We would rather be boring and checkable than first to a keynote.
Workflow pages
Operating pages (packs, specs, loops, eval) are normative: they recommend a default. They are not lab studies. When we say “write tests first,” that is editorial judgment from watching one-shot work fail, not a 10,000-run benchmark.
You can disagree with the defaults. Change the weights. Keep some tests.
Landscape pages
Agents in 2026 describes families, not a leaderboard. We label features as announced vs something you should verify in current vendor docs. Dates on the page are editorial-pass dates.
We will not invent GA. If we cannot tell, we say so.
Scorecards
The tool scorecard is a rubric. Default weights are ours. Trials we describe are 20-minute shapes you can run; they are not paid bake-offs unless we say so (we have not).
AI assist
Drafting and research may use models. Claims still have to survive About rules: no fake studies, no secret pipeline reveal, no owner PII.
Corrections
Contact. Named product errors take priority over tone notes.
What this methodology is not
It is not ISO certification, not financial advice, not a red-team report. It is how a small publisher tries not to lie.