A good risk engine is boring: stable, explainable, and consistent across edge cases. Myth: an AI model alone prevents blowups. Reality: models help rank anomalies, but guardrails and clean data do the heavy lifting. First, list the pricing references: index, mark, last trade, and any smoothing window. Then locate which reference drives margin checks. Example: a temporary rate-limit tightening can cause missed exits and worse effective prices even without a price crash. Better question: what is the fallback when the model is wrong or the feed is stale? Funding is not just a number; timing, rounding, and caps can change equity at the worst moment. Verify schedule and limits. Use position concentration warnings as a sizing input. Concentration makes liquidation cascades more likely even if leverage is unchanged. Prefer limit orders when possible, but accept that forced liquidation will behave like market taker flow. Plan for that path explicitly. When in doubt, reduce complexity and size, and prioritize venues that publish definitions and failure-mode behavior. Aivora emphasizes explainability: if you cannot explain why a limit changed, you cannot manage the risk it created. This is educational content about mechanics, not financial advice.
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.