Counterparty — early warning for the sponsor-bank layer

The problem

Fintechs don't have bank charters. To offer a checking account, a debit card, or hold deposits, a fintech rents a real bank's charter — the “sponsor bank” sitting quietly behind the app. Most consumers have never heard of these banks, and that is exactly the problem: when a sponsor bank gets into trouble, the people who lose access to their money are the fintech's end-users, who never chose that bank and are almost always the last to find out.

The Synapse–Evolve collapse is the template. The warning signs — consent orders, leadership churn, compliance failures, litigation — are public, but they are scattered across regulators, courts, and filings that no one reads together. By the time the freeze hits and accounts go dark, the tells have been sitting in plain sight for a year. Counterparty was built on a simple premise: the layer with the most systemic risk in consumer fintech is the least-watched, and the leading signals are public — if someone bothers to fuse them.

The approach

Counterparty is two products in one codebase: a public weekly newsletter, and the intelligence engine that feeds it. The engine is the hard part.

Multi-source sweeps. Every cycle it pulls from eight public sources — FDIC call reports, SEC EDGAR filings, regulatory enforcement actions, CFPB consumer complaints, CourtListener litigation, state licensing, macro data, and press — each behind its own scheduled job, normalizing wildly different formats into one event stream.

A deliberately two-axis score. The risk model keeps DISTRESS (how likely a bank is unravelling) separate from EXPOSURE (how many fintech programs ride on it). A healthy bank carrying seven programs and a dying bank carrying one are different problems; collapsing them into a single number hides exactly the distinction a reader is paying for. Every rule that fires records its own contribution, so a score can always be decomposed and defended.

An entity graph. Fintechs, sponsor banks, and middleware are nodes; their relationships are timestamped, append-only edges. That makes contagion a query rather than a rebuild — “Bank X just got a consent order; which programs sit on it?” — and lets complaint volume roll up from individual fintech programs to the sponsor bank behind them.

A freeze detector, human-gated. A complaint-trajectory model flags anomalous funds-access spikes that look like an account freeze forming. Nothing it surfaces touches a published score automatically — every signal is a candidate a human reviewer confirms first.

From engine to newsletter

All of that machinery exists to produce one calm artifact: a Thursday briefing. The engine's signals become “What We Watch” and “How We Rate”; the analyst's read becomes the issue. The result is early warning for the people with the most to lose — fintech operators choosing a sponsor, the banks themselves, investors, and compliance teams — delivered before the breakup, not the week after.

What this means for your business

Counterparty is a working answer to a general question: how do you turn a firehose of messy, public, multi-source data into a defensible signal a person will actually trust? The same architecture — scheduled multi-source ingestion, an append-only entity graph, a transparent additive scoring model, and a human in the loop before anything ships — generalizes to any domain where the early warnings are public but no one is reading them together: vendor and counterparty risk, competitive intelligence, compliance monitoring, and beyond.