Anomaly Detection Audit Trail

The problem

Finance teams deploy AI agents to monitor thousands of transactions per day for anomalies: duplicate payments, unusual vendor activity, transactions at suspicious times. But when an auditor asks how the organization governs these agents (what they flag, why they flag it, how accurate they are, and who reviews their work), most teams have no structured answer.

Thirteen anomaly subtypes

Sigmodx records the anomaly subtype the agent detected, including duplicate payments (same vendor, similar amount, close dates), revenue reversals after period close, velocity anomalies (unusual transaction frequency), and split transactions designed to avoid approval thresholds. It never stores raw transaction data.

Reliability signals

Five signals are computed from human review: false positive rate, false negative rate, detection precision, escalation rate, and severity accuracy. False negative rate carries the most weight. An agent that misses real anomalies moves to LIMIT above 2% and BLOCK above 5%.

Integration

pip install sigmodx

from sigmodx import SigmodxClient

client = SigmodxClient(api_key="...", agent_id="...")

input_hash = client.hash_inputs({
    "txn_ref": "TXN-2026-4421",
    "amount": 14250,
})

result = client.submit_anomaly_decision(
    decision_type="flag",
    input_hash=input_hash,
    rationale="Transaction matches prior payment within 0.4%.",
    anomaly_subtype="duplicate_payment",
    severity="high",
    transaction_amount=14250,
)

pip install sigmodx sigmodx-integrations

from sigmodx import SigmodxClient
from sigmodx_integrations import SigmodxCallbackHandler, ANOMALY_DETECTION_CONFIG

client = SigmodxClient(api_key="...", agent_id="...")
handler = SigmodxCallbackHandler(
    client=client,
    config=ANOMALY_DETECTION_CONFIG,
)

# Every tool call is automatically logged to Sigmodx
agent_executor.invoke(
    {"input": "Monitor transaction TXN-2026-4421"},
    config={"callbacks": [handler]},
)

pip install sigmodx-mcp

SIGMODX_API_KEY=your-key \
SIGMODX_AGENT_ID=your-agent-uuid \
sigmodx-mcp

{
  "mcpServers": {
    "sigmodx": {
      "command": "sigmodx-mcp",
      "env": {
        "SIGMODX_API_KEY": "your-api-key",
        "SIGMODX_AGENT_ID": "your-agent-uuid"
      }
    }
  }
}