Stress-test fraud rules before they cost good revenue.
RiskReplay is a visual fraud rule builder that replays nested rules against labeled synthetic transactions and shows capture rate, false positives, and approval impact before and after.
From raw heuristic to quantified trade-off in three steps
Select or generate labeled synthetic data
Start with 20,000 E-Commerce CNP authorizations, Account Takeover login sessions, or Refund Abuse rings — or dial your own seed, volume, and fraud rate.
Compose nested fraud rules visually
Combine velocity counters, geo distance, BIN risk, AVS/CVV codes, and account age inside nestable AND/OR groups with an instant plain-English preview.
Replay, sweep thresholds & share
Compare baseline vs candidate in milliseconds: inspect capture rate, false positive rate, $ good revenue declined, overlap, and export a read-only report.
Built for risk engineers who care about false declines
Visual Nested AST Rule Builder
Build multi-level AND/OR condition trees across 18 payment & identity signals with priority ordering, version tracking, and plain-English translations.
Sub-Second 20k-Txn Replay Engine
Compiled first-match evaluation replays 20,000 transactions across 10 rules in under a second directly in your browser or on the server.
Interactive Threshold Sweeps
Sweep any numeric cutoff (amount, IP distance, velocity, account age) to plot Capture Rate vs False Positive Rate and find the precision elbow.
Rule Contribution & Overlap
See exact first-match attribution, rule-level precision, and pairwise rule intersections so redundant rules don't hide behind priority #1.
Shareable Portfolio Reports
Freeze any baseline-vs-candidate run into a read-only shareable URL with one-click Print/PDF and crisp high-DPI PNG executive cards.
AI Rule Critique (Server-Side)
Flag overfitting and false-positive blind spots before shipping. Uses OpenAI or Gemini when configured, or deterministic Demo AI with zero keys.
What RiskReplay will NOT do
Transparency matters in risk engineering. RiskReplay is built for rule prototyping, education, and portfolio demonstration — never as a drop-in production processor.
- Simulation OnlyRuns exclusively on deterministic synthetic datasets. Not a live authorization or decisioning engine.
- No Real PII or PANsDemo mode uses fake names, example.com emails, 555 phones, and test BINs only. Never upload live cardholder data.
- Trade-offs, Not GuaranteesSynthetic recall, FPR, and approval rates illustrate rule mechanics — not guaranteed performance on any live portfolio.
- No Production ML ScoringFocuses on transparent, auditable deterministic rules and first-match priority semantics rather than black-box neural scoring.
Start in the free sandbox. Upgrade when your team is ready.
No live checkout in v1 — Individual and Team plans join the early-access waitlist unless payment links are enabled.
Interactive demo sandbox with bundled synthetic data (no cloud save).
- 20k E-Commerce CNP bundled dataset
- Visual nested AND/OR rule builder
- Sub-second before/after replay & sweep
- Zero login or API keys required
For solo fraud analysts, risk PMs, and portfolio builders.
- Unlimited versioned rules & rule sets
- All bundled datasets (CNP, ATO, Refund Abuse)
- Custom seeded synthetic dataset generator
- Persistent shareable reports & PNG/PDF export
- Server-side AI rule critiques
For payment risk teams calibrating rules collaboratively.
- 5 organization seats with Supabase RLS
- Shared versioned rule libraries
- Anonymized CSV dataset upload (planned)
- Side-by-side policy review reports
- Priority dataset scenario presets
Common questions
Can I connect RiskReplay to my live payment gateway or Stripe/Adyen stream?+
No. RiskReplay v1 is strictly an offline simulation and rule-design workbench on synthetic data. It is not a production authorization or decisioning engine.
Where does the transaction data come from?+
Every bundled dataset is generated deterministically at runtime from a seeded PRNG (mulberry32) with correlated fraud archetypes and intentional legitimate overlap. All emails use example.com, phones use 555 ranges, and BINs are test ranges only.
How are True Positives and False Positives defined when rules use Review vs Decline?+
Both Decline and Review count as interdicted (positive predictions) for Capture Rate (Recall) and False Positive Rate, while Decline vs Review counts and dollar impacts ($ Good Revenue Declined vs Reviewed) are broken out separately in every confusion matrix.
Does the demo require an account or API keys?+
No. Open /demo immediately with zero sign-up. You can edit starter rules, run live 20k-transaction replays, inspect threshold sweeps, and generate shareable simulation reports right away.
What will Team plans include when paid seats launch?+
Team ($149/mo) is planned for up to 5 seats, shared versioned rule libraries scoped by organization with Postgres Row Level Security, and custom anonymized CSV dataset uploads.
Join the RiskReplay waitlist
Get notified when hosted org workspaces, CSV uploads, and saved rule libraries roll out. Meanwhile, the interactive demo is open right now.