A semester on the farm
The work behind this site started outdoors and unglamorously: a semester spent on a working regenerative pasture, where the difference between a healthy field and a tired one is something people talk about constantly and measure almost never. Farmers judge soil by how it smells, how it crumbles, how fast water disappears into it — good instincts, built on an ecosystem nobody on the field can see.
That gap is what this project is about. A teaspoon of that soil holds on the order of a billion bacteria and thousands of distinct kinds, almost none of which will grow in a laboratory. Reading their DNA is currently the only practical way to take attendance.
What the project turned into
Soil samples from under five different plants in the same pasture were sequenced, and the resulting data — along with some earlier analysis of it — became the material for a full re-analysis: a scientific manuscript, a reproducible codebase, and this site.
A good deal of that work was undoing rather than adding. An inherited set of "fungal guild" results turned out to be an artifact of matching non-fungal DNA against a fungus-only database. A headline ratio circulating in earlier drafts mixed two levels of classification and had to be recomputed. A public dataset picked as the best available comparison turned out, on inspection of its actual reads, to contain three different marker genes and no usable control. None of that is exciting, and all of it changed what the study is allowed to claim.
The habit that came out of it: check the data against itself, and write down what you cannot say. Both the paper and this site have a section listing what the study does not establish, and it is not a short one.
Why a website and not just a paper
A manuscript is written for reviewers. The person with the most direct stake in these results is the farmer whose soil it is, and a PDF full of PERMANOVA tables is not written for them.
So the same analysis feeds two outputs. Every number on these pages is filled in from the same generated file the paper's figures read, which means the plain-language version cannot quietly fall out of date when a definition changes — the usual fate of the friendly summary. How that is arranged is described on the methods page.
Whose data this is
The soil belongs to the farm, and so does what was found in it. The cores were taken from working pasture with the owners' permission and at their invitation, and the data from them — the sequences, and every result computed from those sequences, including the figures on this site — is the farm's. This project analysed that data. It does not own it, and it is not the farm's published record of its own ground unless the farm chooses to make it so.
What is not the farm's is the analysis code and the writing: those belong to the author, and so does responsibility for any error in them. Where this site interprets a result — and particularly where it says what the study cannot establish — that is the author's reading, not a statement the farm has made about its own soil.
Getting in touch
Questions and corrections are welcome through the project repository, linked at the top of every page. Corrections especially: several of the results here exist in their current form because someone questioned an earlier version of them.
Requests for the underlying data are a matter for the farm rather than for this project, so they are passed on rather than answered here. The analysis code is a separate question, and the repository is the place for it.