# Grows Here

> Most advice is right for the person who gave it. Grows Here checks one piece of advice against your own situation: rate the soil it grew in (the giver's safety net, spare time, appetite for risk, experience in this area, people in their corner and room to move) against your plot, see it printed as a seed packet, and get either a fit check (what it was rooted in, what each gap means for you, how sure it sounds, the plain version that fits you, a small trial) or the strongest honest case for the opposite advice.

URL: https://grows-here.skillsafe.ai/
API tutorial: https://grows-here.skillsafe.ai/api.html

## Free, in the browser
- The soil test: six growing conditions, each rated little, some or plenty for you and for the giver (the giver's side may be "not known"). The same level is a match, one step apart a small gap, little against plenty a big gap, a giver's side you do not know is unknown.
- The seed packet: the advice on the front with a sprout drawn for the verdict (in the ground, in a pot, grafted, or wilting on the compost); the growing conditions on the back - the giver's level and yours on each condition, joined by a line coloured by the gap.
- Sweeping words: a plain word list ("always", "never", "everyone", "the only way", "must", "guaranteed" and a few more) run over the advice.
- The verdict rule, checked in order: compost (2 or more big gaps), graft (1 big gap, or 3 or more small gaps), pot first (3 or more unknowns, or it would cost you a lot, or you could not easily undo it), plant (everything else). The opposite leans the same way: compost - lean to the opposite, graft - take a bit of both, pot first - let a trial decide, plant - keep the original.
- Exports: Markdown, JSON and a soil sheet (.csv, one row per condition); kept packets in a seed tin (in the browser, or in your account when signed in).

## Metered (gpt-terra), two tasks on the same advice
- fit: the verdict by the rule, 2-4 roots (each quoting the advice word for word, with what had to be true and the condition it leans on), one note for every gap the page marked, how sure the advice sounds (naming any sweeping words), the plain version that fits your plot, a trial sized to the window you chose, and one question to ask yourself.
- flip: the lean by the rule, the reasonable opposite advice, its three strongest reasons, who it is right for, 2-3 conditions under which the original still wins, the common ground, a trial that would settle it, and one question.
Both use only what you wrote: no invented facts about you or the giver, no he/she, no real names or brands, no numbers that are not in the input, and trials that are small, safe and reversible (a qualified professional for medicine, the law or large sums). The page re-checks every reply: the verdict or lean against its own rule, every quote against the advice, every marked gap, and every number, capitalised name, first-person slip and field name against the input.

## Sources
Built on two SkillSafe skills: @tjboudreaux/thinking-circle-of-competence - https://skillsafe.ai/skill/@tjboudreaux/thinking-circle-of-competence - (check where a claim's evidence ends before trusting it) for the fit check, and @tjboudreaux/thinking-steel-manning - https://skillsafe.ai/skill/@tjboudreaux/thinking-steel-manning - (build the strongest faithful opposing case, state when you would agree) for the opposite.
The trend note on the page: an essay arguing that most advice is right for the person giving it (coached.com) was on the Hacker News front page on 21 October 2026.