The model never writes the advice
This is the part most worth knowing, because it is unusual and it is the reason the rest of the page can be short.
When you photograph a patch, the vision model does exactly one job: it distributes confidence across a closed set of 16 causes and guesses the species. It does not write a word of what you read afterwards. Every sentence about what is happening to the grass, every physical check, and every step of every treatment plan was written by hand, in advance, and reviewed before it shipped. A cause the model tries to invent outside that set is dropped rather than shown to you.
The Sward team builds Sward, and we build software rather than study turf for a living. That is not an aside — it is the whole argument for the design above. Advice generated on the spot by a language model is advice nobody checked, and nobody could have. Advice written once, in a file, is advice that can be read carefully, argued about, corrected, and compared against a published source. So that is what the app contains.
The horticultural facts here and in the app — how grubs feed, what a dormant crown looks like, how much water a lawn wants in a week — follow the published guidance of university extension services. Every article in the library names the ones it drew on and links them, so you can go and check rather than take our word for it.
Why the articles refuse to be certain
The app is built around a rule that is easy to state and was hard to keep: it names a single most likely cause only when the evidence genuinely favours one, and otherwise it shows a shortlist and the physical check that separates the candidates. An app that always picks a winner is guessing on the days it does not know, and you cannot tell those days apart from the rest.
The same rule binds this site. An article here will tell you what separates two causes and how to check which one you have. It will not tell you which one you have, because it cannot see your lawn — and it gives you the check for free, in full, rather than holding it back until you install something.
What we get wrong
Probably several things. If you find one, write to us — and if it is about a diagnosis the app gave you, say what the lawn actually turned out to be. That is the single most useful thing anyone sends us, and it is how the written material gets better.