The BULK Exchange BTC Liquidation Cascade, Explained: How a Thin Order Book Printed Bitcoin at $123K

Disclosure: BuiltOnBulk participates in BULK Exchange’s referral program and links to app.bulk.trade elsewhere on this site. We’re publishing this piece anyway because a fair accounting of what happened matters more to our credibility than looking away from it — but you should know the relationship exists as you read.

A note on sourcing: this draws on an independent trade-level reconstruction by trader @Rutto02 on X, an official statement from BULK co-founder @junbug_sol (also posted to BULK’s own Discord with the same figures), BULK’s Discord announcement channel more broadly, and BULK’s public project documentation. We have a working permalink for Jun’s X post but couldn’t fetch/render its content directly (X blocks automated retrieval for us), so we’re relying on the text as relayed to us rather than an independent re-read of the tweet itself; we still don’t have a permalink for @Rutto02’s thread. BULK also referenced a fuller internal incident report on Notion, which is access-restricted and we could not retrieve. We have not independently verified the underlying on-chain data ourselves. Where accounts disagree, we’ve said so rather than picking one as authoritative. Treat the specific dollar figures below as reported, not as independently confirmed.

On September 6, 2026, an account executing roughly $2 million in sell orders pushed BULK Exchange’s BTC perpetual — a Solana-based perps market — to print as high as $123,221, against a Bitcoin price of roughly $79,600 on major centralized exchanges at the same moment. That’s according to the independent reconstruction; BULK’s own statement confirms a related but not identical sequence of events. Neither account describes a hack, an exploit, or a smart-contract breach. Both describe something arguably more uncomfortable for a young exchange: every liquidation and auto-deleverage fired exactly as the code was written to do, and the code still produced a 55%+ price dislocation.

This piece reconstructs both accounts of what happened, flags where they don’t agree, and covers the fixes BULK has said it’s shipping.

The setup: a thin book on a new exchange

BULK is a Solana-native perpetuals exchange using an on-chain central limit order book and its own consensus mechanism (BulkBFT) across validators, targeting sub-400ms settlement. It raised an $8 million seed round in 2025, ran a public testnet from November 2025, and launched mainnet trading on September 5, 2026 — one day before this incident.

BULK’s own statement describes the period as a “liquidity-bootstrap phase,” which is a fair characterization: a market that traded only around 7 BTC in the prior hour and 43 minutes, per the independent reconstruction, does not have the depth to absorb a $2 million market sell without a severe price impact. A thin book on a brand-new exchange isn’t unusual by itself. What turned it into a five-figure-per-BTC dislocation is what happened next.

Timeline, as reconstructed by @Rutto02

One important caveat up front: the independent reconstruction times the event starting around 21:32 UTC, while BULK’s own statement frames its version of events in a 14:15–15:15 UTC window. That 14:xx window isn’t a one-off — BULK’s team gave the same framing twice, independently: first in an initial Discord announcement (“At 14:52 UTC the BTC market on BULK experienced unusual price action…”), and again in Jun’s follow-up post on X restated in BULK’s own Discord (“Between 14:15–15:15 UTC…”). Two of BULK’s own channels agreeing with each other doesn’t resolve the discrepancy with the independent reconstruction’s 21:32 UTC framing — it’s still possible that’s a timezone artifact, a different reference point, or a genuine error on one side — but it does shift our confidence somewhat toward BULK’s stated window being the internally consistent one. We’re presenting both without inventing a bridge between them.

BULK’s Discord also adds one detail not in the original X statement: the incident happened because “the book [was] still being seeded in its first day” — mainnet had launched barely 24 hours earlier, on September 5.

Per the independent reconstruction (times as given in that thread):

If accurate, the full dislocation ran roughly five and a half minutes from first liquidation to mean-reversion.

What the mechanism reportedly was

Neither source claims a hack. The independent reconstruction’s read on why the mechanism produced this outcome points to three compounding design factors:

1. No price band on market orders. A market order is described as having had no cap on how far it could walk the book, which is how a stale resting ask at $115,115 got filled at face value once real depth ran out.
2. A mark price that reportedly leaned on the local book under volatility. BULK’s mark price is described as a median of an external-oracle-adjusted price, the live book price, and a smoothed book price — but under high volatility, the local book component is reported to have dominated. On a book only a few BTC deep, that means the same forced sell that moves the tradable price can also move the reference price used to calculate other accounts’ liquidation thresholds.
3. Closeout logic reportedly tied to instantaneous local liquidity, rather than to a robust external reference — meaning a temporary, self-inflicted liquidity gap could directly worsen the numbers for accounts already being forced to close.

Combined, if this reconstruction is accurate, the sequence is: a large forced seller drains the book → the drained book distorts the mark price → the distorted mark price triggers further liquidations at worse levels → those liquidations find no real liquidity and print against a stale order → the resulting print exceeds what margin can cover → auto-deleveraging pulls the shortfall from unrelated, profitable accounts.

What BULK’s own statement says

BULK co-founder @junbug_sol’s statement frames the same underlying event with its own account-level detail: the trigger account is described as generating roughly $1.5 million in sell volume across 400 fills, partially liquidated eight times before the main event. A protective short-close order is described as activating near a mark price of $81,450 at 14:36:45 UTC (note: this timestamp is from BULK’s stated 14:xx window, not the 21:xx window in the independent reconstruction), converting into a reduce-only market buy that consumed available ask liquidity and landed on the resting ask at $115,115. Three further protective closes are described firing near $82,000–$82,500.

Per BULK’s figures: BTC open interest fell from roughly 26.21 BTC to 6.82 BTC over the event. ADL is stated to have covered about $48,480 of the resulting shortfall, with BULK’s risk vault absorbing an additional ~$66,149 that ADL didn’t recover — roughly $114,600 combined. Across the full incident, BULK’s statement cites approximately 26.55 BTC in forced liquidations and 17.69 BTC in matched ADL reductions.

Where the two accounts diverge — an open question, not a verdict

Here’s the discrepancy worth flagging plainly rather than glossing over: the independent reconstruction puts total liquidation damage at roughly $3.01M across 33.67 BTC and 28 accounts, plus a separate ~$2.82M ADL event across 34.75 BTC and 248 accounts. BULK’s own statement cites combined ADL-plus-risk-vault losses of roughly $114,600 — smaller by close to an order of magnitude — even though the BTC volumes each account describes (33.67–34.75 BTC on one side, 26.55–17.69 BTC on the other) are in a broadly similar range.

We want to be careful about what this gap does and doesn’t tell us. It could reflect a genuine definitional difference — “total value liquidated across the whole cascade” is a much broader number than “shortfall not covered by ADL and absorbed by the risk vault,” and figures denominated in BTC-at-time-of-trade versus USD-at-time-of-report will diverge sharply across a range that spanned $78,000 to $123,221 within the same event. It’s also possible one or both reconstructions contain errors — trade-level reconstructions from public data can miscount, and official statements can understate scope, intentionally or not. We don’t have enough independently verified information to say which explanation is correct, and we’re not going to guess. What we can say is that the two public accounts of the same incident don’t reconcile cleanly, and if you’re trying to size the real impact of this event, that’s a gap worth watching for further clarification from BULK rather than treating either number as settled.

The fix BULK has announced

Per BULK’s statement, the platform’s next network upgrade targets the mechanical root causes directly:

• Mark price anchored more heavily to external data. During the liquidity-bootstrap phase, mark price is said to lean more on external market data, require a minimum depth threshold before local book inputs can move it, and cap how far mark can diverge from the oracle.
• Equity and liquidation eligibility decoupled from instantaneous local liquidity. Rather than valuing positions against an estimated cost-to-close on the live book, the stated plan is to evaluate equity and liquidation eligibility against the external market price, so a temporary depth gap shouldn’t by itself trigger a liquidation.

BULK’s statement also says it is reviewing every affected liquidation and ADL account for reimbursement, and states zero tolerance for deliberate manipulation of thin-liquidity conditions, without naming the trigger account as a bad actor at time of writing.

Update (September 7, 2026): reimbursements and the upgrade are underway

A pinned message from BULK’s team account on Telegram, posted the same day, gives a concrete timeline for both open items above:

• Reimbursements: BULK says it will begin distributing reimbursements within 1–2 hours of that post, credited directly to affected accounts and accessible instantly — plus an extra 10% on top of the expected reimbursement amount, described as a gesture of appreciation for early-user support during the disruption.
• The network upgrade, targeted for completion 2–3 hours after that post, is described with more specificity than the original statement: (1) mark price is now more robust in how it follows the local best bid/offer, with safety bounds that prevent extreme price dislocation relative to the oracle; (2) liquidations now follow the fair price of the book and discount for available liquidity, which BULK says should prevent low-liquidity-driven liquidations and ADL, and “hardens” the exchange against the kind of low-liquidity manipulation that produced this cascade; (3) additional robustness improvements to isolated margin, withdrawals, and other areas. BULK says the upgrade requires about one hour of downtime, and that the exchange will stay in post-only mode afterward until liquidity partners are properly back online.

We’re citing this as a screenshot of a pinned Telegram post, not a linkable permalink, so treat the exact wording as reported rather than independently verified against a canonical source — but directionally, it lines up with and sharpens the fix BULK had already announced (mark price anchored to external data, liquidation eligibility decoupled from instantaneous local liquidity), and it resolves the open reimbursement question with an actual timeline rather than an open-ended “we’re reviewing accounts.” We have not independently confirmed reimbursements have landed or that the upgrade completed on schedule.

Update: the upgrade now has a confirmed date and time

BULK has since published a formal “Mainnet Upgrade Notice” with a fixed schedule, superseding the “2–3 hours after that post” estimate above:

“At 1PM UTC, Sept 8 the BULK Network will undergo an upgrade to introduce following features: (1) Mark price is now more robust and follows the local BBO and has new safety bounds that prevent any extreme price dislocation to oracle. (2) Liquidations now follow fair price of the book and discount the impact based on liquidity — this should prevent low liquidity liquidations and ADL on the network. This also hardens the low liquidity manipulation that can cause liquidations. (3) Robustness improvements on isolated margin, withdrawals, and other areas of the network. Expect 1hr of downtime. After the upgrade, markets will stay on reduce-only + post-only mode to allow market makers to replace the liquidity. Public trading will begin soon after.”

In plain terms, for anyone who traded through the September 6 cascade: mark price will no longer lean primarily on the local order book it’s supposed to protect, with hard bounds against dislocating from the oracle; and liquidation eligibility will price in how much available liquidity actually exists before forcing a close, rather than treating a temporarily thin book as if it were a fair market — the exact mechanical gap this cascade exposed. Expect roughly an hour of downtime starting 1PM UTC on September 8, 2026, then a reduce-only/post-only window while market makers rebuild depth before normal trading resumes. See How BULK Exchange Liquidations Work for the updated mechanics in full once live.

What this means for traders, regardless of which numbers are right

Thin liquidity feeding into a mark-price formula that references the very book it’s meant to protect is a known failure pattern for on-chain perpetuals broadly — it isn’t unique to BULK. Other large perp DEXs, including Hyperliquid, have publicly dealt with related incidents (a March 2025 market-manipulation event and later oracle-driven cascades on thinly traded markets are documented elsewhere). We’re not attempting a detailed one-to-one comparison here, since we haven’t independently re-verified those other incidents’ specifics for this piece — the point is only that the category-level risk (thin book + leverage + a pricing mechanism that can be distorted by the conditions it’s supposed to price around) is a recurring one across this class of exchange, not evidence of a BULK-specific defect.

The practical takeaway for traders on any young, on-chain perp exchange: size leverage to actual book depth, not just to account equity. A thin, self-referencing order book can turn a normal-looking short into a forced closeout at a price the rest of the market never saw — no exploit required.

Next steps

• BULK’s mainnet is invite-only — you’ll need a free access code to create a new account, in addition to a referral link.
• Want the full list of what’s tradable and BULK’s current market lineup? See every live market on BULK.
• Ready to trade once you’re in? app.bulk.trade.

This piece will be updated if BULK publishes further clarification on the loss figures discussed above, or if we’re able to independently verify the source threads. Not financial advice. BuiltOnBulk is an independent site and is not BULK Exchange.