A Critique of the CAP Theorem - paper
The theorem is not wrong. It is just less useful than it sounds.
Metastable Failures in Distributed Systems - paper
The failure mode that does not show up in a runbook. Systems that recover from individual faults but collapse under the combination.
Filed every assumption I was making about network behaviour.
We Wish to Inform You that Tomorrow We Will Be Killed with Our Families -
My late Dad gave me this in high school. Heavy, unsparing.
Designing Data-Intensive Applications -
The one I return to most. Replication, consistency, and the gaps between what databases promise and what they deliver.
Data and Reality - 1978
Predates the web by decades. Every problem he describes is familiar to anyone who has fought with entity-relationship modelling.
Networks drop packets at random. User behavior is a distribution. Distributed systems fail randomly. This reframed those facts as something I could reason about.
The Matrix Calculus You Need For Deep Learning - paper
The Jacobian, vector reductions, and element-wise operations explained as the physical mechanics of backpropagation. Gave me a working mental model of how error signals move through weight matrices.
Programming and Mathematical Thinking -
Discrete math - logic, sets, relations, graphs - without the formalism that usually gets in the way.
Read this in undergrad. The framework for approaching a problem you have never seen before.
Picked it up expecting hagiography. Found something closer to a manual.
Go To -
Call Me Maybe: Elasticsearch -
Most of the loopholes have since been closed. The thinking behind the analysis has not.
The series I return to most. CockroachDB in particular changed how I think about consistency guarantees in distributed databases.
Metastability and Distributed Systems -
Same failure mode as the Charapko paper, from the practitioner side.
A financial ledger built in Zig with correctness as the first constraint. The blog is the decision log.