The math is the product. Here it is, in full.
A recommendation you can't audit is an ad. This page documents every step between your raw transactions and the number on your Leakage Report, including the assumptions, the conservative defaults, and what we deliberately don't do.
Categorization: a waterfall, not a guess
Every transaction is assigned a rewards-relevant category by the first layer that matches. Confidence matters more than coverage: only high-confidence spend enters the headline leakage number.
When your card feed carries a merchant category code, we map it directly.
The vendors that dominate SMB spend: Google Ads, Meta, AWS, FedEx, Delta, Uline. Every correction any customer makes improves the table for everyone.
Long-tail descriptors are classified by a language model. High-dollar unknowns route to a human review queue instead of silently entering the math.
Whatever we can't classify appears as an explicit bucket on your report and stays out of the headline number.
The optimizer: exhaustive, not clever
We take your three most recent months of spend, scale them to a 12-month pace, and score every wallet combination of up to 4 cards from the catalog. Brute force, no heuristics. For each candidate wallet:
Caps deplete realistically
Each category routes to the wallet card with the best effective rate; bonus caps are tracked and depleted, and spend past a cap earns the base rate.
Merchant-restricted bonuses
Amazon's 5% and Marriott's 6x count only against spend at matching merchants. A card never gets credit for a bonus your vendors can't trigger.
Fees are subtracted in full
A card must pay for itself in incremental rewards, or it doesn't make the wallet.
Meaningful thresholds only
Wallets cap at 4 cards (you can change this), and improvements under $250/year are suppressed. We don't recommend paperwork for pennies.
The report always leads with the zero-signup wallet: the best you can do by re-routing spend across cards you already hold. That number requires no application, no credit pull, and no trust in our incentives.
Point valuations: deliberately conservative
Points are only worth what you can actually redeem. Defaults are cash-equivalent floors, not blog-post fantasies, and co-branded currencies are discounted further, because points locked to one airline are worth their sticker rate only if your travel happens to match.
| Currency | Default ¢/pt |
|---|---|
| Cash back | 1.00¢ |
| Chase Ultimate Rewards | 1.25¢ |
| Amex Membership Rewards | 1.10¢ |
| Capital One miles | 1.00¢ |
| Brex points | 0.80¢ |
| Southwest Rapid Rewards | 1.00¢ |
| Delta SkyMiles | 0.90¢ |
| Marriott Bonvoy | 0.55¢ |
Every valuation is overridable in your report's assumptions panel, and the leak recomputes live. If you fly Southwest weekly, tell us, and the math will reflect it.
The card catalog: versioned, dated, reviewed
Earn rules change quietly and often. Every rule carries an effective date and a last-verified date, the catalog is reviewed monthly, and material changes trigger re-optimization and an alert for affected customers. A recommendation built on stale terms is worse than no recommendation; catalog freshness is the ongoing grind we signed up for.
Neutrality, mechanically enforced
The optimizer is bounty-blind
Affiliate bounty data lives outside the scoring engine. The engine's inputs are earn rates, caps, fees and your spend. Nothing else.
Disclosure on the recommendation
When a recommended card pays us a bounty, the badge sits next to the recommendation, not in a footer.
"You're already optimal" is a valid output
If re-routing your current cards captures the gap, that's the recommendation, and it earns us nothing.
We issue no card, move no money
Read-only access, credentials handled entirely by Plaid/OAuth. Basel initiates no payments, ever.
What the math is not
Rewards optimization is arithmetic on published card terms; it is not investment, legal, or tax advice. Issuer approval is never guaranteed, and issuers occasionally code merchants unpredictably. Where known coding quirks exist we encode them per-merchant, and where they don't, we default to the less favorable assumption.
Audit it against your own statements. That's the point.
Run the report