Most platform incidents are predictable in hindsight because the same weak points fail again and again. Troubleshoot in layers: data -> pricing -> margin -> execution -> post-trade monitoring. AI monitoring is useful when it remains auditable. Pair it with deterministic guardrails so a single model output cannot flip the market behavior. First confirm whether marks diverged from index. Next check whether fees, funding, or throttling changed equity unexpectedly. When risk limits are tiered, confirm how tiers are computed and updated. Silent tier changes can invalidate backtests. Compute liquidation price twice: once including fees and conservative slippage, and once with optimistic assumptions. The gap is your uncertainty budget. Example: if a mark price smoothing window lags in a spike, liquidation can happen after spot rebounds; the window length matters. Treat cross margin as a correlated portfolio, not a set of independent positions. Correlations tend to converge in selloffs. Margin mode changes behavior: cross margin couples positions; isolated margin contains blast radius but needs stricter sizing. Aivora discusses these topics as system behavior: define inputs, test edge cases, and keep controls auditable. Derivatives are risky; use independent judgment and test assumptions before scaling 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.