Made it to the top #1 new releases! " The Electronic Trading Stack " Collected Field Notes on Architecture, Latency, and Execution Quality in High-Frequency and Institutional Trading Thanks all for the feedback received!!

My last book is a collection of all the articles I've been writing in the past year. Rearranged for easy reading. All printed in a book and in one place. Stays next to my desktop for reference. (online articles remain fully for free, so having this book is not necessary for reading them) The book: The Electronic Trading Stack: Collected Field Notes on Architecture, Latency, and Execution Quality in High-Frequency and Institutional Trading Find it on Amazon now!!

Today, retail traders get live level 2 quotes on their phones. In the 80s and early 90s, access to market data was not as early accessible: • Reading ticker tape and chalkboard boards • Green-screen Quotron machines with 15-minute delays • Early Reuters and Bloomberg terminals costing fortunes If you had the data 30 seconds (HUGE) before someone else, you had an infinite edge. How the flow of market information changed over the decades. 👇 Source: Daily Mail — Experts By Decade: Wall Street Trader Life #marketdata #bloomberg #quants

When electronic trading arrived, traders thought it would never replace human instinct and relationships. Then Instinet, Island ECN, and matching engines showed up. Spreads collapsed, pit trading evaporated, and human market makers were replaced by low-latency algorithms and high-frequency trading firms. How people felt back then when the biggest technological disruption came down in Wall Street? 👇 Source: Daily Mail — Experts By Decade: Wall Street Trader Life #electronictrading #HFT #fintech #wallstreet

How people traded before optical fibers and electronic matching engines. 1. Client calls broker on a landline. 2. Broker scribbles an order on a physical paper ticket. 3. A runner sprints across the floor. 4. Shouting and hand signals in the pit with a specialist. 5. Trade executed, stamped, and phoned back. Total round-trip time: minutes (if you were fast). Today: microseconds. Veteran traders break down how order execution transformed from human adrenaline to server racks.👇 Source: Daily Mail — Experts By Decade: Wall Street Trader Life #trading #stockmarket #orderflow

What were trading desks like in the 1980s? Cigarette smoke, shouting matches, paper flying everywhere, and desks jammed shoulder-to-shoulder with 5 landlines ringing simultaneously. How trading desks evolved across 4 decades: 1980s vs 90s vs today. 👇 Source: Daily Mail — Experts By Decade: Wall Street Trader Life #trading #wallstreet #finance #marketmicrostructure

Six Study Tiles, One NASDAQ Day: The Theory, the Screen, and What the Data Supports visualhft.substack.com/p/six…

"Step 1: Measure Judge Accuracy". @sh_reya and @HamelHusain , Evals for AI Engineers, chapter 5, on estimating true success rates with imperfect judges. "The investment firm shall review its automated surveillance system at least once a year to assess ... its ability to minimise the generation of false positive and false negative surveillance alerts." MiFID II RTS 6, Article 13(6), which covers investment firms engaged in algorithmic trading. Surveillance gets step 1 at least once a year. An LLM judge should also get it every time its own model changes.

This is Thariq Shihipar of Anthropic SEC Rule 15c3-5, the market access rule: "The financial and regulatory risk management controls and supervisory procedures ... shall be under the direct and exclusive control of the broker or dealer ..." At a broker-dealer, that decides which constraints you get to delete.

Susquehanna took Platinum at CppCon 2026, and Gold went to Bloomberg, Citadel Securities, Hudson River Trading, Jump Trading and Optiver. Five of the six gold sponsors trade for a living. #cppcon #cpp

India vs HFTs interesting video from Bloomberg Originals #hft

How to use guardrails-ai at its FULLEST potential to build validation pipelines like hedge funds: this is guardrails-ai on GitHub, 7.4K stars. I read through the whole repository. From its creator: "the first of its kind benchmark comparing the performance and latency of 24 guardrails across 6 common categories ..." and, "this is the EXACT validator i have run for 5 days & results are INCREDIBLE so far" It prints its latency benchmark, cites on_fail as what its validator checks against, and never executes that check. Its validator fires when regex matches, and on_fail with noop passes every time. A pre-trade risk check does not have a noop option. Five milliseconds on the execution path does not wait for a fix. Its benchmark table of 24 validators runs from toxic language to regex. The risk gate at the perimeter is not on it.

Exchange feed drops one packet: the venue fix is discarding your order book, pulling a 5,000-level snapshot, and buffering deltas. Takes 80ms to >1s. Quoting through is fail-open. The buffer timeout that survives volatility without thrashing is the number worth comparing. #hft

Inside Bracket22 Future of Wall Street: one single hedge fund manager, working with one or two people under them, supported by a swarm of AI-agents beneath. What do you think? Fiction or real?

I had the chance to work with a team reviewing an exchange to move from Erlang to C++. Their C++ engineers did not understand RabbitMQ. The recovery boundary was still open. Nobody had written down what the system rebuilds after a crash and they've described a recovery scenario in which the existing system could come back within one or two seconds. The setup was a RabbitMQ cluster sitting behind it and was described as persistent. The plan was to replace components one at a time, in a progressive way. RabbitMQ persistence is useful, but It's not a recovery contract on its own. The contract has to say what the trading system rebuilds, where it gets the record, and what a client sees while that happens. Without that, a restart will produce an operations problem, even though the technical improvement is real. The C++ work could continue one component at a time. Each component needed defined results before traffic moved. A lower-latency component must preserve accepted-order state when a process disappears. For reference: in June 2025 MIAX Sapphire's test environment connected to production and injected data into its live matching engines, and, trading stopped. What happened next? Fifteen minutes of trades were canceled. The exchange reviewed +2,200 claims and proposed a $525,000 liability cap. This may seem obvious, but not always is. Before approving a rewrite, kill the process mid-session. Four checks: acknowledged orders present, no duplicates, ordering held, rebuilt state matching the client's. Until that test runs, the recovery boundary is still open.