latency.lab
Why this course exists

The roadmap

Getting genuinely strong enough for firms like Citadel Securities, Akuna, Jane Street, Jump, HRT, or DRW takes three layers, not one book: deep modern C++, algorithms and problem-solving, and systems/performance/trading-shaped projects. This page is that plan, made concrete.

The one thing to avoid

Don't start with a book called "Build a High-Frequency Trading System in 30 Days." That's backwards. The order is C++ → CS fundamentals → algorithms → systems → performance → trading systems — not trading buzzwords first.

Project progression

Eight levels, each harder than the last

Vector<T> to a low-latency matching engine and market-data system.

1C++ mastery — build the containers firstTinyVector, TinyHashMap, MemoryPoollive now2Algorithms, with the discipline of asking whyA real lesson per topic, then curated practicelive now3Limit order bookAdd + cancel graded now; matching nextlive now
4Matching engineSequencing, fills, execution reportsplanned
5Market data engineParse → normalize → book → strategyplanned
6ConcurrencyAtomics, memory order, lock-free SPSCplanned
7Performance engineeringCache locality, branch prediction, false sharingplanned
8Interview simulationOne problem, progressively extendedspec written
01
C++ mastery — build the containers first
TinyVector, TinyHashMap, a fixed-block MemoryPool — before trusting the STL versions, build your own and feel why each design decision exists.
02
Algorithms, with the discipline of asking why
Not a LeetCode dump — every topic teaches the tool first: what it is, why it works, the pattern that tells you when to reach for it, and the mistakes people actually make. Problems come after, graded on complexity too, not just correctness.
03
Limit order book
BUY/SELL, LIMIT/MARKET, GTC/IOC/FOK, submit/cancel/modify, price-time priority. Add + cancel is graded now; matching (the next slice) isn't built yet.
04
Matching engine
Order book → sequence numbers, timestamps, partial fills, execution reports. Then: how many orders per second?
05
Market data engine
A simulated feed: parse → normalize → book → strategy, at millions of messages, measuring p50/p95/p99/p99.9.
06
Concurrency
std::thread, atomics, memory_order — then a lock-free SPSC queue between producer and consumer.
07
Performance engineering
Deliberately make it slow, then explain why: cache locality, branch prediction, false sharing, pointer chasing.
08
Interview simulation
One problem, progressively extended — "now make it thread-safe," "now 10 million orders" — not disconnected puzzles.
Reading list, in order

The stack

One book to learn C++ properly is not enough on its own — this is the curriculum around it.

  1. 01C++ Primer — Lippman, Lajoie & MooThe foundation. Not how to copy an order book — why the code works.
  2. 02Effective Modern C++ — Scott MeyersTurns "I can write C++" into actual expert instincts.
  3. 03The Algorithm Design Manual — Steven SkienaA method for approaching a problem you've never seen, not a solutions list.
  4. 04Computer Systems: A Programmer's Perspective — Bryant & O'HallaronWhat the machine is actually doing underneath your C++.
  5. 05C++ Concurrency in Action — Anthony WilliamsThreads, atomics, memory ordering — done right, not by accident.
  6. 06Systems Performance — Brendan GreggHow to find out *why* something is slow instead of guessing.
  7. 07Designing Data-Intensive Applications — Martin KleppmannDistributed/data-system thinking — the market-data-feed layer.
  8. 08The C++ Interview Book — cppinterview.comInterview-specific drilling — after the foundation, not instead of it.
A different angle

The quant-finance track

Pricing and derivatives, not systems engineering — a real alternative if that's the actual target, not a substitute for the reading list above.

  1. Introduction to C++ for Financial Engineers — Daniel J. DuffyBinomial models, Black-Scholes, Monte Carlo — the pricing side, not the systems side.
  2. C++ Design Patterns and Derivatives Pricing — Mark JoshiClean OO structure for complex payoffs — a different skill than a matching engine.
  3. High-Performance C++ for Finance — —SIMD, hardware optimization, bridging C++ engines with Python/R.

What actually gets evaluated

  • Design, programming, code review, and extending one real problem as constraints change — not disconnected puzzles.
  • C++ fundamentals: STL, memory management, low-level optimization, multithreading.
  • The ability to reason out loud: "why is this slow," "what happens at 10x," "now make it thread-safe."
  • Roughly 250–350 carefully chosen problems beats 1,000 solved mindlessly — the goal is "I've never seen this exact problem" not being scary.