BondBridge Intelligence
What it is, what it requires, and what is not built yet
BondBridge Intelligence is a proposed rule-based analysis layer that would surface patterns in a user's financial data — cash flow timing, charge anomalies, receivable gaps, debt-maturity collisions — and present them as structured evidence cards for the user to review.
No part of this system is live. No user data is collected, processed, or analyzed. This page documents the architecture, the 7 signal detectors under design, the 8 synthetic test cases written to validate them, and the backend, partner, and legal dependencies that must be resolved before any live data collection can begin.
Route and content audit
Signal detector catalogue — 7 hypotheses
These are design hypotheses only. No detector is implemented. No live data is processed. Each detector requires the backend and partner dependencies listed below before it can be built or tested against real data.
Consequence-before-amount
A user commits to a financial obligation (signs a lease, accepts a loan offer, books a service) before confirming the total cost including fees, penalties, and compounding. The detector would flag transactions or document patterns where a binding commitment precedes a complete cost disclosure.
Timing mismatch
Income arrives after recurring obligations are due, creating a predictable overdraft or late-payment window. The detector would identify the gap between payroll credit dates and fixed-debit dates across a rolling 90-day window.
Duplicate charge
The same merchant charges the same amount within a short window (3–7 days), suggesting an accidental double-billing rather than a recurring subscription. The detector would distinguish known subscription cadences from anomalous repeats.
Receivable vs. cash need
A user has outstanding receivables (invoices sent, payments owed) that, if collected, would cover an upcoming cash need — but the cash need is being met by credit instead. The detector would flag the gap between receivable age and credit utilization trend.
Hidden total cost
A product or service is presented at a headline price that excludes material costs (origination fees, mandatory add-ons, compounding interest on deferred payments). The detector would compare the headline figure in a document or transaction against the total obligation derived from contract terms.
Maturity collision
Two or more debt instruments mature or reset within the same 60-day window, creating a concentrated refinancing or repayment demand. The detector would map maturity dates from account data or uploaded documents and flag overlapping windows.
Cross-path milestone
A user's activity across two separate financial paths (e.g. credit-building and investing) reaches a point where a decision on one path materially affects the other — such as a credit utilization change that would affect a pending mortgage application. The detector would identify milestone intersections across paths the user has explicitly enrolled in.
Synthetic test cases — 8 validation scenarios
These are fictional scenarios written to validate detector logic before any live data is used. All names, amounts, and institutions are invented. They do not represent real users, real accounts, or real financial outcomes.
Jordan — timing mismatch, biweekly payroll
Target: Timing mismatch
Scenario
Payroll credits on the 1st and 15th. Rent debits on the 28th. Utilities debit on the 3rd. Three months of data show a consistent 2-day overdraft window between the 1st and 3rd.
Expected output
Detector flags a recurring 2-day gap. Evidence card shows the gap dates, the amounts involved, and a prompt to review whether a payroll advance or timing adjustment would close it. No recommendation to take any specific product.
Incorrect output guard: Detector must not recommend a specific financial product. It must not claim the user will overdraft — only that the pattern has appeared.
Alex — duplicate charge, gym membership
Target: Duplicate charge
Scenario
A gym charges $49.99 on the 1st (monthly subscription). The same merchant charges $49.99 again on the 5th of the same month.
Expected output
Detector flags the second charge as a potential duplicate. Evidence card shows both transaction dates and amounts. User is prompted to verify with the merchant before disputing.
Incorrect output guard: Detector must not initiate a dispute or contact the merchant. It must not classify the charge as fraudulent — only as a potential duplicate requiring user verification.
Morgan — consequence-before-amount, car finance
Target: Consequence-before-amount
Scenario
A financing agreement document is uploaded. The headline monthly payment is $389. The total obligation over 72 months including origination fee and GAP insurance add-on is $31,104 — 12% above the headline figure implied by the monthly payment alone.
Expected output
Detector flags the gap between the headline monthly figure and the total obligation. Evidence card shows both figures and the calculation. User is prompted to confirm they have reviewed the full contract before proceeding.
Incorrect output guard: Detector must not advise the user to reject the agreement. It must not claim the agreement is unfair or predatory.
Riley — receivable vs. cash need, freelance
Target: Receivable vs. cash need
Scenario
Three invoices totalling $4,200 are 45 days outstanding. In the same period, a credit card balance increased by $1,800 to cover operating costs.
Expected output
Detector flags the receivable age and the credit increase in the same window. Evidence card shows the receivable total, the credit increase, and the overlap period. User is prompted to review their collections process.
Incorrect output guard: Detector must not claim the invoices will be paid. It must not advise the user to take any specific collection action.
Sam — maturity collision, two loans
Target: Maturity collision
Scenario
A personal loan matures in 38 days. A business line of credit resets its rate in 44 days. Both require refinancing decisions within the same 60-day window.
Expected output
Detector flags the 6-day separation between the two events. Evidence card shows both instruments, their maturity/reset dates, and the overlap. User is prompted to consider whether they want to address both in a single planning session.
Incorrect output guard: Detector must not recommend a specific refinancing product or lender. It must not claim the user will face difficulty refinancing.
Casey — hidden total cost, BNPL
Target: Hidden total cost
Scenario
A buy-now-pay-later agreement shows 0% interest for 12 months. The fine print includes a $25 origination fee and a 29.99% deferred interest rate applied retroactively to the full original balance if not paid in full by month 12.
Expected output
Detector flags the deferred interest clause and the origination fee. Evidence card shows the headline cost, the origination fee, and the worst-case total if the balance is not cleared. User is prompted to confirm they have read the deferred interest terms.
Incorrect output guard: Detector must not advise the user to cancel the agreement. It must not characterize the product as predatory.
Drew — cross-path milestone, credit and mortgage
Target: Cross-path milestone
Scenario
User is enrolled in the credit-builder path and has a mortgage pre-approval in progress. A new credit card application (detected via a hard inquiry in credit data) would increase utilisation and potentially affect the pre-approval.
Expected output
Detector flags the intersection: a new credit application while a mortgage pre-approval is active. Evidence card shows the timing of both events and the potential utilization impact. User is prompted to consult their mortgage broker before proceeding.
Incorrect output guard: Detector must not advise the user to cancel the credit card application. It must not claim the mortgage will be affected — only that the intersection exists and warrants review. Credit data access requires FCRA permissible purpose — this test case cannot run without that clearance.
Null case — no pattern present
Target: All detectors
Scenario
90 days of clean transaction data: payroll arrives 3 days before all fixed debits, no duplicate charges, no outstanding receivables, no maturing debt, no cross-path intersections.
Expected output
No evidence cards surfaced. The system presents a neutral status: no patterns detected in the current data window.
Incorrect output guard: The system must not surface a card when no pattern is present. It must not suggest the user take any action when the null case is confirmed.
Dependencies that must be resolved before any live data collection
This list is not exhaustive. It represents the minimum set of dependencies identified at the time of writing. Additional requirements may emerge during legal review, partner negotiation, or technical implementation.
Legal and regulatory
- FCRA permissible-purpose determination by qualified counsel before any credit data is accessed or stored
- FTC Safeguards Rule compliance review and written information security program before any financial data is ingested
- State-by-state review of financial data aggregation licensing requirements (some states treat data aggregation as a regulated activity)
- Consent receipt architecture: a separate, documented consent per data class (bank feed, credit data, document upload, provider-authorized account data) with version control and revocation mechanism
- Data retention and deletion policy reviewed and approved before any data is stored
Partner and integration
- Plaid (or equivalent) data aggregation partner agreement — including liability terms, data use restrictions, and end-user license agreement requirements
- Credit bureau data agreement (Experian, Equifax, or TransUnion) — requires FCRA permissible purpose clearance first
- Tokenised credential store: no user credentials may be stored in plaintext or passed through BondBridge servers; partner tokens must be stored in an isolated, auditable vault
- Partner sandbox access for integration testing before any production data flows
Backend and infrastructure
- Secure data ingestion pipeline: isolated from the main application database, with field-level encryption at rest
- Rule engine with evidence-card schema, confidence threshold configuration, and human-review gate implementation
- Audit log: every data access, rule evaluation, and evidence card surfaced must be logged with timestamp, user ID, and data source
- Incident response plan for data breach or unauthorized access, reviewed before any live data is ingested
Product and design
- Evidence card UX: each card must show the source data, the rule that triggered it, the confidence level, and an 'incorrect / not relevant' control that suppresses the card and logs the dismissal
- No automated money movement: the system surfaces evidence cards only; it does not initiate transfers, payments, or applications on the user's behalf
- Professional referral gate: any card that touches a regulated decision (investment, credit, tax, legal) must include a prompt to consult a qualified professional before acting
This page is noindexed and intended for internal review only. It documents the design state of BondBridge Intelligence as of 2026-09-25. Nothing on this page constitutes a commitment to build, launch, or offer any of the described capabilities. No user data is collected, processed, or stored by any system described here.