Synapse bankruptcy: the FBO ledger shortfall across four partner banks, timeline to September 2026

Synapse bankruptcy timeline from public records

The trigger for this post is the CFPB's second Civil Penalty Fund allocation for Synapse end users, dated 29 May 2026 on the Bureau's allocation schedule. Every row below comes from a court filing, a regulator page, or press coverage listed in the sources; we re-read all of them on 22 September 2026.

The trustee's third status report (filed 20 June 2024) names the four partner banks as American Bank, AMG National Trust, Evolve Bank and Trust, and Lineage Bank. The CFPB complaint (filed 21 August 2025) alleges that Synapse failed "to maintain adequate records of the location of consumers' funds" and to ensure "those records matched the records maintained by its partnering banks", and that the partnering banks found "a shortfall of between $60 and $90 million".

Failure class: FBO ledgers that reconcile in total and fail per bank

An FBO account, in the trustee's definition, is "a pooled account that allows a company to manage funds on behalf of, or 'for the benefit of,' one or more of their users without assuming legal ownership of that account." Three facts in the third status report (filed 20 June 2024) define the failure class:

• One ledger, several banks. End-user funds sat in omnibus FBO accounts and demand deposit accounts across the four banks. Each bank saw its own FBO balance; only the Synapse ledger said which end user owned which part.
• Money moved between banks. The partner banks discussed sharing historical interbank transfer records, and at least one bank reported transfers to American Deposit Management Company (ADMC), which ran a sweep network used for certain Synapse accounts from September to December 2023.
• Unallocated credits. About $31 million of Synapse-related interest and rebate payments at certain banks explained the width of the $65 million to $96 million range, pending reconciliation to decide how much belonged to end users.

The same report states that the ledger records sat in MongoDB and Amazon Web Services, and that the trustee was still arranging read-only access for the banks in June 2024. The cited sources do not identify the cause of the shortfall, and this post does not assign one.

The failure class is multi-bank FBO reconciliation: an owner-level ledger held by middleware, cash held at several banks, and transfers between those banks. A reconciliation that compares total ledger balances with total FBO cash can agree while the ledger attributes funds to the wrong bank, and it cannot say where an unallocated credit belongs.

Why it matters for testing in 2026

The FDIC's proposed 12 CFR 375.3 (89 FR 80135, 2 October 2024) targets this class. It would require beneficial ownership records in the format of its appendix A (375.3(a)), reconciliations "no less frequently than at the close of business daily" (375.3(b)(2)), and, where a third party keeps the records, "direct, continuous, and unrestricted access" for the bank, "including in the event of business interruption, insolvency, or bankruptcy of the third party" (375.3(c)(1)). Status verified on 2026-09-22: the Federal Register lists only the proposal and the comment extension under RIN 3064-AG07, with no final rule and no withdrawal, so the proposal is not in force.

The consequence without the rule is on the record. The CFPB enforcement page states that consumers could not reach their funds while the banks reconciled, and that many have not received their full balance. By 29 May 2026 the Civil Penalty Fund allocations for Synapse reached $55,214,058 in total, against the CFPB's $60 million to $90 million shortfall figure. The Banking-as-a-Service segment page lists what partner banks and supervisors ask a middleware platform to show.

The test that would have caught the gap

Our practice for a platform that spreads one ledger across several FBO accounts is a daily per-bank check before any total check. The data below is synthetic: five end users at three banks, with two injected breaks. A sweep of 40,000 minor units from bank-a to bank-b posted at both banks, and the ledger still places EU-0002 at bank-a. Bank-c credited 2,150 minor units of interest to its FBO account, and the ledger allocated it to no one.

as_of,end_user_id,bank,balance_minor
2026-09-17,EU-0001,bank-a,125000
2026-09-17,EU-0002,bank-a,40000
2026-09-17,EU-0003,bank-b,310000
2026-09-17,EU-0004,bank-b,5500
2026-09-17,EU-0005,bank-c,72000

as_of,bank,fbo_account,closing_minor
2026-09-17,bank-a,FBO-A-01,125000
2026-09-17,bank-b,FBO-B-01,355500
2026-09-17,bank-c,FBO-C-01,74150

Load the first file as ledger_positions and the second, built from each bank's closing statement balance, as bank_fbo_closing:

WITH ledger AS (
  SELECT bank, SUM(balance_minor) AS ledger_minor
  FROM ledger_positions
  WHERE as_of = '2026-09-17'
  GROUP BY bank
)
SELECT b.bank,
       COALESCE(l.ledger_minor, 0) AS ledger_minor,
       b.closing_minor,
       b.closing_minor - COALESCE(l.ledger_minor, 0) AS difference_minor
FROM bank_fbo_closing AS b
LEFT JOIN ledger AS l ON l.bank = b.bank
WHERE b.as_of = '2026-09-17'
  AND b.closing_minor <> COALESCE(l.ledger_minor, 0)
ORDER BY b.bank;

Output with SQLite 3.45.3:

bank    ledger_minor  closing_minor  difference_minor
bank-a  165000        125000         -40000
bank-b  315500        355500         40000
bank-c  72000         74150          2150

The totals are 552,500 in the ledger and 554,650 at the banks, so a total-only check reports a 2,150 surplus and misses the misplaced 40,000. The per-bank query reports both breaks. A test passes only when the query returns exactly the injected rows. Our suite adds two more cases: an owner-level export that the bank can read with the middleware switched off, and an interbank transfer that must post on both banks' ledger lines on the same business day.

The ledger reconciliation testing guide holds the full procedure with break injection and business-day windows. BaaS and partner bank integration testing is the service that runs these checks against a sponsor bank's files, and after a reconciliation incident sets out the sequence once a gap is already open. The test files hold end-user balances, which is account data, so we build them from synthetic records only.