TracewiseCase ConsoleFraud Investigation Agent
all cases

HHG-019

legitimateClosed — Legitimate

no pattern identified

Case HHG-019 was reviewed after a real-time model scored transaction 3503878 ($99.92, online) at 0.90, but the final verdict is legitimate with a fraud probability of 0.12. Key evidence indicates the transaction originated from a device marked 'New' (id_15=New), yet the product category and billing region align with the customer's established history, and disconfirming precedents CC-0678 and CC-0388 support a legitimate classification. The agent's request to expand the investigation to the shared-device neighborhood using device_neighbors(device_id=device_id_15) was refused because the parameter appeared to be a raw dataset column name rather than a valid entity ID. Consequently, the final action remains CLOSE_NO_FRAUD (auto, R0: policy default, settled low, no rule triggered), as the initial assessment was upheld without changes.

exposure
$0.00
affected txns
0
connected cards
0
prior cases cited
5
likely legitimate
fraud probability
0.150.300.700.85

stop_reason — Probability 0.12 with at least two independent evidence sources supporting no fraud; stopping per policy §6.

recommendation historyunchanged

Initial recommendation

before · on the evidence gathered so far
  1. CLOSE_NO_FRAUDAutoR0: policy default, settled low, no rule triggered

Final recommendation

after · no revision required
  1. CLOSE_NO_FRAUDAutoR0: policy default, settled low, no rule triggered

The evidence settled the question without changing the recommended course of action.

entity neighbourhood1 cards · 0 devices
HHG-019-CARD
shared devicecard under investigationflaggedconnected card
policy trace · R1–R100 of 10 predicates fired

Every rule below is a deterministic predicate evaluated against this case’s evidence — not a model’s assertion of an outcome. The engine checks all ten; most do not apply here, and that is shown deliberately.

R1
did not fire
Single signal only, fraud probability below 0.70 — must precede any block.
R2
did not fire
Customer was asked to validate and denied the charge.
R3
did not fire
Customer confirmed the charge as their own.
R4
did not fire
No customer reply within 24 hours.
R5
did not fire
3+ sub-$5 online authorizations within an hour on one card, then a larger purchase.
R6
did not fire
Device profile, billing region, or recipient email shared across multiple cards in one window.
R7
did not fire
Disputed charge matches the customer's own recurring pattern — merchant, amount, monthly.
R8
did not fire
Verdict uncertain with exposure over $500, or evidence conflicts.
R9
did not fire
Fits none of the five documented patterns, and abuse is coordinated or repeated across customers.
R10
did not fire
BLOCK_ALL_CARDS was proposed — requires 2+ cards confirmed fraud or confirmed compromised credentials, else downgrades to a single-card BLOCK_CARD.
precedent retrieval · graphRAG5 cases retrieved

The closed-case history runs 4,665 confirmed fraud to 900 cleared — a single merged ranking over that history returns almost nothing but incriminating cases and pushes every verdict toward blocking. Retrieval runs as two separate pools, confirming and exonerating, so that imbalance cannot silently decide the case.

Confirming precedent(3)

  • CC-0783
  • CC-0278
  • CC-0031

Exonerating precedent(2)

  • CC-0678
  • CC-0388

Drawn from 5,565 closed investigations.

evidence ledger · 11 itemsprovenance & signals

  • 1 online transaction(s) on card C07987-K2 -- T3503878 ($99.92) -- came from a device first marked 'New' for this account (id_15=New)

    Graph
    query:card_window(card_id=C07987-K2)T3503878

  • Precedent (confirming): CC-0783 (confirmed fraud, $99.92 exposure): semantically similar analyst notes (cosine distance 0.28), no direct structural overlap

    Graph
    query:prior_case_candidates(case=CC-0783)CC-0783

  • Precedent (confirming): CC-0278 (confirmed fraud, $50.07 exposure): semantically similar analyst notes (cosine distance 0.29), no direct structural overlap

    Graph
    query:prior_case_candidates(case=CC-0278)CC-0278

  • Precedent (confirming): CC-0031 (confirmed fraud, $20.36 exposure): semantically similar analyst notes (cosine distance 0.29), no direct structural overlap

    Graph
    query:prior_case_candidates(case=CC-0031)CC-0031

  • Precedent (disconfirming): CC-0678 (cleared as legitimate): semantically similar analyst notes (cosine distance 0.22), no direct structural overlap

    Graph
    query:prior_case_candidates(case=CC-0678)CC-0678

  • Precedent (disconfirming): CC-0388 (cleared as legitimate): semantically similar analyst notes (cosine distance 0.22), no direct structural overlap

    Graph
    query:prior_case_candidates(case=CC-0388)CC-0388

  • Flagged transaction's product category and billing region match this customer's established history

    Graph
    query:customer_baseline(customer_id=C07987)T3503878

  • Agent elected to expand the investigation to the shared-device neighbourhood. Reasoning: YES The new device marker combined with in-character activity suggests a potential shared device compromise that could extend to other cards.

    Graph
    llm_decision:expand_device_neighbourhood(case=HHG-019)C07987-K2

  • Agent proposed device_neighbors(device_id=device_id_15), refused: that looks like a raw dataset column name, not a real entity id -- running it would waste a query rather than gather evidence. Full params: {'device_id': 'device_id_15', 'anchor': '2016-12-01 22:28:53', 'hours': 720}

    Graph
    llm_plan:rejected(column_as_id, depth=1)

  • Agent proposed device_neighbors(device_id=device_id_15), refused: that looks like a raw dataset column name, not a real entity id -- running it would waste a query rather than gather evidence. Full params: {'device_id': 'device_id_15', 'anchor': '2016-12-01 22:28:53', 'hours': 720}

    Graph
    llm_plan:rejected(column_as_id, depth=2)

  • Confidence check on gathered evidence: medium -- The investigation has reached its maximum depth limit (3 of 3). While the device neighborhood expansion was blocked due to invalid entity IDs, the existing evidence provides a defensible basis for the policy engine to make a final determination. The signals are mixed: a high real-time risk score (0.90) and a new device marker suggest potential fraud, but the transaction aligns with the customer's established baseline (product/region) and there are disconfirming precedents. This creates a 'medium' confidence scenario where the evidence is sufficient to hand off to the deterministic policy engine, which is designed to weigh these conflicting signals.

    Graph
    llm_decision:confidence_gate(case=HHG-019, depth=3)
prior cases · memory5 cited

Confirming precedent

  • CC-0783Supports fraud
  • CC-0278Supports fraud
  • CC-0031Supports fraud

Disconfirming precedent

  • CC-0678Exonerates
  • CC-0388Exonerates
Stop reasonProbability 0.12 with at least two independent evidence sources supporting no fraud; stopping per policy §6.
Tool calls0
Tokens0
Latency0.0s