Electronic Trading Architect | Strategic Advisor @sisSoftware

Electronic trading systems architect | Executive Advisor | 20yr production HFT. hftAdvisory.com New York City 🗽 Joined January 2010

Tweets

7K
Followers

4K
Following

2K
Likes

774

Electronic Trading Architect | Strategic Advisor @sisSoftware

14 hours ago

Full piece: the four criteria, the measurement mechanics, the OPRA case study, and a three-tier acceptance bar. Run the three-question check against your last report. Where it breaks is the comment worth reading. electronictradinghub.com/what-and-how-t… #marketmicrostructure #hft #trading

0 0 0 44 0

View Details

Electronic Trading Architect | Strategic Advisor @sisSoftware

14 hours ago

Measuring each stage's P99 separately and summing them ignores compounding. Google's Tail at Scale: on a 100-server fan-out, 63% of requests hit at least one component at its own P99-slow moment. OPRA: P99 fell 543.5 to 57.5 microseconds, 89%, while the median barely moved.

1 0 0 52 0

View Details

Electronic Trading Architect | Strategic Advisor @sisSoftware

14 hours ago

OPRA's P99 latency fell from 543.5 to 57.5 microseconds after a 2024 upgrade. The median barely moved. Most desk latency reports would have scored that as nothing happened. 4 things a real latency number needs, that most reports miss: #hft #lowlatency

1 0 2 435 4

View Details

Electronic Trading Architect | Strategic Advisor @sisSoftware

2 days ago

One case stays hard even after all four: cross-venue reconstruction, each venue with its own clock and its own capture boundary. Full walkthrough + the six leaks: electronictradinghub.com/designing-mark… Solved cross-venue clock skew cleanly? That's the thread worth reading. #hft #marketdata

0 0 0 81 0

View Details

Electronic Trading Architect | Strategic Advisor @sisSoftware

2 days ago

Decision 4: verify fidelity before you size storage. A byte-level diff against production output turns replay from a hypothesis into a control. Every signal you trade on it, effective spread, adverse-selection markout, order-book imbalance, becomes auditable back to the packet.

1 0 1 70 0

View Details

Electronic Trading Architect | Strategic Advisor @sisSoftware

2 days ago

Feed a market-data path the same packets from 09:31:07.442, same order, same timing. It should reproduce the same output bit for bit. Most paths can't. The gap surfaces during an incident, when it's already too late to add. A design breakdown:

1 0 4 370 2

View Details

Electronic Trading Architect | Strategic Advisor @sisSoftware

3 days ago

Kernel by pass cuts network latency but the bottleneck migrates to the order processor at Scale. Here is the architectural diagnosis and a practical checklist. open.substack.com/pub/hftadvisor…

0 0 1 321 2

View Details

Electronic Trading Architect | Strategic Advisor @sisSoftware

5 days ago

Why Strong Engineering Teams Fail at HFT Put fifteen strong backend engineers on a sub-millisecond trading system build without prior HFT experience, and you’ll watch twelve to eighteen months disappear before anyone realizes the architecture is wrong. electronictradinghub.com/why-strong-eng… #electronictrading #hft #quants

1 2 12 1K 21

View Details

Electronic Trading Architect | Strategic Advisor @sisSoftware

5 days ago

@quant_arb That +100 to -100 flip is the tell. Everyone computes the same imbalance at 1s; I've seen two desks run near-identical signals and book opposite P&L purely on tick-to-trade.

0 0 0 114 0

View Details

Electronic Trading Architect | Strategic Advisor @sisSoftware

5 days ago

@AgustinLebron3 Surviving 18 months live means the veto layer held too. Every pre-trade spec I've written for an agent that can't explain itself ends up gating position delta and notional, blind to the why.

1 0 1 569 0

View Details

Electronic Trading Architect | Strategic Advisor @sisSoftware

5 days ago

@TuongvyLe12 @Stanford_JBLP @SECGov Where does the paper handle sequencer discretion? Timestamp assignment, tie-breaks, all of it happens before commit, so the layer deciding who gets filled first never enters the verifiable record.