Systems Notes
Field notes, not a feed.
Short, specific technical notes drawn from real projects. New entries arrive when there's something worth saying properly, not on a schedule.
Draw the boundaries before you parallelize
Decomposing a monolith into services and parallelizing its computation look like one project. Doing them in the wrong order makes the second one much harder.
The reconciliation layer earns its keep
A validation framework isn't a tax on a performance redesign — on the benchmark system, it's what made the aggressive part of the redesign possible.
On unfamiliar hardware, tooling isn't overhead
An assembler and debugger for a custom processor. Diagnostic tooling for a fax protocol nobody could see fail. Neither shipped as a feature, and both were the reason anything else could.
What might come next
Topics pulled from real project history that haven't been written up yet.
- What building your own assembler teaches you about every system you build afterward
- Migrating a proprietary data store to SQL Server without losing data integrity
- What "FIPS-aligned" actually requires in day-to-day implementation decisions
- Debugging protocol timing issues on shared, constrained hardware
- Automating index-rebalancing workflows: where the operational risk actually lives
- Data contracts: what breaks first when they stay implicit
- Parallelizing computation without quietly breaking correctness guarantees
- What 15+ years of C/C++ changes about how you read someone else’s code