Hedging a position it never held

On 24 June 2026, a panel of the Chicago Board of Trade's Business Conduct Committee accepted a settlement with Spartak Trading LLC, a US proprietary trading firm, and fined it $80,000. The headline number is small, but the mechanics behind it are the stuff of every risk manager's nightmares. During the Treasury futures roll in late August 2024, an automated trading system Spartak operated used an auto-hedger parameter set to an inappropriate value. When a large fill event hit the market, the hedging system misread it, concluding that the entire first level of the order book had traded in full. On that false belief it entered large aggressor orders to hedge a position it did not actually hold.

What happened next took seconds. A Spartak trader aggressively sold contracts at the same price to offset the phantom position the machine had created, and the combined activity produced a significant decline in liquidity in both the 10-Year and 5-Year futures spreads, volume aberrations at the close of a roll-period trading day, and significant latencies in the exchange's match engine and market data. A single misconfigured setting, in other words, did not just cost the firm money; it briefly degraded the market itself. CBOT found that Spartak had failed to diligently supervise its automated trading system, in breach of its rules and the exchange's advisory notice on automated trading.

The case is worth far more attention than its fine, because it is a clean, public illustration of how trading algorithms actually break. Real-world algo failures are rarely dramatic acts of manipulation; far more often they are a wrong number, an unhandled edge case, an assumption that held until the one morning it did not. This piece uses Spartak to lay out the common ways trading algorithms fail, what the enforcement record shows about where these cases land, and how firms can test their own risk frameworks against these failure modes before an exchange does it for them.

The panel found two distinct violations. The first concerns the auto-hedger episode: Spartak violated CBOT Rule 432, which requires diligent supervision, and the exchange's automated-trading advisory (RA2403-5), which requires any party operating an automated trading system to monitor and supervise it while running and to take immediate, reasonable corrective action if it disrupts the market or the exchange's systems. The inappropriate parameter, the resulting phantom hedge, the aggressor orders and the market disruption together amounted to a failure to diligently supervise the automated system.

The second violation is separate and behavioural rather than accidental. Between August 2024 and May 2025, in markets that use a pro-rata matching algorithm, Spartak entered multiple maximum-quantity orders in an effort to capture a greater share of passive fills. Pro-rata matching allocates fills in proportion to displayed size, so posting maximum-size orders games the allocation in the firm's favour, at the expense of other passive participants. The panel treated this as inconsistent with just and equitable principles of trade. It also flagged a latent capital risk: had all of those maximum-quantity orders on one side executed at once, the resulting margin call, in that instant, could have exceeded Spartak's immediately available, liquid, unencumbered assets.

Neither finding required proof that Spartak set out to manipulate anything, and the firm settled without admitting or denying the findings. That is the important point. The exchange sanctioned the firm for the state of its controls and the design of its strategy, not for intent. In automated markets, the rulebook increasingly treats a poorly governed algorithm as a violation in its own right, because at machine speed the gap between a configuration error and a market event is measured in seconds.

Spartak is a useful anchor because its failure was not one failure but a chain, and each link is a recognisable, recurring failure mode. Start with the trigger: a single parameter set to the wrong value. Auto-hedgers, throttles, size caps and price collars are all governed by parameters, and a parameter that is reasonable in normal conditions can be catastrophic in an edge case such as a large fill during a thin roll period. Most algo incidents begin exactly here, with a value that was never stress-tested against the conditions that eventually arrived.

The second link is a false view of the world. The hedger concluded that the entire first book level had traded when it had not, and acted on that phantom state. Algorithms act on their model of the market, not the market itself, and when the input is misread, stale or hallucinated, the machine will confidently do the wrong thing. The third link is speed and scale: within seconds the system had entered large aggressor orders and a human trader was scrambling to offset them. Automation removes the natural friction that lets a person catch a mistake, so a small error propagates into a large position before anyone can intervene. The fourth link is the market impact: liquidity fell, volume distorted and the match engine lagged. An algorithm does not fail in isolation; it fails into a shared venue, and one participant's runaway logic becomes everyone's latency.

Beyond this specific chain, the enforcement record shows a consistent catalogue of ways algorithms go wrong, and it is worth naming them plainly.

The connective tissue across all of these is governance, not code. Spartak's fine was not for having a bug; every system has bugs. It was for failing to supervise the system diligently: no parameter validation caught the bad value, no real-time monitoring flagged the runaway hedge quickly enough, and no control stopped a strategy that could, in the worst case, have outrun the firm's own capital. The algorithm failed the way algorithms usually fail, and the controls that should have caught it were not there.

The controls that would have caught Spartak are well established, and the exchange rulebooks now effectively require them. Drawing on the control design RegLabs derives from this case, four are non-negotiable for any firm running automated strategies.

The second is pre-trade risk controls with a working kill switch. Every order should pass a pre-trade engine that enforces maximum order size, order-rate limits, position limits and loss limits before it reaches the exchange, and an authorised person or automated trigger must be able to halt a strategy or the whole firm instantly. Spartak's runaway hedge and its maximum-quantity strategy are both exactly what pre-trade limits and a kill switch exist to contain.

The third is real-time monitoring of the algorithm's behaviour, not just its P&L. The exchange advisory requires firms to supervise an automated system while it operates and to take immediate corrective action. In practice that means live dashboards and alerts on order volume, aggressor-to-passive ratios, fill rates, latency and market-impact indicators, so that a system hedging a position it never held is flagged in seconds, not reconstructed after the fact. Detection has to run at the speed the algorithm does.

The fourth is strategy design that respects the market and your own balance sheet. The maximum-quantity finding is a reminder that a strategy can be a violation by design, not by malfunction: gaming a pro-rata matching engine harms other participants, and stacking maximum-size orders created a tail risk that could have exceeded the firm's liquid capital. Strategies should be reviewed for their market impact and their worst-case capital draw before deployment, with a net-capital buffer sized for the position the algorithm could actually build, not the one it usually holds.

Spartak is one small entry in a large and active enforcement theme. Across the RegLabs record, algorithmic-trading governance failures account for more than 580 catalogued actions and over $6 billion in penalties, spread across the SEC, the CFTC, FINRA and, above all, the exchanges themselves, which police automated trading on their own venues. The Spartak fine sits at the small end by dollars, but the failure mode it represents sits right in the middle of the distribution: a control breakdown around an automated system, handled by the venue where it happened.

Zoom into the venue where Spartak landed, and the pattern is precise. Across CME Group's catalogued algorithmic-trading cases, the single most common finding is failure to supervise, followed by disruptive trading, wash sales and quote-stuffing or quote-flickering, and the firms on the receiving end are overwhelmingly proprietary trading houses and market makers operating in futures. In other words, the Spartak profile, a prop firm, a supervision failure, a futures product, a disruptive market impact, is not unusual. It is the modal case.

The strategic reading is that the fine is the least of it. A parameter error that jams a match engine is one bad configuration away from a genuine loss event, a client harm, or a market-wide dislocation, and the same governance gaps that produce an $80,000 exchange fine are the ones that produce a nine-figure trading loss when the numbers are larger. The firms that stay clear of both are the ones that treat their algorithms as supervised systems with tested parameters, hard pre-trade limits, real-time behavioural monitoring and a kill switch, and that prove those controls work against the specific ways algorithms fail rather than assuming they hold.