Engineering

rrweb records the DOM. Flutter has no DOM. Here is how we rebuilt session replay for Flutter on top of the widget tree, and what the privacy trade-offs look like.

The follow-up to our SQLite observability benchmark: a full week of production-shaped traffic against the single-binary stack, with the failure modes nobody benchmarks for.

Minified stack traces are useless. Here is what actually happens between a browser stack trace and a readable one: VLQ mappings, debug IDs, and why symbolication is a cache problem.

What separates good restaurants from bad ones is consistency. When you order the same meal twice, you expect the same meal twice. That predictability is what makes them reliable. A distributed system needs the same property.

A beginner-friendly walkthrough for building your first AI SRE agent. In this tutorial we'll catch a production error, open an issue, and set up an automatic workflow that runs Claude Code to investigate and fix the root cause. When it's done it will open a PR linked to the original bug, or leave a comment explaining why the issue could not be automatically addressed.

Same $16.49/month server, same TracePath binary, two embedded databases. DuckDB writes 4x to 15x faster than SQLite, serves dashboards at 100x the row count, and stores a billion metric points in 10.8 GB. Full numbers and methodology inside.

TracePath now speaks MCP: a remote server with OAuth login built in, and a local stdio one that rides the CLI session. How it works, the three ways to authenticate, and when I'd still reach for the CLI instead.

The shape of your data can make a service fast or slow. A walk from the slow version to the fast one, built on a real example: turning unreadable production crashes back into readable code.

Post 1 only measured metrics writes. This time I ran spans, metrics, and logs against the same $16.49/month CCX13 box, and probed real dashboard queries at 100k, 1M, and 10M rows.

What actually happens between your TypeScript and a production stack trace, decoded entirely by hand.

Most observability stacks optimize for throughput your startup will never use. We optimized for accessibility instead, easy and cheap to run, and hit 58K metric points/sec anyway. Here is the full benchmark.

The missing layer of mobile observability. Open source session replay for Flutter: architecture, benchmarks, and everything I ruled out first.