What shipped
Newest first. Short entries — what you can now do, and whether it fully works yet. The longer stories about how any of it went live on the blog, and the how-does-this-work pages are in help.
-
We said we had looked and found nothing. Four times, we had looked in the wrong place
Eight crops on this site carried no container size at all. Each one had a note attached from whoever tried: **we searched, nothing published a figure, so we are not making one up.** We went back and checked those notes against the pages nobody had read. **Four of the eight were wrong.** **Leaf lettuce** has its own row in Maryland Extension's container table — half a gallon. Two separate people recorded that no source states a size for it, and neither had opened that page. **Fennel** is named individually on both Maryland pages, agreeing at five to six gallons. **Dill** is in Maryland's large-herb line at five gallons or larger. **Watermelon** turned up in Alabama Extension's series at one plant per five-to-seven-gallon container — a family of pages none of the original searches had touched. **The other four stand, and now say what was read.** Sweet corn is not named in any container guide we can find. Garlic has a real published figure and it is a *depth* — at least eight inches — which is a different measurement from volume and one this field cannot hold. **Kohlrabi and onion are the interesting refusals.** Alabama gives a number for both — but as plants per pot, not pot per plant. Four to six kohlrabi in a five-gallon container. Dividing one by the other would have produced a tidy per-plant figure **that no source anywhere states.** So we left them empty. **The rule this bought us: a refusal is only as good as the sources it was made against.** A written note saying "we searched and found nothing" is the most convincing thing in a dataset. It is signed, it is dated, it reads as settled — and it quietly stops anybody looking again. Those four notes were honest. They were just made against the wrong shelf of the library. Every refusal on this site now names the pages that were actually read, so the next person can check the search instead of trusting it.
-
We listed kohlrabi among the plants that shrug off a hard freeze. It does not
Our frost-protection guide had a table telling you which of your vegetables to worry about on a cold night. **Kohlrabi was in the wrong row.** It sat with kale, collards, cabbage and Brussels sprouts, under a line saying most of these shrug off a hard freeze. Kohlrabi is half-hardy — it takes damage in the high twenties, which is exactly the night that table was telling you to leave it alone. **That is the direction that costs you a plant.** Five other crops on the same table were also in the wrong row, and every one of those errs the harmless way: we told you to cover something that did not need it, and you lost an evening. Spinach, kale and onions are tougher than we said; peas and radishes are tougher than we said. **Kohlrabi was the only one we made sound tougher than it is**, and it is the only one where following our page and doing nothing was the wrong move. We are not giving you a temperature for it, because we do not have a sourced one. We know its tier and we know what that tier means; inventing a number to look precise is the thing the tier system exists to prevent. What went wrong is duller than a bad source. **We had the right answer already.** Every one of those six crops carries a frost rating in our plant database, sourced, and five of the six disagreed with the table on the page. The table had drifted away from our own data — in both directions, which is how we know it was drift rather than a deliberate simplification. The page now sends you to the frost-tolerant vegetables guide, which reads from the same ratings the database holds, so there is one list instead of two that can disagree. If you grew kohlrabi this year and lost it to a cold night, we are sorry — that one is on us.
-
We were marking about half our estimates, and the other half looked sourced
Every plant page shows a small **est** tag next to a number we could not find a published source for. That tag has been there a while. **It was appearing on roughly half the numbers that deserved it.** That is worse than having no tag at all. A mark that shows up on some numbers is a claim about every number it skips — you see three tagged values on a page and reasonably conclude the other twenty are sourced. **370 times across the site, that was not true.** The cause is dull and worth saying anyway. The page was reading a shortcut flag stamped into our data file rather than the citation records themselves. The flag missed one whole category (numbers a person set by hand) and 221 values it simply never got stamped on. **The page was checking something that stands in for the answer instead of the answer.** It now reads the citation records directly, so it no longer depends on what the export remembered. And the line at the bottom of each plant page used to say we carry "a cited fact per field". It now says how many — **"22 of the 23 values on this page carry a citation"** on tomato, 11 of 20 on orach — a number you can check by counting the tags yourself. **Some pages look thinner now, and the herbs and flowers look thinnest.** That is what our coverage has always been; the difference is that you can see it.
-
Who says this variety is drought tolerant
Every crop page now has a column saying what each variety is known for — **and who said so.** Not "drought tolerant" on its own. **"Fedco Seeds describes this as drought tolerant."** The difference matters more than it looks. **About seven in ten of these claims come from the company selling the seed.** Printed flat, that is us repeating a seller’s marketing as our own finding. Printed with the name on it, it is completely true, and you already know how much weight to give a shop describing its own stock. The other three in ten come from somewhere like NC State Extension or a university citrus collection, and those read differently on purpose — **an extension service "records" where a catalog "describes"**. We are not flattening the two into one voice, because the difference between them is the useful part. **And where nobody is on record, the row says nothing at all.** Not a bare tag, not a borrowed name — around a hundred varieties have a trait we cannot attribute to anyone, and a trait with no author is a claim we would be making ourselves. That is the thing we decided not to do. We are also showing 27 traits rather than all 155 we hold. Thirty-six of them appear on exactly one variety, and a row wearing twelve badges teaches you to stop reading badges.
-
You do not need to know the gardening word for it
**"Deadhead." "Pot up." "Thin." "Hill."** If you know those words the app understood you perfectly. If you did not, it mostly did not. We measured it rather than assumed: every sentence the app understands, rewritten the way somebody who has never gardened would say it. **31% survived.** Not because the app is bad at English — because we had been checking it against our own vocabulary, which is a test you cannot fail. So it now understands the job **described** rather than only the job **named**. *"Put some radish seeds in."* *"Put straw round the apple tree."* *"Cut some branches off the plum."* *"Took some of the seedlings out so the others have room."* *"The beans have come through."* *"The peppers are starting to form little peppers."* All of those now land on the right entry, and so do the proper terms — nothing was traded away. **The phrasings did not come from us.** We took them from published gardening glossaries: pages written by people who know the words, for people who do not. That is the exact translation direction we needed, and it is a better source than our own guesses about how a beginner talks.
-
Our lemon page was listing a mint, and a two-record rounding error is why we found it
If you looked up lemon on this site before today, we told you it had seventeen varieties. Five of them were lemon balm — a mint, not a citrus, not remotely the same plant, and not something you would ever plant expecting fruit. Lemon has twelve varieties. It says twelve now. The cause is small and stupid in the way these usually are. Our database gives every plant an id: lemon is `lemon`, lemon balm is `lemon_balm`. To find a crop’s varieties the code asked which ids start with `lemon_` — and every lemon balm does. Out of 151 crops that is the only pair where one name is the start of another, so it went unnoticed. It now matches the longest name that fits, which puts each variety with the plant it actually belongs to. What we want on the record is how it surfaced, because nobody found it by looking. Two of our own counts disagreed by two — one part of the system said 3,181 varieties and another said 3,179. Two records out of three thousand is exactly the kind of gap that gets rounded away as a rounding error. We went after it instead, and it was not two records at all: five varieties counted twice, and three counted never — three lentils that were quietly dropped because the lentil crop record itself is not finished. Plus five, minus three, and the two that showed up were the leftovers. The real number is 3,176, which means both of the numbers we had were wrong. The dropped lentils now get named out loud when it happens, rather than disappearing. And a small unexplained discrepancy in our own data is not something we intend to round off again.
-
Tell it four things and get a picture back, not a form to proofread.
Last week the chat learned to take four things in one sentence. What it handed back was four forms — which turned twenty seconds of talking into a paragraph of proofreading, the exact tax the feature existed to remove. Now it is one card, and it is a picture. Your beds are drawn as beds with the plants inside them: six tomato icons in the back bed, three peppers beside them, the new raised bed as its own box, and anything that is not a plant on a line underneath. One button saves the lot. The reason is not that it looks nicer. **You check a picture by looking and a list by reading**, and ten lines of correct-looking text get a tap and no real attention. A pepper drawn sitting in the wrong bed is wrong before you have read a word. Tap a row to fix it. Tap the small ✕ and that one is gone while the rest stay — without which one invented plant would force you to accept it or abandon four correct things beside it. And there is a box on the card that says don't ask me again. Check it and what you say saves as you say it — though it still tells you, one line, with an undo. It stops the asking, not the telling. Profile has a switch to bring the card back, because a setting you cannot reverse is not a setting. Three bugs in this one were invisible in the code and obvious the moment somebody opened it: a bed labeled with its own name twice over, a new bed drawn below a watering as though it were an afterthought, and "6 tomato" printed next to six tomato pictures. A fourth was worse and only turned up by pressing the buttons — undo left behind the bed the app had quietly made on its way to holding a plant, under a message reading "nothing was kept".
-
A place to look at your week.
The card on your home screen used to read you the first job on the list and offer one button: start, at whatever happened to be first. It spent most of its space on a task you had not chosen. It now shows the shape of the morning instead — "10 things to do", then 2 now, 1 today, 6 this week, 1 no rush, each one a button that opens just those. Tap the number itself and you get the new week view. That screen is for looking, not for being managed. Seven days across the top as small bars so you can see at a glance where the busy days are, then the days themselves down the page with the jobs on each. Today is split by urgency because today is the only day with a mixed bag. Tap any single job to do just that one, instead of being walked through everything from the top. It is also the first place your further-off reminders show up. Something you asked to be reminded about in six weeks was deliberately kept off the daily list — a job six weeks away is not this morning’s business — but it was not visible anywhere else either. Now it sits at the bottom under "further off", where you can see it without it nagging. There is no progress bar and no completion score. Looking at your week and deciding to do none of it today is a perfectly good use of the screen, and an app that scolds you for that is an app you stop opening in July. Putting it all on one page also caught something we could not see before: an apple tree was being told to mulch twice, once by the part of the app that tracks a plant’s stages and once by the part that tracks once-a-year jobs. Two different corners naming the same work. In a scrolling list they looked like two jobs; side by side on one page it was obvious. Fixed.
-
The app kept promising you a page it never handed over.
Buried in the rules the AI answers under, written the day the rules were written, is this: "if the app has given you a page of ours on this problem, point to it once." The app never gave it one. Not once, in any answer, since the day it shipped. Four guides were live, sourced, argued over and fact-checked line by line — and the part of the product most likely to need them could not see they existed. Now it can. Name a hornworm, get our page on how to actually find one at nine in the morning. Squash bugs, get the one on why spraying the ones you can see does not work. Yellow leaves on a tomato or a cucumber, get the page for that crop — they are different problems that look identical, and sending you to the wrong one of the pair would be worse than sending you nowhere. Which is the whole design. Four problems have a page today and everything else gets no link, because a link makes you lower your guard and stop checking. And the app looks the page up AFTER the answer is written, never before — show an AI a list of diseases alongside a healthy plant and it will pick one. We measured that two weeks ago and deleted the feature that did it. This one runs the other way around, so there is nothing to pick from. The bug we caught building it makes the point better than we could. The rule for yellow leaves was one line of pattern-matching, and a misplaced choice in it meant any answer that so much as said the word "leaves" matched — "early blight, a fungal disease of the leaves" came back with our yellowing guide attached. A confidently wrong citation, written by the same hand that wrote the warning against them, four lines above it. Running the thing found it in a second. Reading it had not.
-
A question we could not answer is now the note about what to write next.
The chat answers questions about the app from our help pages, and says so plainly when we have not written a page about what you asked. That answer is honest, and on its own it wastes something: you have just told us exactly where our documentation is behind. So it reaches us now, in the words you typed rather than a summary of them, because the sentence somebody actually chose is the thing worth reading in three months. Nothing is asked of you and nothing appears on screen to say it happened. One person asking once is just a question; the same question three times is a page we owe you, and only the words make that visible. We also found the chat holding two instructions that disagreed. One rule said never talk about how the app works — written after it kept narrating its own plumbing mid-answer — and the new one said to answer questions about the app from our pages. Both were in front of it at once. The fix was to say what the old rule always meant: never describe the app from your own idea of it. Read out what we have written, or say we have not written it.
-
A reminder is just a task.
Ask the app to remind you about something and it used to build you a second list. "You asked to be reminded" sat on the home screen as its own headed section, with its own paragraph explaining what it was, above the jobs the app had worked out for you. Two lists of things to do is not a feature. It is a second place to lose one. So the section is gone and the jobs are in the same place as everything else — counted in the number on the button, worked through one at a time in the same walkthrough, closed with the same tap. Anywhere you can ask for a reminder, including in the chat, now puts it there. They come near the top, because a job you chose beats one we worked out, and an overdue one comes above a job due today: the one whose day has already passed is the one most likely to be quietly abandoned. Two things still outrank them, and only two — a frost coming tonight, and seedlings still outside in the cold. Those are the ones that cost you a plant by morning. The rest works the way it did. A reminder that wants a photo still opens the camera and keeps the picture with the job that asked for it, so the answer sits next to the question. "Not needed" still drops it for good. And a job you set for six weeks from now still stays out of the way until it is close.
-
It builds beds and jobs too, not just plants
Say "I built a raised bed out back, it is 4 by 8" and it offers to add the bed, size and all. Say "remind me to prune the apple tree in December" and it offers to put that on your list, dated — in your words, not wrapped in ours. The refusals carry over: it will not convert a bed size from meters, because a bed recorded as 4 feet when you said 4 meters is a garden that does not fit, and it will not pick a date for "at some point", because a reminder that fires on a day nobody chose is the one you would have trusted.
-
The chat could answer anything about your garden and nothing about itself.
Eighteen help pages, a changelog, and the part of the app people are most likely to ask could not see a word of it. Ask how many free questions you get and it changed the subject. Now it answers — from these help pages, in the same words you would read here. How the free questions work, whether it needs a signal, where your photos live, what changed lately. If we have not written a page about what you asked, it says so rather than guessing. That last part is the design, not a limitation. **It has never been shown how the app is built — only what we have published about it.** So the worst it can be is out of date, and a stale page is a fault with an owner and a fix. A guess about our own product would sound exactly as confident as a fact about it, and there is no gardening extension service to check that against. The bit that took the longest was keeping the two kinds of question apart. "Where are my photos stored?" is about the app. "Why does it think my tomatoes need water?" only sounds like it is — that is a question about your tomatoes, and it gets answered from the plant. Building it turned up something we would not otherwise have noticed: asked how many free questions you get, it correctly said five a month for the chat and kept that separate from the fifteen-then-three for photo answers. Two meters, two pages, and it did not blend them. We only found out because we asked it.
-
It could tell you that was a sunflower and then do nothing about it.
Reported from the app: "it identified this as a sunflower and it should have some logic to let me click add plant and go through the workflow." Right — it knew the plant, and the only way to get it into your garden was to go and type it in again. Now there is a line under the answer offering to add it, and it opens the ordinary add-a-plant screen with the crop already chosen. It stays away when the plant is already in your garden. **It does not guess when you planted it, and that was the interesting decision.** The ask included "maybe assume a planting date and ask to edit or adjust", and we planned to suggest one from how far along the plant looked — showing the reasoning rather than just asserting a date. Then we went and measured whether our own records could carry it. **They hold more than we first thought.** Every one of our 151 crops records how long each stage lasts and what it counts from, and 110 of them carry days to maturity. The reason it still cannot answer is not absence, it is width: a tomato flowers anywhere from 35 to 65 days after going in the ground, and that is the tightest link in the chain — working further back to when it was sown widens it again. **A date offered with a month of slack in it is the plain question wearing a confident coat**, and a planting date is the one field that quietly ruins every other date the app works out. So it asks. Worth saying plainly, because we got this wrong twice before publishing it: our first check looked for the figure at the top of each crop record, found nothing, and reported that we did not have it. It was one level down the whole time. **A check that does not look inside returns zero and reads exactly like absence.** Two things nearly shipped broken and were caught by following the link rather than trusting it. The first version passed the plant's NAME where the screen expects its catalog id, so the button landed on an empty form — the link existed, the label read correctly, and it did nothing. The second tried to pass the variety along too, until we noticed the add screen has no variety field at all. A link that does not keep its promise is worse than no link.
-
It can see tonight’s weather now
Ask about watering, frost or heat and the answer is built on tonight’s real forecast low for your grid square — not a thirty-year average for the date. That is the difference between "it can get cold in August" and "41 tonight", which is the sentence that actually makes somebody go out and cover something. Where we have no forecast it says nothing about the weather rather than guessing. Photo answers also name a second possibility worth ruling out now, or say plainly that there is not one.
-
Not every job is the same job.
The list always knew which jobs mattered most — it has been ordering them that way since the day it shipped. It just never told you. Seven things to do, in a considered order, presented as seven identical things to do. Now the ones that matter say so. A job is marked "Now" or "Today", or it is left alone, and the button splits its count: "1 now · 2 today · 4 no rush" instead of one number that flattens a frost warning and a bag of garlic into the same seven. The part worth explaining is what moves a job up. Watering is routine most of the year. At 96°F it becomes a job for today, and the card says why — "96°F today, so this dries out about twice as fast as usual." At 104°F it becomes a job for now, because a pot in triple digits can go from damp to dead inside a day. We were already checking twice as often in heat. We were not saying it any louder, and saying it louder is the point. Nothing gets promoted silently. If a job has moved up, the reason is on its face, in a sentence you can check against your own sky — because an urgency you cannot account for is one you learn to ignore. And if we have no forecast, nothing moves at all. A guess about the weather is worse than no opinion about it. "Now" is meant to be empty most days. Frost coming tonight, seedlings still outside in the cold, pots in a heat wave — that is close to the whole list. If everything is urgent then nothing is, and a warning you have learned to scroll past is worse than no warning.
-
One cell said "sourced" and was not, so we went looking for the rest
Every value in our plant database carries a label saying how well we know it — whether we read it in an extension publication, whether a person set it, or whether it is ordinary horticulture that nobody bothers to publish. Those labels are the point. If we are asking you to trust a number, you should be able to see where it came from and go read it yourself. One of them was wrong. Blackberry’s pollination entry — whether a blackberry sets fruit on its own or needs a second variety nearby — was marked as read-in-a-publication, and had no publication attached to it. The value itself is right: blackberries are self-fruitful, and Iowa State says so in four words. What happened is that two days earlier somebody upgraded the entry, found the Iowa State page, wrote a note about it, added the page to our source list — and never linked the two. The claim and the evidence sat in the same file and did not know about each other. Strawberry’s identical entry, done the same day from the same page, was linked correctly. So we checked every other cell that claims a source. There are 3,386 of them and blackberry was the only one, which is the answer we were hoping for and not the one we assumed. Then we made it impossible to repeat: our build now refuses to run if any value claims to be sourced without a source behind it. We broke a cell on purpose to watch the check catch it, because a check nobody has seen fail is not evidence of anything. Blackberry’s page now shows thirteen sources instead of twelve. That is the whole visible change, and it is a fair description of how much of this work you will ever notice.
-
Say four things at once and it kept one.
You can tell GardenTrack what you did in your own words and it builds the record for you. Say four things in one message, though, and it built the first and dropped the other three — under a reply that said it had got them. Not refused, not flagged. Gone, quietly, which is the worst way for something to go in an app whose whole promise is that it remembers. "I put in six tomatoes and three peppers on Saturday, built a bed out back, and remind me to net the cherries in June" now comes back as four cards, one for each thing, each with its own button. Press the ones you want. Nothing is saved until you do. Two things had to change underneath, and the second is the interesting one. The app had a sensible limit of three buttons under an answer, because five stacked buttons on a phone is a wall. But it was counting two different things as one: our SUGGESTIONS — look closer at that, here is how to treat it — and things YOU had just said. Cutting a suggestion loses you an idea. Cutting the fourth thing you said loses the thing you said. The limit now counts only ours. The other half is which number belongs to which plant. "Six tomatoes and three peppers" is two plants and two numbers, and usually it can follow that. When your sentence is genuinely ambiguous you still get both cards, with the number box empty. That is on purpose: an empty box is a question you answer in a second, and a wrong number is an answer you would never think to check.
-
Scroll the varieties when you do not know the name
Tapping the variety box on a plant used to show you an empty search field, which is no help at all when you cannot remember what you planted or you just want to see what is in here. The whole list is there now before you type a thing. On the crops where we hold it, it is sorted by the shape the plant grows into — vining ones that keep going until frost, bush ones that set their crop over a few weeks and stop — because that is something you can answer standing over the plant, and "70 days" is not. Where we do not hold the shape, the list is A to Z rather than pretending. And every row now carries what its variety is known for, in plain words: takes the heat, keeps a long time after picking, no thorns.
-
It can now tell you how the plant has changed since you last photographed it.
Any app can look at a leaf and guess. **Only this one is holding the photograph you took of the same plant three weeks ago** — and until now it never looked at it. Ask about a plant you have photographed before and the answer opens with how it compares: better, worse, or no real change, and what is different in the words you would use. "The lower leaves were green in the earlier one and are yellow now." Two decisions worth explaining. It compares against a picture **at least ten days old**, not your most recent one — two photos three days apart show the light and the angle changing, and that is the commonest pair there is, because a worried gardener photographs a plant again the next morning. And **"no real change" is a proper answer, not a blank one.** That sounds like a detail and it is the whole design. We test the AI against real photographs of a real garden, and the last run had it name a problem on four of twelve healthy plants — one of them a confident leaf-spot diagnosis, well argued, on a sunflower that was perfectly fine. Something built only to report differences will find them. So the AI has to choose one of the three every time, and "it looks the same, and at this stage that is what it should look like" is one it can pick. The first version of this failed its own test and the reason was not what we expected. Given two photos of the same healthy tree it kept reporting progress — because **one photo was the whole tree and the other was a close-up of one branch.** It described what it was shown, correctly, and called it change. It read zoomed-out as grew. That is not a rare case. It is the likeliest one: you photograph the whole plant while it is fine, and you photograph the one leaf that worries you. So now it has to name the thing it can see in BOTH pictures before it is allowed to say better or worse at all — and when there is no such thing, it says so and tells you which photo would answer the question. **That is the most useful sentence in the feature**, because you can act on it in ten seconds.
-
Tell it what you did and it writes it down
"I watered the tomatoes this morning, gave them about two gallons" now becomes a real entry — on that plant, dated this morning, two gallons — instead of a nice reply and no record. It is the sentence people say most: you add a bed once and water all summer. The refusals hold as always: say "a good soak" and the amount stays empty rather than becoming a made-up number, and anything it cannot classify is filed as a note in your own words rather than guessed at, because pulling out a dead tomato is not a harvest.
-
Tell it what you planted and it builds the plant
Say "I put in two rows of Sungold on Saturday in the back bed" and the app offers to add it — crop, variety, bed and date already filled in, nothing saved until you look at the card and tap. The date is the part that matters: every job that plant ever gets is worked out from it, so the app shows you the actual day it worked out from your words, and leaves it blank rather than guessing if you said something like "in the spring". Same with how many — say "two rows" and it will not pretend that means two plants.
-
Tell it why, and it remembers the why
Say "remind me to water the tomatoes Thursday, I’m away for the weekend" and the reason is saved with the job. On Thursday you see your own words, not a bare instruction you have to work out again. What it will not do is bump the job up the list — an urgency somebody set by hand goes stale, and then it sits above a real frost warning, and then the top of the list stops meaning anything. Some things are worth recording and not acting on.
-
The app now tells you if a plant will survive your winter
We have held a hardiness zone range for every perennial crop since the database was built — rosemary 8 to 10, fig 7 to 10, blueberry 3 to 8, apple 4 to 9 — and until today the app read none of it. So somebody in a cold zone could add a fig and we would cheerfully schedule its whole year without ever mentioning that it does not live through February. That is the one question a zone genuinely answers, and we were silent on it. Now, when you add a plant that is meant to come back every year, the app says plainly whether it survives your winter outdoors — and what to do instead, with the pot size the crop's own sources give. It never stops you planting anything: growing a fig in a pot you carry into the garage is a real and sensible thing to do, and the point is only that nobody is surprised in February. Two things stay quiet on purpose. An annual gets no message, because a tomato was never going to overwinter and warning you about it is how you learn to ignore us. And a perennial that is fine where you are gets no message either — that silence is the answer. The app needs one thing from you to do any of it: your planting zone, which we ask for once, in the moment it matters, because there is no honest way to work it out from anything else we hold.
-
The plant database is free to download, and you can check every number in it
Every crop we hold — 151 of them, with 3,176 named varieties underneath — is now a free download. Spacing, sowing depth, days to maturity, container size, hardiness, when to start relative to your last frost, growth stages, pests and diseases. CC BY 4.0, so you can use it commercially, redistribute it, or build something on top of it. Credit us and that is the whole license. The part we care about is the second file. It records where every single value came from: 4,486 of 5,412 values cite a named publication, mostly university extension services, 983 of them in all, each with its title and a link. And the 926 values that cite nothing each say why — 743 because nobody publishes that fact and we filled it from ordinary horticulture, 183 because a person set it without a citation. So you never have to guess whether a gap means we did not check, or nobody knows. Two things on the page we would rather you heard from us: our own first draft of the readme claimed seed catalogs are never used for horticultural facts, and that was wrong — 194 values cite one, and the page now says so with the count. And of the 226 sourcing batches that were audited, only 96 name who audited them. We would rather print that than let the number sound better than it is. Superseded on August 7: the counts above describe the file as it shipped on this date, which was our entire internal record. The download still carries every crop and every variety and now carries fewer fields — see the August 7 entry.
-
The pot size everyone quotes is for a different plant
New guide on growing tomatoes in pots, and it starts by taking apart the advice everybody gives. Five gallons is the number you always see, and it is real — Clemson lists a five-gallon bucket, or a 14 to 20 inch container, as suitable for COMPACT tomato cultivars. That last word is the one that keeps getting dropped. In the same guidance Clemson says an indeterminate tomato, the kind that keeps growing taller until frost, will not be successful in a 12-inch pot and is difficult to grow in containers at all. So the advice is not wrong. It is an answer about a small plant, and most people asking have bought a big one. We looked at our own 115 tomato varieties: 82 of them are indeterminate. That is 71%, so the odds are against the pot you were told to buy. The guide lists all 33 determinate and dwarf varieties we hold, with days to maturity, because those are the ones a container actually suits. It also admits what we cannot tell you: Clemson says to go by the mature height on the plant label, and we hold a height for only 10 of our 115 varieties, so where a packet prints one, believe the packet over us.
-
The question box was a quarter of the screen wide and did nothing when you typed in it.
Reported from inside the app: "the text box at the top of this page is kind of weird. It cuts off a little visually and when I type my question in there it doesn't actually do anything." Both halves were real, and "cuts off a little" was generous. It had no styling at all, so the browser sized it by its own default — **147 pixels inside a 375 pixel phone**, showing "Ask anything — or leav" of a sixty-character prompt, in the browser's font rather than ours. The only unstyled control on an otherwise finished screen. It was also under the size where an iPhone zooms the whole page in the moment you tap it. And typing really did do nothing, because the question does not send on its own — it rides along with the next photo you take. That is the right behavior and nothing on screen said so. Now the box is full width and readable, it is labeled, the line under it changes to say where your question is going the moment you start typing, the button changes to "Take the photo", and **pressing return opens the camera** — which is what he was trying to make happen. The third thing in the same report — a new picture appearing at the bottom of the list instead of the top — was already fixed. We checked by putting three photos in and looking, rather than by reading the code and assuming.
-
Tap "start" and get a blank screen.
The worst kind of bug, because it looked like an answer. You tapped through to today's jobs and got an empty page — which reads as "nothing to do", not as "this is broken". It was broken. Every visit, for everyone, since the harvest-amount question went in a day earlier. One line of that change sat four lines lower than it should have, and the screen threw the moment your garden loaded. There was nothing on it to suggest a crash, so the only way to find out was to be told. Fixed, and the rule it broke is now checked automatically before anything can ship — the tool that catches it has existed for years and we simply never turned it on. It found no others anywhere in the app. Two more, found while proving the first one was really fixed. Tapping a job twice — the thing a thumb does on a phone — used to log it twice and skip the next job without ever showing it to you. A job you never saw, quietly gone. And watering a bed twice in quick succession, or changing your mind from "normal" to "deep soak" before the first tap landed, put both in your records. Now the app takes one tap as one thing done, everywhere, for every kind of entry — while still keeping two entries that genuinely differ, because a splash and a soak are two different facts about your garden and it should remember both.
-
Our tomato page stopped giving one pot size for two different plants
This morning the tomato page said "Container size · 5 gal" and stopped there. That is the exact advice our new pots guide spends a page taking apart, so we went and re-sourced it. It now reads 5 to 10 gallons. The low end is the one everybody quotes and it is well supported — Maryland Extension lists a five-gallon minimum for a standard tomato, Oregon State puts a determinate in a five-gallon nursery pot, Clemson names the five-gallon bucket. The high end comes from the same Maryland guidance, which adds the part that usually gets dropped: for an indeterminate tomato, ten gallons would be better. California's Master Gardeners say the same thing a different way, that a tomato needs anywhere from five to 15 gallons depending on which one you bought. One honest caveat, and it is on the record in our data: what none of them publish is a clean two-number rule where the word on the tag decides the pot. They size by how big the plant actually gets. Tennessee makes the point for us by putting an indeterminate cherry in the same five gallons as a bush slicer. Growth habit is the main thing driving size, so it is a good guide — but the neat determinate-or-indeterminate labeling of our two numbers is our shorthand, not a claim we are putting in anyone else's mouth.
-
"The try again button doesn't work." It did — for nine milliseconds.
Reported from a phone, with a screenshot: a chat message that would not send, saying there was no signal when there plainly was, and a Try again button that appeared to do nothing. It was doing something. We measured it: pressing it re-sends, the connection fails instantly rather than timing out, and the message is back in exactly the same state **nine milliseconds later** — with "Sending…" on screen for five of them. **A button that works and cannot be seen to work is a button that does not work.** It now says "Tried again — same problem" when the second attempt hits the same wall. Worse, the app already knew WHY and threw it away. The real reason was written to the phone the moment it failed, and the line that draws the message on screen printed a shrug instead. Four lines below it, the photo path does the opposite and prints the reason — same file, same mistake, fixed once and left alone here. It now reads "though you are online (Failed to fetch)", which is a fact somebody can act on. And the one screen where all this happens was the one screen with no way to report it. The feedback pencil had been hidden on the chat page because it sat on top of the Send button — a real collision, solved by removing the button instead of moving it. It is back, above the composer. Our first attempt at moving it still covered Send, and we only found that by opening the page at phone size and measuring the overlap.