Field notes on my projects, techology, philosophy and whatnot.
7 posts found
Exposure logging alone is not enough for trustworthy experimentation. A/B testing needs a clean chain of eligibility, assignment, exposure, and outcome with explicit identities and consistent joins.
Small, regular upgrades reduce security and stability risk. This post explains why a maintenance loop matters more than any one bot, and how update tooling must be paired with controlled delivery and rollback.
Asynchronous processing introduces concrete operational risks like retries, ordering, poison failures, and visibility debt. This post argues for adding queues only when asynchronous ownership is part of the product, not a default optimization.
Stable APIs usually fail on inconsistent error contracts. This post proposes a boring structure using status, problem code, human-safe detail, and request identifiers to reduce contract drift over years.
Impression logs (exposure data) turn clicks into accountable metrics. Learn the request_id join pattern that makes recommender evaluation and debugging sane.
A practical guide to engineering RFCs: when to write one, how to keep it lightweight, and a template that reduces risk without slowing delivery.
A pragmatic 2026 playbook for software supply chain security for mid-sized orgs: SBOMs, provenance, dependency risk, and an automation-first good enough bar.
