TradeXYZ Incident

Incident Summary#

• On July 28, one SK hynix share traded at ₩1.272M ($868.17)—29.96% below the previous close—on a thin NXT pre-market order book.¹ The NXT cash price recovered to the ₩1.7M range within about two minutes,² but TradeXYZ's SKHX mark—a liquidation reference, not a closing price—temporarily fell 18.7%, from $1,127.90 to $917.25.³
• Up to 10x leverage, long concentration, and cross margin amplified the move. MarketsAlpha estimated that 960 accounts and $57.4M in long positions were liquidated, causing $17.3M in realized losses.⁴
• Multiple providers relayed the same NXT trade, so the price inputs were not truly independent.³ The trigger remains unknown; neither error nor manipulation has been confirmed. This was a pricing-design failure, not a code exploit.
• On August 6, an 11-share trade at ₩1.168M—29.98% below the previous close—repeated the NXT anomaly. Dynamic VI then triggered, and the cash-market decline narrowed to 3–4%.⁵

Why the NXT Pre-Market Matters#

• Nextrade (NXT) is South Korea's first alternative trading system and its second stock venue alongside the Korea Exchange (KRX). Its pre- and after-market sessions extend trading to 12 hours a day, improving access while adding competition on fees and execution.⁶
• NXT's pre-market runs from 8:00 to 8:50 a.m. KST. (TradeXYZ's current documentation lists external pricing from 8:10, although its post-mortem says the oracle was already in external-pricing mode at 8:01 during the incident.) ⁷⁸³
• TradeXYZ offers three Korean single-stock perpetuals: SKHX, SMSN, and HYUNDAI.⁸
• NXT uses continuous matching in the pre-market: orders can execute as soon as trading opens, rather than waiting for a single-price opening auction. A marketable sell order fills against the highest available bids first. On July 28, a thin order book allowed one share to execute at the low price; the exact order type and arrival sequence remain unconfirmed.⁹
• The mark price—not necessarily the latest execution price—is the reference used for P&L, margin, and liquidations; at 10x leverage, initial margin is roughly 10% of exposure.

Incident Flow#

The sequence below covers the July 28 incident. The August 6 event may be technically different. The specific Relayer values shown below are from 08:00:21.678 KST. TradeXYZ's official post-mortem reports the $917.25 Mark Price (−18.7%) at 08:01 KST.³

Unable to render diagram.

* Mark Price Calculation#

1. The TradeXYZ Relayer receives SK Hynix price data from the data providers—shown as externalPerpPxs in the Explorer—and converts it to USD.

2. On the first abnormal update, the Discovery Bound capped oraclePxs within ±10% of externalPerpPxs. Subsequent Oracle updates moved toward externalPerpPxs in observed steps of no more than 1%.

Discovery Bound: SKHX’s 10x maximum leverage set its Discovery Bound at ±10% of the external price (1 / maximum leverage). When the external price fell to $868.17, the permitted range became $781.35–$954.99. The Oracle first moved to the upper bound, then approached $868.17 in observed 1% steps about every three seconds. See the Oracle column in the table below.¹⁰¹¹

Current documentation may differ from the rules in effect during the incident.

3. Input A was an oracle-based Mark input. At 08:00:03, it was published at $954.99 and then declined through 3-second updates. Its exact transformation and rate limit are not publicly documented.

4. Input B was calculated from oraclePxs and an EMA (Exponential Moving Average) using a formula of the form B_raw = Oracle + EMA.¹² The EMA weights recent values more heavily; here, a 150-second time constant smooths the price gap between the order book and the Oracle. Based on the observed values, the selection algorithm appears to have been min(B_raw, Discovery Bound upper limit), but its mechanism is not publicly documented.

5. HyperCore calculated Input C onchain from three real-time values in the SKHX order book: bestBid, the highest buy order; bestAsk, the lowest sell order; and lastTrade, the most recent execution on the SKHX book. Input C was calculated as median(bestBid, bestAsk, lastTrade).¹³ At 08:00:21, the last observed trade was $1,020, suggesting C ≈ $1,020.¹⁴

6. HyperCore first calculated a raw target from median(A, B, C)—At 08:00:21, median($921.96, $954.98, ~$1,020) = $954.98—then limited the effective Mark to within 1% of the previous Mark per update (See the Mark Price (reconstructed) column in the table below).¹³

* Raw Target is reconstructed assuming C > B; because A ≤ B throughout the incident updates, median(A, B, C) = B.

Simplified Low-Price Execution Scenarios#

The actual order types and arrival sequence are unconfirmed. Both scenarios assume a marketable order requiring immediate execution.

Sell-side order error:

1. A buyer places a low limit bid at ₩1.272M ($868.17).
2. A seller mistakenly submits a marketable sell order.
3. With no higher bids on the book, the order executes at ₩1.272M ($868.17).
4. After the price recovers, the buyer can profit from the gap. The seller forgoes the difference.

Buy-side order error:

1. A low ask at ₩1.272M ($868.17) remains on the book.
2. A buyer mistakenly submits a marketable buy order.
3. With no lower asks available, the order executes at ₩1.272M ($868.17).
4. After the price recovers, the buyer can profit from the gap. The seller forgoes the difference.

In either low-price execution, the buyer benefits and the seller loses.

Root Cause and Defensive Opportunity#

• Undisclosed Trigger Event

  • The cause of the first abnormal trade has not been disclosed. Fat finger, manipulation, and market/system bugs remain possible root causes.

• Thin Liquidity at NXT Pre-market Open

  • A short pre-order window and continuous matching may have allowed one share to trade at an abnormal price before sufficient depth formed.
  • Defensive Opportunity: A longer pre-order window; a pre-open call auction, which collects orders and executes them at one price; an indicative price showing the estimated clearing price and imbalance; or Static VI. (NXT plans to introduce Static VI on September 14.¹⁵)

• Single Underlying Price Source

  • Multiple Data Providers relayed the same NXT trade. Their delivery paths were independent, but their price information was not. Using NXT as the primary source was reasonable if it was the only live venue (actually it was in Korea), but such a market was not a healthy sole basis for liquidations.
  • Defensive Opportunity: Add an independent price source. When none is available, lower confidence in NXT prices and sanity-check them against the previous KRX close and the TradeXYZ order book.

What does “independent” mean?

https://x.com/tradexyz/status/2082260930751082821

• Insufficient Price-quality Validation

  • A one-share trade affected the Mark Price without confirmation from trade size, market depth, or subsequent trades.
  • Defensive Opportunity: Validate or weight prices using trade size, depth, and persistence; for example, do not give a one-share trade the same weight as thousands of shares trading near the same price.

• Alert / Live Monitoring

  • A one-share reference trade, low trade count, thin depth, or a large deviation from the previous close were detectable. Timely intervention may have prevented the liquidations.
  • Defensive Opportunity: Alert on these conditions and pause or review Oracle updates and liquidations.

• Correlated Mark-price Inputs

  • A and B both depended on the Oracle. Although median(A, B, C) appeared to be 2-of-3, the Oracle effectively had two votes.
  • Defensive Opportunity: Do not treat A and B as independent price inputs.

• Price Safeguards and Liquidation Design

  • The Discovery Bound and 1% per-update Mark clamp were directionally useful but proved insufficient to prevent the liquidations.
  • Defensive Opportunity: Reassess how price safeguards interact with liquidation. This article does not propose a specific solution because the system's internal constraints are not fully known.

Potential Exploit to this System#

• An attacker could short SKHX, then execute a low-priced cash trade in the thin NXT market to push down the Mark Price and trigger Long liquidations.
• A Short trader reportedly realized about $2.2M through ADL. There is no evidence that this trader caused the NXT trade, and manipulation in this Incident remains unconfirmed.¹⁶

Incident Response and Post-Incident Measures#

• TradeXYZ said it would reimburse attributable liquidation losses and review its pricing system, but did not disclose the changes yet.³
• On August 6, 11 shares traded at ₩1.168M, 29.98% below the previous close. The 08:00:02.021 TradeXYZ Relayer update reported a normal price, and no liquidation cascade followed; any mitigation remains undisclosed.⁵¹⁷
• NXT plans to introduce Static VI on September 14. When Dynamic or Static VI is triggered, trading will switch to a two-minute single-price auction. (This change was planned before the Incident.)¹⁵

References#