9. Market and Order Lifecycle

Interview outcome

Speak the domain language needed to build a trading platform and model an order as a state machine rather than a database row.

Minimum market vocabulary

• Bid/ask: best displayed buy/sell prices; spread is ask minus bid.
• Mid: (bid + ask) / 2; a reference value, not an executable guarantee.
• Market order: prioritizes execution, not price.
• Limit order: constrains price, not execution.
• Liquidity: ability to trade size with limited price impact.
• Slippage: execution price versus chosen reference; define sign convention.
• Maker/taker: liquidity-providing/removing behavior and often fee categories.
• Position: signed inventory; distinguish gross, net, and exposure by factor/book.
• PnL: realized plus unrealized under explicit marking, fees, FX, and funding rules.

Order lifecycle

CREATED → VALIDATED → SENT → ACKNOWLEDGED → PARTIALLY_FILLED → FILLED
    │          │         │          ├──────────────→ CANCEL_PENDING → CANCELLED
    └→ REJECTED└→ REJECTED└→ UNKNOWN└──────────────→ REJECTED/EXPIRED

Real venues vary. A cancel and fill can race. A timeout after submission produces UNKNOWN, because absence of acknowledgement is not proof of absence at venue.

Important quantities:

0 ≤ cumulative_filled ≤ original_quantity
leaves_quantity = original_quantity - cumulative_filled

Updates may duplicate or arrive late. Use venue/source sequence and cumulative quantity to make state application idempotent. Never generate extra position from a duplicated fill.

Risk before and after the order

Pre-trade controls include max quantity/notional, price collars, position limits, restricted instruments, market-data freshness, and message rate. Post-trade controls reconcile positions, cash, fills, PnL, and broker/venue truth.

Risk should not be only a function called by strategy code. Enforce critical limits independently at an execution boundary so a faulty or bypassing strategy cannot skip them.

Market data models

• Trades: executed events, not the whole available liquidity picture.
• Top of book: best bid/ask and sizes; updates can often be coalesced.
• Depth/book deltas: incremental state; one missing update may invalidate all later state until resnapshot.
• Bars: derived aggregates; define time zone, session, corporate actions, and whether incomplete bars are visible.

Drill

Receive: acknowledgement, partial fill 30, duplicate partial fill 30, cancel acknowledgement, then late fill with cumulative 40. State the final filled and leaves quantities and whether CANCELLED may still receive a venue-authoritative late fill. Explain how you prevent double-counting.

Answer frame

I model orders as explicit state transitions driven by idempotent events. I distinguish local intent, transport acknowledgement, and venue truth. Cancel/fill races and ambiguous timeouts are first-class states, and position is updated from deduplicated incremental or cumulative fill evidence.