The build log

Days to maturity is two different numbers printed identically — counted from sowing, or from setting a transplant out. We hold both for 56 crops and measured the gap: a median of 22 days, and a month or more on 20 of them. Including where our own pages still get it wrong.

The sentences most likely to be wrong in any app's help pages are the ones that say 'not yet' — because they stop being true without anybody touching them. We found three of ours in one day, and built the thing that catches the next one.

Most gardening apps answer every question about every crop. We leave a blank where nobody has published an answer — so we can count our own ignorance. Then the count misled us three times running, and there is a mechanism behind that.

The finger test — push in to the first knuckle, water if it is dry — is the most repeated watering advice in gardening. It is the test for an established plant, and it is handed to people standing over a seedbed, where the only soil that matters is above where the test reaches.

Basil next to tomatoes is the pairing everyone knows. There is a three-year double-blind trial on it, and it says the famous part is wrong and something else is right. We had that research on one page of this site and not on another, for three weeks, and nothing broke — because nothing was comparing them.

We checked 967 varieties of frost-tender crops against a 70-day countdown to first frost. 486 of them cannot finish. Not one of 55 winter squash varieties. All 54 cucumbers can. Our own app holds both numbers and has never once put them together.

Of the 115 crops our database says you can sow, 92 have a stretch after sowing where no growth stage has started yet — a median of seven days, up to fourteen. That is the same week the surface of the seedbed dries out and kills the seed, and it is the week every calendar leaves blank.

We searched the extension literature for a frost-kill temperature on every crop we hold. 27 of 151 have one, and not a single one of those is an ordinary vegetable on a frost night. Here is what the published numbers are actually for.

We shipped hardening off as a single tap. A gardener pressed it, it vanished, and he was right to be annoyed. Why a week-long process cannot be a checkbox.

Our app checks whether tonight’s low will kill your plants. It ran on all 151 crops, threw no error, and returned “no warning” every time. Nothing crashed.

Asked what you could still plant on 30 July, our engine said go ahead on 122 of 151 crops, apple and asparagus among them. Three fixes, three kinds of wrong.

We set out to source a hardening-off schedule per crop family. There isn’t one. What we found instead was better, and it exposed advice that ruins a crop.

Attempt 2 ran the two fixes we pre-registered and scored 6 of 6 on paper. We reject the pass: the model we would ship measures 45.6%, under the 50% bar.

Our free frost lookup now computes last and first frost for your exact spot, elevation-corrected from 26 years of NASA daily data. The accuracy numbers.

Five of six pre-registered criteria passed. Cache Valley failed: the model caught 19 days of a real 45-day gap where we had declared half. Nothing ships.

Our frost map corrects for elevation and is blind to cold-air pooling. NOAA stations show 45- and 62-day season gaps it misses. Here is the bar to clear.

USGS names map tiles by their north edge; Copernicus by the south. That one-degree difference put wrong elevations under 3.3 million cells of our frost map.