Last Week Is Not the Last Seven Days: Defining Periodic Export Windows
How to define a weekly period independently of job execution, convert its boundaries to UTC, and handle retries and late records separately.
Notes
Why My Publisher Does Not Retry POST After a Timeout
How persisting intent before POST blocks another publication when the outcome is unknown, and why a success log alone cannot guarantee duplicate-free delivery.
Why IP Rate Limiting Starts with Proxy Trust
How a trusted proxy allowlist protects an IP rate limiter, why forwarded headers need validation, and what application tests cannot prove.
Where LLM Responsibility Ends
How a maintenance request generator separates model-written text, user requests, and rules the application enforces with ordinary code.
Where Deployment Actually Ends
Why a successful GitOps flow does not prove publication, and how checking the expected revision ties a release to the site response.
Why Quality Gates Need Tests Too
How bugs in commitlint and a shell check produced false success, and which regression tests now protect the checks themselves.
When a Contract Test Knows Too Much
How LLMs can accelerate contract-test overengineering and why observable behavior matters more than implementation details.
Documentation as a Contract: Keeping CI Documentation in Sync with the Pipeline
How to turn critical claims about CI into contract tests that make documentation drift visible when the pipeline changes.
Prompts Define the Work, Checks Define Done
Why the definition of done should move from prompts into a verifiable contract, with an independent review of the agent’s result.
Why the Second AI Agent Should Review Without the First Agent’s Context
Why independent review works better when the second agent sees the task and result but does not inherit the implementer’s explanation.