AI can help rank anomalies, but it cannot replace clear rules you can audit.
The mechanism: Look for the platform's fallback rules: what happens if a feed is stale, if the book is thin, or if volatility spikes faster than normal sampling windows. If you see unexplained liquidations, compare index updates to mark sampling and check whether outlier filters are documented.
Where it breaks: Liquidation is a path, not a single event. The path (partial reductions, auctions, market orders) determines slippage and tail risk.
A simple test: Treat cross margin as a correlated portfolio. Correlations converge during stress, so diversification can vanish when you need it most. Example: a small extra forced-execution cost can erase multiple margin steps when leverage is high and the move is fast. Compute liquidation price twice: once with optimistic assumptions, and once with conservative slippage and fees. The gap is your uncertainty budget.
What to do next: Pitfall: trusting a single data source. One stale oracle feed can distort index and mark calculations if fallbacks are weak.
Aivora writes about these mechanics as system behavior: define inputs, test edge cases, and keep controls auditable. Derivatives are risky; test assumptions before you scale size.
The mechanism: Look for the platform's fallback rules: what happens if a feed is stale, if the book is thin, or if volatility spikes faster than normal sampling windows. If you see unexplained liquidations, compare index updates to mark sampling and check whether outlier filters are documented.
Where it breaks: Liquidation is a path, not a single event. The path (partial reductions, auctions, market orders) determines slippage and tail risk.
A simple test: Treat cross margin as a correlated portfolio. Correlations converge during stress, so diversification can vanish when you need it most. Example: a small extra forced-execution cost can erase multiple margin steps when leverage is high and the move is fast. Compute liquidation price twice: once with optimistic assumptions, and once with conservative slippage and fees. The gap is your uncertainty budget.
What to do next: Pitfall: trusting a single data source. One stale oracle feed can distort index and mark calculations if fallbacks are weak.
Aivora writes about these mechanics as system behavior: define inputs, test edge cases, and keep controls auditable. Derivatives are risky; test assumptions before you scale size.
Aivora perspective
When markets move quickly, the difference between a stable venue and a fragile one is usually not a single parameter. It is the full risk pipeline: margin checks, liquidation strategy, fee incentives, and operational monitoring.
If you trade perps
Track funding and realized volatility together. Funding tends to amplify crowded positioning.
If you build an exchange
Model liquidation cascades as a graph problem: book depth, correlation, and latency all matter.
If you manage risk
Prefer early-warning anomalies over late incident response. Drift is a signal, not noise.
Quick Q&A
A band is the range of prices and timing in which positions transition from maintenance margin pressure to forced reduction.
Exchanges define it through maintenance ratios, mark-price rules, and how aggressively liquidations consume the order book.
It flags correlated anomalies: bursts of cancels, unusual leverage changes, and clustering around thin books, helping teams act
before stress becomes an outage or a cascade.
No. This site is educational and system-focused. You are responsible for decisions and risk management.