Talk people down more often than you talk them up.

This site exists because “what stack should I use?” is the wrong question asked of the wrong oracle. Language models have a gravity well: Next.js if you said website, Python if you said data or AI. Both can be fine. Neither is a mental model.

A short questionnaire about your idea — who it is for, what a session looks like, whether anything must be remembered, whether anything must stay awake — produces a more honest answer, and more importantly it names the distinctions. After one pass you should be able to disagree with a chatbot.

This is not a chatbot

An LLM as the oracle recreates the problem: unstable answers, framework gravity, no citation trail. If a “describe your idea” box ever appears here, it will only fill the same six questions. The tree stays the teacher.

This site is a worked example

It is a public informational site. Strangers read it. It remembers nothing except what you put in the URL. It is only awake when someone visits. It does not charge money. It does not call a model.

That maps to Tier 1 — static-info: HTML and CSS on a CDN, two small scripts for the questionnaire, no database, no auth, no HTMX. Internal notes live in docs/ in the repo.

If this were a Next.js app with a database, it would have failed its own test.

Voice

Older sibling, not thought-leader. Recommend a default. Say what you give up. Date anything that mentions prices. Do not moralize handmade HTML. Runtime simplicity, author effort, and how much your editor helps are different scores.