Writing and evidence
Implementation-backed articles and concise design notes on delivery guarantees, failure recovery, capacity, and the evidence still needed to validate each claim.
Backend reliability · Worker correctness
Why persistence precedes acknowledgement, which effects a unique key protects, and how real helper-process exits verify the two commit boundaries.
Read the articleBackend reliability · Durable acceptance
The implemented transaction and dispatcher contract for durable acceptance, confirmed publication, and duplicate-safe recovery after a real subprocess exit.
Read the articleBackend reliability · Capacity and observability
How admission, prefetch, database capacity, and retries interact, with corrected measurements of submission through observed committed results on a local host.
Read the articleAI infrastructure · Recovery contracts
A code-backed examination of atomic checkpoints, configuration identity, RNG restoration, and the exact-equality test that separates a real restart from a plausible-looking continuation.
Read the articleDistributed systems · Delivery semantics
A concrete worker contract for stable message identity, durable acceptance, duplicate suppression, bounded retries, and dead-letter handling—grounded in executable Go tests.
Read the article