Free tools Windows power users keep installed
One-click scans. No signup required.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
JEVelric is a fraud-investigation project built on TigerGraph for TigerGraph’s HHGOA challenge. Given a fraud trigger, it retrieves transaction, identity, device, and prior-case evidence from a graph, assesses risk, requests more evidence when the picture is unclear, and then recommends an action. Its central design choice is that model-generated assessment signals do not select the action. A deterministic policy engine applies rules R1–R10 and attaches an approval route to each recommendation.
Read it as a project implementation and benchmark submission. It is not a proven production fraud-detection product, and the available evidence does not support treating it as one.
How an investigation runs
The project’s documentation describes a fixed sequence. Each stage has a distinct job, and the model’s output only feeds the stages that come before the policy step.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minute- Intake trigger. A fraud trigger starts the investigation. The project exposes a live investigation endpoint and an MCP tool for this purpose, according to the author’s overview.
- Graph evidence retrieval. The system runs GSQL queries against TigerGraph to collect transaction, identity, device, and prior-case evidence.
- Context assembly. The document context assembler combines local policy and pattern files, graph evidence, similar closed cases, and graph-derived hints.
- Risk and pattern assessment. The model produces assessment signals from the assembled context.
- Deterministic policy application. Rules R1–R10 are applied by the policy engine, not by the model.
- Evidence decision. If uncertainty remains, the system gathers more evidence. The documented limit is two evidence-gathering rounds. Simulated customer replies are disclosed as assumptions.
- Final action and approval routing. The recommendation is routed as automatic, team-lead approval, or fraud-manager approval.
- SAR-policy evaluation. The system evaluates the case against suspicious-activity-report policy.
- Case creation and graph write. An investigation case is created and written back to the graph.
- Output validation. The output is checked for structure and for graph-backed identifiers.
The graph model
TigerGraph stores both the linked entities behind a transaction and the system’s investigation memory. The project’s README names eight vertex types:
#1 Best Overall
| Vertex type | What it holds in JEVelric |
|---|---|
| Customer | The customer whose history is checked |
| Card | The payment card under investigation or linked to it |
| Transaction | Individual card transactions (about 590,000 in the HHGOA graph) |
| DeviceProfile | Device profiles used in transactions |
| EmailDomain | Email domains associated with customers or cards |
| BillingRegion | Billing regions associated with transactions |
| ClosedCase | Prior closed investigations |
| InvestigationCase | Investigation cases the system generates |
The README counts 13 edge types connecting these vertices. This article does not list them individually. When an investigation is resolved, the author describes writing it back as case memory linked to its entities and to prior cases, so later investigations can draw on earlier ones.
The six retrieval queries
The author describes six GSQL lookups. Each one targets a specific question that a fraud analyst would normally ask by hand:
- Transaction-window lookup. Pulls a card’s transactions over a defined time window around the trigger.
- Device-neighbor lookup. Finds other cards that share a device profile with the card under review.
- Region-cluster lookup. Examines whether transactions cluster by billing region.
- Email-cluster lookup. Examines whether linked accounts cluster by email domain.
- Closed-case similarity lookup. Compares the current pattern with prior closed investigations.
- Customer-history lookup. Retrieves the customer’s history.
The device-neighbor lookup is the one that answers a question such as “does this card share a device with any other card in the last week?” That is the kind of literal question the graph structure is built to answer: the edge from a card to a device profile makes shared-device checks a single traversal rather than a manual search across tables.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Why the model does not choose the action
The model’s role ends at assessment. It estimates risk, identifies patterns, and decides whether the evidence is sufficient. The policy engine then determines the action from the rules, which keeps each recommendation traceable to a stated rule rather than to a model’s free-text judgment.
The project’s own framing states the limit plainly:
“Risk scores are signals, not fraud verdicts.” — JEVelric project README
Rank #3
Mark Twain Forensic Investigations Workbook, Using Science to Solve High Crimes Middle School Books, Critical Thinking for Kids, DNA and Handwriting Analysis Labs, Classroom or Homeschool Curriculum
- Students build unmatched deductive-reasoning skills as they become crime-solving stars
- Most scenarios have more than one plausible outcome, allowing individuals or groups to broadly interpret evidence
- Includes interpretive handwriting, body language, fingerprinting, and many more activities
Uncertainty controls the loop. When the assessment signals leave a question open, the system can request one of three kinds of follow-up evidence: customer verification, step-up authentication, or input from an analyst. The loop is capped at two rounds, so an unresolved case moves to a routing decision rather than continuing indefinitely.
The numbers and what they measure
The project reports several inventory and test counts. Each measures something specific, and none of them is an accuracy figure.
| Figure | Value | Source and date | What it measures | What it does not show |
|---|---|---|---|---|
| Transactions in the graph | 590,742 | Project README; repository state observed October 7, 2026 | Size of the loaded HHGOA graph | Model accuracy or fraud detection rate |
| Benchmark cases | 20 | Project README; repository state observed October 7, 2026 | Number of held-out cases; all 20 answer files were generated and validated against the live graph | Whether the answers are correct |
| Passing tests | 16 | Project README; repository state observed October 7, 2026 | Test suite results | Semantic quality or calibration |
| Vertex types | 8 | Project README; repository state observed October 7, 2026 | Graph schema size | Data quality |
| Edge types | 13 | Project README; repository state observed October 7, 2026 | Graph schema size | Data quality |
The author’s overview rounds the transaction count to “~590,000 real card transactions.” Use 590,742 when precision matters. The 20 benchmark cases are the author’s own held-out set, not an independent test set.
Rank #4
What validation shows, and what it does not
Validation in the repository is structural. It confirms that the outputs are well formed and that identifiers trace back to the graph. It does not confirm that the recommended actions are right.
- Passed: structural validation and graph-backed ID checks; generation and validation of all 20 answer files against the live graph; 16 passing tests.
- Not yet established: semantic quality and calibration of the investigations, which the README says still need review.
- Not available: a hidden answer key, so outcome accuracy cannot be certified.
- Outside the benchmark path: JEV was not used to generate the case responses or pass benchmark validation. The README states this directly: “JEV was not used to generate the case responses or pass the benchmark validation.”
The repository states the validation boundary in one sentence: “Structural validation and graph-backed ID checks passed, but no hidden answer key is available here to certify outcome accuracy.”
The dataset has no fraud outcome label. The repository cautions against using public IEEE-CIS or Kaggle labels to infer benchmark outcomes. No independently published performance statistic or independent evaluation of JEVelric was found in the sources reviewed for this article.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Implementation status
Several components described in the architecture are not implemented in the repository’s reported state:
- JEV is not wired into the benchmark response path. Its client remains a stub, and the system falls back to an assessment path.
- TigerGraph vector retrieval over a document collection is not implemented. The context assembler uses the local policy and pattern files, graph evidence, similar closed cases, and graph-derived hints instead.
- A dedicated analyst UI is not implemented.
- The latest full run described in the README used NVIDIA NIM as the primary model service and Cloudflare Workers AI as the fallback.
The README lists remaining work: integrating JEV if still required, adding TigerGraph vector search if the submission brief requires it, reviewing investigations for semantic quality and calibration, preparing demo and blog materials, and potentially adding an analyst UI.
Operational lessons from the build
The author’s overview reports three problems encountered while building the system. These are the author’s experience and have not been independently reproduced:
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →- The TigerGraph MCP file-loading tool expects a file path on the TigerGraph server, not on the local machine.
- Loading paths in pyTigerGraph can differ in how they handle headers, so a load that works one way may fail the other.
- An early-access GraphRAG and vector retrieval tool returned a demo-graph response during the author’s evaluation rather than results from the project’s own graph.
Comparing JEVelric with other approaches
JEVelric is not a consumer product, and the available sources contain no independently evaluated competitor comparison. A fair comparison of fraud-investigation approaches should use these axes:
- Graph-based relationship retrieval, versus flat tables or keyword lookups
- Whether model assessment is separated from policy-driven action selection
- How uncertainty is handled and whether the system requests further evidence
- How human approval is routed
- Whether resolved cases persist as memory for later investigations
- Whether a system’s validation is structural or demonstrates outcome accuracy
On these axes, JEVelric is clearly documented for the first, second, third, fourth, and fifth. On the sixth, its own documentation says outcome accuracy has not been certified.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
