Published 4 September 2026 · EarnPark
Status: this document reflects information available to EarnPark as of 4 September 2026. It will be updated as items in section 7 close and as further material information is received; updated versions will be published with prior versions preserved. Next scheduled update of this page: 02 October 2026.
This post-mortem is the technical analysis of the events of 23 August 2026: what happened on our Binance Portfolio Margin Pro account, why our systems did not prevent it, how we responded, and — the main part — what we have changed and are changing so that a gap of this kind is detected against the venue's own numbers and reduced before it becomes a strategy-level outcome. It is written for all EarnPark clients and for anyone evaluating the platform, not only for those affected.
Two documents sit alongside it. The EarnPark Incident Report — 23 August 2026 sets out the facts, figures and methodology for affected participants; the USDT/USDC DeFi Strategy Result Report — 29 August 2026 covers a separate result recognised in the DeFi strategies the same week. Both are available to clients on request. Individual account calculations and our correspondence with the exchange are outside the scope of this document.
1. In short
On 23 August 2026 at 21:11 UTC, Binance liquidated the positions on one of our Portfolio Margin Pro accounts after the account's unified maintenance margin ratio fell below the exchange's threshold. Four strategies operated through that account — Maker Core, BTC LP, ETH LP and ETH DeFi — and recorded a result of approximately −30%, applied to all participants under one methodology as of 23 August.
The cause was a gap in our own risk model: it valued collateral at full value, while the exchange applies collateral rates ("haircuts") that reduce the value it recognises for margin purposes. Our model therefore showed a wider distance to the liquidation threshold than actually existed. The positions themselves were hedged and intended to remain delta-neutral, and effective leverage was approximately 5.5x, within the internal limits then in force. The error was in the collateral valuation inside our risk model; the controls around the model, described in section 4.2, did not catch it. The error was ours, and we take responsibility for it. We are sorry this happened.
This document is about the part that follows from that: what was missing, and what now exists. Three changes are already in force: the exchange's collateral ratios are incorporated into the risk model (25 August); our margin assessment is reconciled in real time against the exchange's own calculation; and any breach of the uniMMR threshold is a standing trigger for reducing position size, with the threshold set individually for each strategy and asset. The full programme, with status and dates, is in section 7.
2. Background: how our exchange strategies are margined
Several EarnPark strategies execute through Binance Portfolio Margin Pro accounts. Under that programme, the exchange does not assess margin position by position; it calculates a single unified maintenance margin ratio (uniMMR) for the account as a whole. Every asset on the account contributes to a common collateral pool, and the exchange applies collateral rates ("haircuts") to certain assets — a discount that reduces the value recognised for margin purposes. When uniMMR falls below the exchange's threshold of 1.05, the exchange liquidates positions on the account under its published rules.
Two consequences of this design matter for what follows. First, margin is an account-level property: strategies that share an account share a margin outcome. Second, the distance to the threshold depends on collateral value after the exchange's rates are applied, not on gross market value. The rate applied to each asset is not uniform: it depends primarily on the type of the asset and on the amount of that asset held as collateral for the position. A model that ignores the rates will systematically overstate the buffer.
3. What happened — the short version
The full timeline is in the Incident Report. The points that matter for the analysis:
- 26 June — Binance updated its published Portfolio Margin Pro Maintenance Margin Ratio Calculation page, which sets out the use of collateral rates in the uniMMR formula.
- 18 August — we asked Binance to clarify how collateral rates apply under Portfolio Margin Pro. Binance explained that maintenance margin is calculated on a unified basis and referred us to its published uniMMR methodology, but did not expressly address the collateral treatment in the example presented.
- 23 August, 21:11 UTC — the account's uniMMR fell below 1.05, and the exchange liquidated its positions.
- 25 August — deposits into the affected strategies were suspended through the platform's Hard Cap function, fixing the calculation base.
- 28 August — a formal Information Request was submitted to Binance, including a request for the exchange-issued liquidation report.
- 29 August — results were reflected in all participants' accounts simultaneously, under one methodology, with individual notifications; deposits reopened within Hard Cap parameters.
- 30 August — accruals resumed from the updated base.
- 31 August — Binance confirmed that our questions are under review and indicated that, given the scope of the request, it needs additional time.
- 2 September — expanded disclosures and more detailed descriptions covering all strategies were published, together with the full text of the previous versions.
- 4 September — this post-mortem.
4. Root cause analysis
4.1 The model gap
Our risk model tracked the account's positions, hedges and leverage. It valued the collateral on the account at full market value. The exchange's margin engine values the same collateral after applying asset-specific collateral rates. The difference between the two valuations is the amount of margin buffer that existed in our model and did not exist on the exchange.
The positions were hedged and intended to remain delta-neutral; price direction was not the trigger. The account's effective leverage of approximately 5.5x was within the internal limits then in force, and the model's liquidation estimate was materially further from the threshold than the exchange's own calculation. In one line: gross collateral → collateral rates → recognised collateral → uniMMR; our model stopped at the first step. When the exchange's calculation crossed 1.05, the model still showed headroom. The trigger was the account's uniMMR falling below 1.05; the sequence of events inside the exchange's engine is part of the open information request described in section 5.
4.2 Why the gap was not caught
Five conditions were present at the time. Each is stated here together with the change in section 7 that addresses it.
- Validation against the model's own inputs. The model was validated against its own inputs, not against the uniMMR reported by the exchange. A daily comparison with the exchange's reported calculation is now in place (7.1); a continuous automated comparison is in 7.2.
- One account, four strategies. The four strategies operated through one exchange account. Under the exchange's account-level margin model, margin sufficiency is assessed for the account as a whole, so the result applies to all strategies on that account. Strategies are now being isolated from each other: each strategy moves to its own venue account, while the asset variants of a single strategy (for example, Maker Core BTC and Maker Core USDT) may continue to share that strategy's account (7.2).
- Alerting. Margin monitoring existed, but not as an around-the-clock system with multiple contacts and escalation at critical levels. It is being extended to that (7.2).
- An open margin question. On 18 August we requested written clarification from the exchange on this margin methodology; the response was general and did not directly confirm or refute the scenario we described. Our standard is now different. We do not open positions until we have studied the venue's documentation in detail and obtained confirmation of the margin mechanics. A decline in uniMMR below the threshold set for a strategy and asset triggers a reduction in position size, regardless of the buffer our own model shows. We are also adding AI-assisted review of venue documentation and risk parameters as a further check on top of human sign-off.
- Exchange rule changes. The exchange's published margin methodology page was updated on 26 June. Changes to published margin rules were not tracked by a dedicated automated process with a mandatory model recalculation; that process is in 7.3.
4.3 What was not a factor
This was not a hack, not a theft, not a compromise of infrastructure or keys, and not a withdrawal of funds by any third party. No strategy outside the four named above was on the affected account. Withdrawals remained available throughout, including the day of the event. Based on the information currently available to us, the liquidation was executed under the exchange's published margin framework. Before the Incident Report was issued, we received an exchange-issued Liquidation Snapshot. It is being reconciled against our own records. The formal exchange-issued liquidation report and Binance’s full response remain pending, and we have not reached final conclusions on the exchange's calculation or execution of the liquidation.
5. How we responded
The response ran on three tracks in parallel.
Containment and calculation. Deposits into the affected strategies were suspended on 25 August so that no new funds would enter while the result was calculated. The result was computed at strategy level on balances as of 23 August and applied to all participants simultaneously on 29 August through the platform's standard result-transaction mechanism — the same mechanism that credits returns. Deposits made on or after 20 August and before the liquidation were excluded from the calculation and remain on balances in full. Accruals for the period after the event were reversed where strategy terms provide for them; the accrual for 23 August itself was preserved. No special adjustment types were created. The founders' and the team's own funds in the affected strategies were treated identically.
Communication. Affected participants were notified individually, an official incident report was prepared and issued, expanded disclosures were published for all strategies together, and we have held direct calls with the most affected clients, to whom we reached out personally. This post-mortem is the next step in that sequence, not the last: this page will be updated against the dates in section 7 and as the exchange track develops.
The exchange. On 28 August we filed a formal Information Request with Binance, including a request for the exchange-issued liquidation report. On 31 August the exchange confirmed that our questions are under review and indicated that, given the scope of the request, it needs additional time. The liquidation snapshot has since been received and is available to affected participants on request through their personal manager; the reconciliation of its contents against our records will be reflected in an updated Incident Report, with prior versions preserved. The full, formal exchange-issued liquidation report has not yet been received. The case is open. We do not promise any outcome, and the exchange has not committed to timelines. We received the exchange-issued Liquidation Snapshot before the Incident Report was published; it is being reconciled against our records.
Any funds received from Binance as a result of this process will be distributed among participants in Maker Core, BTC LP, ETH LP and ETH DeFi in proportion to their positions on 23 August, whether or not they remain on the platform.
6. A separate event the same week: USDT/USDC DeFi
On 29 August an accumulated negative result of approximately −14% was recognised in the USDT DeFi and USDC DeFi strategies. It did not arise from the liquidation. These strategies operate through DeFi protocols, and their result came from the strategies' own positions: adverse movement in the Hydration reference value relative to the liquidity positions, funding costs, and price divergence for the same asset across venues.
Before the result was reflected in participants' balances, it had been offset against EarnPark's performance-fee margin. Continuing to defer recognition could have shifted a disproportionate share of the remaining negative result to participants who stayed in the strategies while others withdrew; the outstanding amount was therefore recognised in full and simultaneously on 29 August. The disclosures published on 2 September now describe the factors that may affect these strategies and the timing with which accumulated strategy results may be reflected in participants' balances. The details are in the USDT/USDC DeFi Strategy Result Report.
7. What we changed — and what is still in progress
The programme below is the substance of this document. Each item is listed with its status as of 4 September 2026, its owner by role, and, where not yet complete, the target date. We will update this page as items close. Until per-strategy venue accounts are in place (7.2), the affected strategies operate under the interim controls already in force: the daily manual check against the exchange's calculation, the standing de-risking rule, and the internal leverage limits then in force.
7.1 In force
7.2 In progress
7.3 Planned
The items below go beyond the direct chain in 4.2 and are structural.
8. What did not change
Some things did not change because they worked, and we are naming them so that the changes above are read in proportion.
- The result of a strategy — positive or negative — is attributed to its participants in proportion to their share, through one mechanism, on one date, with no exceptions and no special treatment. That applied here, including to the founders' and the team's own funds.
- Withdrawals were not suspended on 23 August and continued under each strategy's standard terms. Following the recalculation, requests submitted on the basis of previous balances had to be resubmitted using the updated balances and confirmed through 2FA.
- We publish what happened, in writing, with prior versions preserved, and we speak to clients directly. The disclosures released on 2 September and this document reflect the approach we intend to apply to every strategy going forward.
9. Closing
A modelling gap of this kind is the responsibility of the people who built the model, and that is us. What we owe our clients after the fact is not a promise that nothing will ever go wrong, but a system designed to catch a single modelling gap before it becomes a multi-strategy outcome: independent verification against the venue's own numbers, alerting that reaches a person, each strategy in its own account contour, and a rule that treats uncertainty as a reason to reduce exposure rather than to wait. The model fix, the daily check against the exchange's calculation, and the de-risking rule are in force; the remaining items have dates in section 7, and we will report against them on this page.
Risk limits and the remediation programme are approved by the founders as the company's management; a risk function independent of execution is being established (7.3).
Questions about this document can be sent to your personal manager or to support. Formal claims may be submitted to legal@earnpark.com. Nothing in this document is intended to characterise the conduct of any third party or to create obligations beyond the terms of each strategy.
Eugene Netso, Nick Nazmov
Co-founders & Managing Partners, EarnPark