In, then quickly out
Find a similar-sized outgoing transfer after an incoming one. Timing alone does not trace the same funds.
MONITOR · INVESTIGATE · CHALLENGE
Less noise.
Know what you might miss.
Help a fictional fintech review its transaction alerts. Adjust a rule, inspect the evidence, and document a decision another analyst can challenge.
Loading the synthetic SQLite database…
01 / RULE LAB
Test the tradeoff
Outside expected activity
Compare the day's outbound value with the customer's expected daily volume.
A crowd of senders
Count distinct counterparties paying one account during the observation day.
One observation day: January 15, 2026 · USD only · fixed baseline: 30 minutes / 3× / 5 senders
How these metrics work
Each of 36 fictional customers has one assigned evaluation scenario: 12 seeded review-worthy cases and 24 seeded benign cases. We score one customer–scenario pair once. Precision = detected seeded positives ÷ all alerts. Recall = detected seeded positives ÷ 12. These authored labels are evaluation fixtures, not real-world adjudications. A smaller queue alone is not a success.
02 / ALERT QUEUE
Follow the evidence
| Customer | Scenario | Signal | Review |
|---|
Which seeded cases did this configuration miss?
03 / SQL & DELIVERY
Inspect the work behind the screen
The dashboard executes real SQL in your browser. Data stays on this device. Saved assessments use this browser's local storage; QC is a role-play workflow, not authenticated separation of duties.
What this project demonstrates
Relational modeling, indexed joins, CTEs, aggregation, window functions, data-quality checks, parameterized scenario queries, threshold comparison and traceable reviewer decisions. Independent work sample built with AI-assisted development. This does not represent production Actimize configuration or validated fraud detection.