Multi-Exchange BBO / NBBO Class

Citsec SWE phone-screen prompt that walks the candidate through a stripped-down market-data class. Ingest a stream of orders tagged with exchange id, price, quantity, and order type; expose per-exchange best bid / best ask and the consolidated NBBO across all exchanges. Requirements Input is a list of orders: data = [(exchange_id, price, quantity, order_type), ...] where order_type is bid or ask. Design a class with two methods: get_exchange_bbo(exchange_id) -> (best_bid, best_ask) — the best (highest) bid and best (lowest) ask currently resting on that single exchange. get_nbbo() -> (best_bid, best_ask) — the national best bid and offer aggregated across every exchange seen. The interviewer probes throughput assumptions (high update rate, frequent query) but does not formalize them; clarify before choosing data structures. Notes Naive design: per (exchange_id, side) keep a sorted container of (price, quantity); per-exchange BBO is the head, NBBO is the max-bid / min-ask across all exchange heads. For frequent queries and frequent updates, the cleanest baseline is per-exchange std::map<price, quantity> (sorted by price) plus an overarching cache of (exchange -> top_bid, top_ask) to make NBBO an O(E) scan rather than re-scanning every level. Quantity bookkeeping matters: orders may decrement quantity to zero, at which point that price level must be erased to keep the head pointer accurate. Several candidates lose points by skipping the empty-level cleanup. Order-type clarification is required up front: the prompt does not distinguish new-order vs replace vs cancel; assume the simplest model (each tuple is a resting order, cumulative quantity per price) and state the assumption out loud. Interviewer behavior: this loop reportedly pushes hard on the thinking process — no extended silence allowed. Narrate the design decisions even before writing them. Preparation Implement an in-memory L2 order book once from scratch (per-side sorted map of price levels with aggregated quantity) so the BBO extraction is automatic muscle memory. Practice giving the NBBO talk track in 30 seconds: best bid across exchanges is the per-exchange max of bid heads, best ask is the per-exchange min of ask heads; cache top-of-book per exchange to keep query cost linear in exchange count instead of total levels. Rehearse defending the quantity-zero erase step — it is the most common follow-up. Bring a one-paragraph mental model of what NBBO means in US equity microstructure (Reg NMS, consolidated tape) so the domain framing question is not a surprise.