalies(
self,
contracts_data: List[Dict[str, Any]],
context: AgentContext
) -> List[AnomalyResult]:
"""
Detect suspicious temporal patterns in contracts.
Args:
contracts_data: Contract records
context: Agent context
Returns:
List of temporal anomalies
"""
anomalies = []
# Group contracts by date
date_stats = {}
for contract in contracts_data:
# Try to extract date from different fields
date_str = (
contract.get("dataAssinatura") or
contract.get("dataPublicacao") or
contract.get("dataInicio")
)
if date_str:
try:
# Parse date (assuming DD/MM/YYYY format)
date_parts = date_str.split("/")
if len(date_parts) == 3:
day = int(date_parts[0])
month = int(date_parts[1])
year = int(date_parts[2])
date_key = f"{year}-{month:02d}"
if date_key not in date_stats:
date_stats[date_key] = {
"contracts": [],
"count": 0,
"total_value": 0,
}
valor = contract.get("valorInicial") or contract.get("valorGlobal") or 0
if isinstance(valor, (int, float)):
date_stats[date_key]["total_value"] += float(valor)
date_stats[date_key]["contracts"].append(contract)
date_stats[date_key]["count"] += 1
except (ValueError, IndexError):
continue
if len(date_stats) < 3: # Need minimum periods for comparison
return anomalies
# Calculate average contracts per period
counts = [stats["count"] for stats in date_stats.values()]
mean_count = np.mean(counts)
std_count = np.std(counts)
# Look for periods with unusually high activity
for date_key, stats in date_stats.items():
if std_count > 0:
z_score = (stats["count"] - mean_count) / std_count
if z_score > 2.0: # More than 2 standard deviations
severity = min(z_score / 4.0, 1.0)
confidence = min(z_score / 3.0, 1.0)
anomaly = AnomalyResult(
anomaly_type="temporal_patterns",
severity=severity,
confidence=confidence,
description=f"Atividade contratual suspeita em {date_key}",
explanation=(
f"Em {date_key} foram assinados {stats['count']} contratos, "
f"{z_score:.1f} desvios padrão acima da média ({mean_count:.1f}). "
f"Picos de atividade podem indicar direcionamento ou urgência "
f"inadequada nos processos."
),
evidence={
"period": date_key,
"contract_count": stats["count"],
"mean_count": mean_count,
"z_score": z_score,
"total_value": stats["total_value"],
},
recommendations=[
"Investigar justificativas para a concentração temporal",
"Verificar se houve emergência ou urgência",
"Analisar qualidade dos processos licitatórios",
"Revisar planejamento de contratações",
],
affected_entities=[{
"period": date_key,
"contract_count": stats["count"],
"total_value": stats["total_value"],
}],
financial_impact=stats["total_value"],
)
anomalies.append(anomaly)
return anomalies
async def _detect_duplicate_contracts(
self,
contracts_data: List[Dict[str, Any]],
context: AgentContext
) -> List[AnomalyResult]:
"""
Detect potentially duplicate or very similar contracts.
Args:
contracts_data: Contract records
context: Agent context
Returns:
List of duplicate contract anomalies
"""
anomalies = []
# Simple similarity detection based on object description
for i, contract1 in enumerate(contracts_data):
objeto1 = contract1.get("objeto", "").lower()
if len(objeto1) < 20: # Skip very short descriptions
continue
for j, contract2 in enumerate(contracts_data[i+1:], start=i+1):
objeto2 = contract2.get("objeto", "").lower()
if len(objeto2) < 20:
continue
# Calculate simple similarity (Jaccard similarity of words)
words1 = set(objeto1.split())
words2 = set(objeto2.split())
if len(words1) == 0 or len(words2) == 0:
continue
intersection = len(words1.intersection(words2))
union = len(words1.union(words2))
similarity = intersection / union if union > 0 else 0
if similarity > self.duplicate_threshold:
severity = similarity
confidence = similarity
valor1 = contract1.get("valorInicial") or contract1.get("valorGlobal") or 0
valor2 = contract2.get("valorInicial") or contract2.get("valorGlobal") or 0
anomaly = AnomalyResult(
anomaly_type="duplicate_contracts",
severity=severity,
confidence=confidence,
description="Contratos potencialmente duplicados detectados",
explanation=(
f"Dois contratos com {similarity:.1%} de similaridade foram "
f"encontrados. Contratos similares podem indicar pagamentos "
f"duplicados ou direcionamento inadequado."
),
evidence={
"similarity_score": similarity,
"contract1_id": contract1.get("id"),
"contract2_id": contract2.get("id"),
"contract1_value": valor1,
"contract2_value": valor2,
"object1": objeto1[:100],
"object2": objeto2[:100],
},
recommendations=[
"Verificar se são contratos distintos ou duplicados",
"Analisar justificativas para objetos similares",
"Investigar fornecedores envolvidos",
"Revisar controles internos de contratação",
],
affected_entities=[
{
"contract_id": contract1.get("id"),
"object": objeto1[:100],
"value": valor1,
},
{
"contract_id": contract2.get("id"),
"object": objeto2[:100],
"value": valor2,
},
],
financial_impact=float(valor1) + float(valor2) if isinstance(valor1, (int, float)) and isinstance(valor2, (int, float)) else None,
)
anomalies.append(anomaly)
return anomalies
async def _detect_payment_anomalies(
self,
contracts_data: List[Dict[str, Any]],
context: AgentContext
) -> List[AnomalyResult]:
"""
Detect unusual payment patterns in contracts.
Args:
contracts_data: Contract records
context: Agent context
Returns:
List of payment anomalies
"""
anomalies = []
# Look for contracts with unusual value patterns
for contract in contracts_data:
valor_inicial = contract.get("valorInicial")
valor_global = contract.get("valorGlobal")
if valor_inicial and valor_global:
try:
inicial = float(valor_inicial)
global_val = float(valor_global)
# Check for significant discrepancies
if inicial > 0 and global_val > 0:
ratio = abs(inicial - global_val) / max(inicial, global_val)
if ratio > 0.5: # 50% discrepancy threshold
severity = min(ratio, 1.0)
confidence = ratio
anomaly = AnomalyResult(
anomaly_type="payment_patterns",
severity=severity,
confidence=confidence,
description="Discrepância significativa entre valores do contrato",
explanation=(
f"Diferença de {ratio:.1%} entre valor inicial "
f"(R$ {inicial:,.2f}) e valor global (R$ {global_val:,.2f}). "
f"Grandes discrepâncias podem indicar aditivos excessivos "
f"ou irregularidades nos pagamentos."
),
evidence={
"valor_inicial": inicial,
"valor_global": global_val,
"discrepancy_ratio": ratio,
"absolute_difference": abs(inicial - global_val),
},
recommendations=[
"Investigar justificativas para alterações de valor",
"Verificar aditivos contratuais",
"Analisar execução e pagamentos realizados",
"Revisar controles de alteração contratual",
],
affected_entities=[{
"contract_id": contract.get("id"),
"object": contract.get("objeto", "")[:100],
"supplier": contract.get("fornecedor", {}).get("nome", "N/A"),
}],
financial_impact=abs(inicial - global_val),
)
anomalies.append(anomaly)
except (ValueError, TypeError):
continue
return anomalies
async def _detect_spectral_anomalies(
self,
contracts_data: List[Dict[str, Any]],
context: AgentContext
) -> List[AnomalyResult]:
"""
Detect anomalies using spectral analysis and Fourier transforms.
Args:
contracts_data: Contract records
context: Agent context
Returns:
List of spectral anomalies
"""
anomalies = []
try:
# Prepare time series data
time_series_data = self._prepare_time_series(contracts_data)
if len(time_series_data) < 30: # Need sufficient data points
self.logger.warning("insufficient_data_for_spectral_analysis", data_points=len(time_series_data))
return anomalies
# Extract spending values and timestamps
spending_data = pd.Series([item['value'] for item in time_series_data])
timestamps = pd.DatetimeIndex([item['date'] for item in time_series_data])
# Perform spectral anomaly detection
spectral_anomalies = self.spectral_analyzer.detect_anomalies(
spending_data,
timestamps,
context={'entity_name': context.investigation_id if hasattr(context, 'investigation_id') else 'Unknown'}
)
# Convert SpectralAnomaly objects to AnomalyResult objects
for spec_anomaly in spectral_anomalies:
anomaly = AnomalyResult(
anomaly_type=f"spectral_{spec_anomaly.anomaly_type}",
severity=spec_anomaly.anomaly_score,
confidence=spec_anomaly.anomaly_score,
description=spec_anomaly.description,
explanation=self._create_spectral_explanation(spec_anomaly),
evidence={
"frequency_band": spec_anomaly.frequency_band,
"anomaly_score": spec_anomaly.anomaly_score,
"timestamp": spec_anomaly.timestamp.isoformat(),
**spec_anomaly.evidence
},
recommendations=spec_anomaly.recommendations,
affected_entities=self._extract_affected_entities_from_spectral(spec_anomaly, contracts_data),
financial_impact=self._calculate_spectral_financial_impact(spec_anomaly, spending_data)
)
anomalies.append(anomaly)
# Find periodic patterns
periodic_patterns = self.spectral_analyzer.find_periodic_patterns(
spending_data,
timestamps,
entity_name=context.investigation_id if hasattr(context, 'investigation_id') else None
)
# Convert suspicious periodic patterns to anomalies
for pattern in periodic_patterns:
if pattern.pattern_type == "suspicious" or pattern.amplitude > 0.5:
anomaly = AnomalyResult(
anomaly_type="suspicious_periodic_pattern",
severity=pattern.amplitude,
confidence=pattern.confidence,
description=f"Padrão periódico suspeito detectado (período: {pattern.period_days:.1f} dias)",
explanation=(
f"Detectado padrão de gastos com periodicidade de {pattern.period_days:.1f} dias "
f"e amplitude de {pattern.amplitude:.1%}. {pattern.business_interpretation}"
),
evidence={
"period_days": pattern.period_days,
"frequency_hz": pattern.frequency_hz,
"amplitude": pattern.amplitude,
"confidence": pattern.confidence,
"pattern_type": pattern.pattern_type,
"statistical_significance": pattern.statistical_significance
},
recommendations=[
"Investigar causa do padrão periódico",
"Verificar se há processos automatizados",
"Analisar justificativas para regularidade excessiva",
"Revisar cronograma de pagamentos"
],
affected_entities=[{
"pattern_type": pattern.pattern_type,
"period_days": pattern.period_days,
"amplitude": pattern.amplitude
}],
financial_impact=float(spending_data.sum() * pattern.amplitude)
)
anomalies.append(anomaly)
self.logger.info(
"spectral_analysis_completed",
spectral_anomalies_count=len(spectral_anomalies),
periodic_patterns_count=len(periodic_patterns),
total_anomalies=len(anomalies)
)
except Exception as e:
self.logger.error(f"Error in spectral anomaly detection: {str(e)}")
# Don't fail the entire investigation if spectral analysis fails
return anomalies
def _prepare_time_series(self, contracts_data: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
"""Prepare time series data from contracts for spectral analysis."""
time_series = []
for contract in contracts_data:
# Extract date
date_str = (
contract.get("dataAssinatura") or
contract.get("dataPublicacao") or
contract.get("dataInicio")
)
if not date_str:
continue
try:
# Parse date (DD/MM/YYYY format)
date_parts = date_str.split("/")
if len(date_parts) == 3:
day, month, year = int(date_parts[0]), int(date_parts[1]), int(date_parts[2])
date_obj = datetime(year, month, day)
# Extract value
valor = contract.get("valorInicial") or contract.get("valorGlobal") or 0
if isinstance(valor, (int, float)) and valor > 0:
time_series.append({
'date': date_obj,
'value': float(valor),
'contract_id': contract.get('id'),
'supplier': contract.get('fornecedor', {}).get('nome', 'N/A')
})
except (ValueError, IndexError):
continue
# Sort by date
time_series.sort(key=lambda x: x['date'])
# Aggregate by date (sum values for same dates)
daily_aggregates = {}
for item in time_series:
date_key = item['date'].date()
if date_key not in daily_aggregates:
daily_aggregates[date_key] = {
'date': datetime.combine(date_key, datetime.min.time()),
'value': 0,
'contract_count': 0,
'suppliers': set()
}
daily_aggregates[date_key]['value'] += item['value']
daily_aggregates[date_key]['contract_count'] += 1
daily_aggregates[date_key]['suppliers'].add(item['supplier'])
# Convert back to list
aggregated_series = []
for date_key in sorted(daily_aggregates.keys()):
data = daily_aggregates[date_key]
aggregated_series.append({
'date': data['date'],
'value': data['value'],
'contract_count': data['contract_count'],
'unique_suppliers': len(data['suppliers'])
})
return aggregated_series
def _create_spectral_explanation(self, spec_anomaly: SpectralAnomaly) -> str:
"""Create detailed explanation for spectral anomaly."""
explanations = {
"high_frequency_pattern": (
"Detectado padrão de alta frequência nos gastos públicos. "
"Padrões muito regulares podem indicar manipulação sistemática ou "
"processos automatizados não documentados."
),
"spectral_regime_change": (
"Mudança significativa detectada na complexidade dos padrões de gastos. "
"Alterações bruscas podem indicar mudanças de política, procedimentos "
"ou possível manipulação."
),
"excessive_quarterly_pattern": (
"Padrão excessivo de gastos trimestrais detectado. "
"Concentração de gastos no final de trimestres pode indicar "
"execução inadequada de orçamento ou 'correria' para gastar verbas."
),
"unusual_weekly_regularity": (
"Regularidade semanal incomum detectada nos gastos. "
"Padrões muito regulares em gastos governamentais podem ser suspeitos "
"se não corresponderem a processos de negócio conhecidos."
),
"high_frequency_noise": (
"Ruído de alta frequência detectado nos dados de gastos. "
"Pode indicar problemas na coleta de dados ou manipulação artificial "
"dos valores reportados."
)
}
base_explanation = explanations.get(
spec_anomaly.anomaly_type,
f"Anomalia espectral detectada: {spec_anomaly.description}"
)
return f"{base_explanation} Score de anomalia: {spec_anomaly.anomaly_score:.2f}. {spec_anomaly.description}"
def _extract_affected_entities_from_spectral(
self,
spec_anomaly: SpectralAnomaly,
contracts_data: List[Dict[str, Any]]
) -> List[Dict[str, Any]]:
"""Extract affected entities from spectral anomaly context."""
affected = []
# For temporal anomalies, find contracts around the anomaly timestamp
if hasattr(spec_anomaly, 'timestamp') and spec_anomaly.timestamp:
anomaly_date = spec_anomaly.timestamp.date()
for contract in contracts_data:
date_str = (
contract.get("dataAssinatura") or
contract.get("dataPublicacao") or
contract.get("dataInicio")
)
if date_str:
try:
date_parts = date_str.split("/")
if len(date_parts) == 3:
day, month, year = int(date_parts[0]), int(date_parts[1]), int(date_parts[2])
contract_date = datetime(year, month, day).date()
# Include contracts within a week of the anomaly
if abs((contract_date - anomaly_date).days) <= 7:
affected.append({
"contract_id": contract.get("id"),
"date": date_str,
"supplier": contract.get("fornecedor", {}).get("nome", "N/A"),
"value": contract.get("valorInicial") or contract.get("valorGlobal") or 0,
"object": contract.get("objeto", "")[:100]
})
except (ValueError, IndexError):
continue
return affected[:10] # Limit to first 10 to avoid overwhelming
def _calculate_spectral_financial_impact(
self,
spec_anomaly: SpectralAnomaly,
spending_data: pd.Series
) -> Optional[float]:
"""Calculate financial impact of spectral anomaly."""
try:
# For high-amplitude anomalies, estimate impact as percentage of total spending
if hasattr(spec_anomaly, 'anomaly_score') and spec_anomaly.anomaly_score > 0:
total_spending = float(spending_data.sum())
impact_ratio = min(spec_anomaly.anomaly_score, 0.5) # Cap at 50%
return total_spending * impact_ratio
except:
pass
return None
def _generate_investigation_summary(
self,
contracts_data: List[Dict[str, Any]],
anomalies: List[AnomalyResult]
) -> Dict[str, Any]:
"""Generate summary statistics for the investigation."""
total_value = 0
suspicious_value = 0
# Calculate total contract value
for contract in contracts_data:
valor = contract.get("valorInicial") or contract.get("valorGlobal") or 0
if isinstance(valor, (int, float)):
total_value += float(valor)
# Calculate suspicious value
for anomaly in anomalies:
if anomaly.financial_impact:
suspicious_value += anomaly.financial_impact
# Group anomalies by type
anomaly_counts = {}
for anomaly in anomalies:
anomaly_type = anomaly.anomaly_type
anomaly_counts[anomaly_type] = anomaly_counts.get(anomaly_type, 0) + 1
# Calculate risk score
risk_score = min(len(anomalies) / max(len(contracts_data), 1) * 10, 10)
return {
"total_records": len(contracts_data),
"anomalies_found": len(anomalies),
"total_value": total_value,
"suspicious_value": suspicious_value,
"risk_score": risk_score,
"anomaly_types": anomaly_counts,
"high_severity_count": len([a for a in anomalies if a.severity > 0.7]),
"medium_severity_count": len([a for a in anomalies if 0.3 < a.severity <= 0.7]),
"low_severity_count": len([a for a in anomalies if a.severity <= 0.3]),
}
def _anomaly_to_dict(self, anomaly: AnomalyResult) -> Dict[str, Any]:
"""Convert AnomalyResult to dictionary for serialization."""
return {
"type": anomaly.anomaly_type,
"severity": anomaly.severity,
"confidence": anomaly.confidence,
"description": anomaly.description,
"explanation": anomaly.explanation,
"evidence": anomaly.evidence,
"recommendations": anomaly.recommendations,
"affected_entities": anomaly.affected_entities,
"financial_impact": anomaly.financial_impact,
} |