Short posts about software, tools and slow thinking
Before writing a new check, look at what already fails when the thing breaks. Most systems already tell you, just not loudly and not in the place you are looking. Reading the existing signal is free. Adding a new one costs maintenance forever.
One file, append-only, dated entries, no structure beyond that. The value is not in reading it back, which I rarely do. The value is that writing the line forces me to decide what actually happened, and that decision is usually the hard part.
Configuration that computes things is code without tests. If a value can be written literally, write it literally. The duplication you are avoiding is cheaper than the evening you will spend working out why the computed value is wrong on one host.