Process before polish
Show the working.
Know the limits.
A confident answer is easy to produce. A useful answer needs a better foundation: evidence, context, tested assumptions, and a clear idea of what remains uncertain.
Understand the system.
Start with the decision someone is trying to make. Find the dependencies, constraints, and likely failure points. A generator’s headline wattage means little until you know what needs power. A page’s traffic means little until you understand the visitor’s next step.
Go back to the source.
Use the source best placed to answer the question: government datasets for occupational data, a regulator for licensing requirements, a manufacturer for a specific device’s capabilities. Check scope, date, units, and geography. A search result or an AI summary is a lead to investigate.
Make assumptions visible.
Separate measured values from estimates and defaults. Ask what changes if an input is wrong, and whether the result implies more precision than the evidence supports. A calculator should explain its method and make its limitations hard to miss.
Build, check, and revisit.
Test representative cases and edge cases. Compare results with independent calculations or source examples where possible. Check what actually happens in the browser, in the data, and along the user’s path. When something fails, fix the process that let it fail too.
Different kinds of evidence deserve different words.
- Researched
- I investigated sources, documentation, and assumptions. That says something about the research process; it doesn’t imply I’ve used the product.
- Built or methodology developed
- I developed or directed the tool, its research, or its method and checked its outputs. The work may involve collaborators and AI-assisted development.
- Personally tested
- I actually used or tested the product or system, with a documented method and observations. This label belongs only on work with that evidence.
- Professional expertise required
- Some questions belong with an electrician, plumber, HVAC professional, clinician, engineer, manufacturer, or code authority. Homeowner research can help prepare the questions; it cannot grant their qualifications.
Yes, I use AI. A lot.
I use AI in research, analysis, development, and editorial workflows. It helps me explore more possibilities, work through data, write and revise software, and move from a question to something testable.
The important part is what happens around the output: checking sources, reviewing assumptions, testing behavior, and deciding what is good enough to publish. A fluent answer can still be wrong. Faster production doesn’t make a weak source stronger.
I don’t claim to have personally typed every line of code. I take responsibility for the decisions I make about the work I publish.
Corrections are part of the process.
A good correction makes the work better. If you find an error, send the URL, the issue, and a reliable source or a reproducible example. I can investigate much more effectively with that than with a vague “this looks wrong.”
Send a correction or question