For teachers and extension educators

You can point a class at this site because every claim on it carries an evidence grade, names the mechanism it works by, links the sources it rests on, and says plainly which of its own numbers it does not stand behind. Where it does not know, it says so.

Milpa Gardens is a garden-planning corpus of rules, folklore checked against the literature, and species records. A person reads each source before a claim built on it is marked verified, and approves every change to the rules. The text was drafted with an LLM and then held to rules that most gardening content, AI-written or not, is not held to. This page sets out those rules in plain language, so you can judge whether it belongs in front of students.

From the maintainer: “I’m new to gardening, and building Milpa is how I’m learning it. I started it because I couldn’t find one place online that gathered what a gardener needs, whether it’s a first bed or a long-tended plot. My background is in delivering software systems, turning complex requirements into tools experts can rely on, and I’ve brought that same care here. I read the sources behind Milpa’s claims one by one and check that each says what we say it does, and a claim I haven’t read yet is marked that way. Every source I read teaches me a little more about gardening, and what I learn goes straight back into Milpa.”

What a grade means

Every rule carries one of five grades. The word is the grade, and the sentence is what it commits to. These counts are read from the live corpus of 136 rules, so they show the real distribution.

Well established 31 rules: The rule names its mechanism and is settled, by sources a person has read and checked or by plain arithmetic and geometry.
Promising 80 rules: The mechanism is named and the literature backs it. A few questions are still open.
A good hunch 19 rules: Plausible by its mechanism, but field studies are thin, so treat it as a lean.
Contested 2 rules: The evidence points both ways, or no study has tested it, so the planner does not enforce it.
Refuted 4 rules: The claim did not survive its sources. It is kept so the folklore can be answered.

A rule is graded “Well established”, “Promising” or “A good hunch” only if it can state its mechanism in one sentence, and the top two never lean on the word “helps”. A claim that only correlates, or works by a mechanism nobody can name, is downgraded or dropped. Sounding right does not earn it the top grade.

What “verified” and “unverified” mean

Verified means a person read a real source, named in the rule’s sources, and confirmed the claim, or that the rule is true by arithmetic or geometry. 127 of the 136 rules were checked against a source by a person, and 8 were settled by arithmetic, geometry or a definition. Unverified means an LLM asserted it from general knowledge and no person has checked a source. It is labeled as such and never presented as fact. None of the rules are unverified. One rule is marked as having no evidence at all, which is a person’s judgment that the literature is silent. Promoting a claim from unverified to verified requires a person reading the source, and the model is never allowed to promote its own work.

Of the 291 sources the rules cite, 270 are university extension guides, 11 are peer-reviewed studies and 10 are government and other publications. Extension guides are written by university specialists from the research, so they are good sources, but they are a step away from the trials themselves: 3 of the 104 source-checked rules graded Well established or Promising cite a study directly.

Numbers get the same treatment at the field level. A species’ size, spacing or timing figure is marked estimated or contested when it is not from a firm source. All 1002 of them are listed on What we don’t know.

What it refuses to do

Each of these is enforced, and each has a reason a student can check.

It gives no fertilizer NPK numbers without a soil test. Asked what to feed a bed, it says “get a soil test” and stops. It will not guess a number without one.
It publishes no companion-planting chart. Folk claims that fail a check against the literature, pairings among them, are marked refuted and kept as corrections: 4 rules and 14 folklore beliefs.
It never plans a planting that does not fit the space you gave it. A plant team too big for your bed is grayed out with the reason and the nearest one that does fit, never hidden and never swapped without telling you.
It invents no growing-degree-day maturity by analogy. A crop with no real GDD model in the literature does not get one, because guessing it from a similar plant would be fabrication.
It recommends no woody nitrogen-fixer without checking the state noxious-weed list first. Several are invasive, so “fixes nitrogen” alone is no reason to plant one.
It applies no elevation temperature correction to frost. On the clear, still nights when frost forms, cold air pools in low ground, so the textbook lapse-rate adjustment predicts the opposite of what happens. The corpus leaves it out.

How to cite it

Every rule, folklore and plant page carries a “Cite this page” block with a ready citation. For a rule or belief, the evidence grade is part of it. Please keep the grade when you quote a claim (“Milpa Gardens rates this Contested”), so your students see how sure it is, the same way you do.

What it is not

It is not a substitute for your local extension office or a soil test, and it is not safety or medical advice. A toxicity note is a general caution, not a poison-control reference. Its climate data is strongest in the United States. Frost dates there come from NOAA’s 1991 to 2020 climate normals at 4,350 weather stations, and hardiness zones from the USDA 2023 Plant Hardiness Zone Map, made with the PRISM Climate Group at Oregon State University. Abroad, a weather model estimates frost dates and zones in and around 2,754 cities of 100,000 people or more, and across five regions (Europe, Southern Canada, Australia, New Zealand and Japan). Farther from its readings, only daylength works. The model is ERA5, and its estimates are graded A good hunch. Canada also gets a card pointing to local gardening help. Mexico also gets its own route to an accredited soil lab. Where Milpa works has the details. It plans gardens from evidence, and it does not replace an agronomist.

You can also take the whole graded set with you: the downloadable corpus is every rule, its mechanism, its grade and its sources in one file, under a share-alike license. You can hand it to a class or check it yourself.

Get in touch

Questions, corrections and reports of a wrong claim go to contact@milpa.garden. Every rule and folklore page also has a link for reporting a problem with that claim. The sources we read are listed by publisher, with how many a person has read.

If your school, extension office or community garden starts using Milpa before the organization plan opens, write to contact@milpa.garden. You’ll get that plan free the day it opens, and a say in what it includes.

Read the rules · Download the data · Sources we read · Folklore, checked · What we don’t know