A clean backtest assumes that a trade can be placed, cancelled, hedged and measured exactly as
intended.
Live crypto trading does not behave like that. Exchanges differ in API behavior, order rules, margin logic,
websocket reliability and liquidation mechanics. Under stress, those differences stop being technical details.
They become PnL.
The problem is not only bad execution. The problem is not knowing that execution state has become unsafe.
"A strategy can be right and still lose if execution state is wrong."
The real problems
Seven execution failures that quietly destroy edge.
01
Cancel/fill race
A stale quote is cancelled, but the cancel confirmation arrives late. A trade hits the quote before the
order is actually removed. The strategy thinks risk was reduced; the account receives inventory at the
worst moment.
The real question: not “was cancel sent?” but “was the order removed
before it could fill?”
02
Websocket book drift
The local order book can become stale because of lag, missed updates or broken sequence IDs. The system
may continue quoting or hedging against a book that no longer represents the venue.
The moat: knowing when the book should no longer be trusted.
03
Partial hedge completion
One leg fills fully while the hedge leg fills only partially. The position is no longer neutral, but the
strategy may still treat it as an intermediate execution step rather than a separate risk state.
Where it hurts: funding trades, basis trades, market making with
inventory hedge and cross-venue spreads.
04
Rate-limit priority failure
During volatility, cancels, replacements, hedges and risk-reducing actions compete for the same API
budget. If every request is treated equally, new alpha-seeking orders may consume capacity while
exposure-reducing orders wait.
Correct priority: cancel unsafe quotes, reduce exposure, confirm
state, complete hedges, then seek new risk.
05
Silent venue rule changes
Tick size, step size, minimum notional, leverage ladders, post-only behavior and margin requirements can
change. The strategy may keep operating with yesterday’s assumptions until orders are rejected or risk is
mis-sized.
The risk: the system continues as if the old venue rules still
exist.
06
Mark price liquidation drift
A perp strategy may look safe using traded price or mid-price, while the venue’s mark price moves closer
to liquidation. Margin risk is often decided by mark price, not by the last traded price.
The better question: what price does the venue use to decide margin
stress?
07
Delisting and ADL unwind risk
Delistings can remove liquidity before positions are fully unwound. ADL, or auto-deleveraging, is when an
exchange forcibly reduces profitable positions during extreme stress because the insurance fund or
liquidation system cannot absorb losses.
The danger: a hedged position can suddenly become directional
because one leg is impaired or forcibly reduced.
The response
Execution needs state awareness, not just connectivity.
A basic connector can place and cancel orders. That is not enough.
A production execution layer must know when its view of the venue is incomplete, delayed or unsafe. In that
moment, the objective should shift from seeking alpha to protecting state accuracy and reducing exposure.
- Slow down quoting when order-book freshness deteriorates
- Prioritize exposure-reducing actions when API capacity is constrained
- Treat partially hedged exposure as an active risk state
- Block orders that violate updated venue rules
- Monitor adverse selection after fills, not just fill completion
- Pause new risk when execution certainty drops
Alphashots view
The trading posture should change before the loss becomes obvious.
At Alphashots.AI, execution risk is treated as a market-state problem. The system should not only watch
price, spread and signal strength. It should also watch whether execution conditions have degraded.
That means monitoring book freshness, cancel pressure, latency deterioration, fill asymmetry, hedge
completion risk, venue-specific stress and adverse-selection pressure after fills.
The output is not a price prediction. It is a posture decision: continue normally, reduce quote size, slow
replenishment, widen spreads, hedge first or pause new risk.
In crypto, the market does not only punish bad predictions. It punishes stale books, late cancels,
broken hedges, false position state and wrong assumptions about venue behavior. That is why
execution is not plumbing. Execution is the moat between a strategy that looks good and a strategy that
survives live trading.