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.
-
The Talk screen stops claiming to know a garden you have not planted yet
The empty Talk screen said "it knows what day it is, what you have planted, when it went in, and what you wrote down" — to somebody who had just installed the app and planted nothing. It was describing what the app becomes after you have used it, to the one person who had not. It now says what is true on a new phone: there is nothing of yours for it to go on yet, and adding a plant is what changes that. Add one and the sentence changes on its own, because the screen counts what you actually have. Two more in the same family. Talk needs an account, and the only thing that ever said so was the failure after you had written a question and pressed send — the requirement is on the empty screen now, with a link, and the box stays live because somebody signed out with a real garden is a real gardener rather than a lapsed one. And Profile told you "everything works, but it lives only on this phone" while one of your four tabs did nothing at all; it now names the exception instead of softening the promise, in the words the sign-in screen has been using correctly all along. Profile also stops opening on two settings for features you may never have touched — your account and where you garden come first, because they are what decide whether any of the rest works.
-
The Talk screen stops offering three questions it cannot answer
Open Talk on a brand-new phone and it suggested three things to ask, every one of which was a dead end. "What did I miss" lands on a screen about yearly jobs like pruning, which belong to trees and bushes you have not got yet. "When will things be ready" has no plants to have dates. "What should I do this week" has no jobs to list. Three doors, all painted on. The screen now offers the single thing that makes every one of them work — add your first plant — and the three come straight back once you have. The wording changed with it, and the first attempt at that was ours to fix too: this screen used to claim it knew what you had planted before you had planted anything, and the sentence we replaced it with opened by telling you what it could not do. Honest, and about as welcoming as a closed shop. It leads with what it is for now and ends with the tap that gets you there, in half the words. Nothing is switched off either way: the typing box and the microphone work from the first second, and you can ask anything you like. The suggestions are simply not offered when we already know the answer would be empty.
-
"Can I still grow this?" is worked out now, not guessed at
Asked whether sweet potato slips were worth planting in mid-August, the chat said there were 61 days until first frost, that sweet potatoes need 90 to 150 days, that this was "tight but workable" — and suggested planting them right away. Both numbers are correct and the conclusion is not: 61 days against a 90-day minimum is 29 days short of the fastest variety there is. Not smaller tubers. No crop. Now the app compares your frost date against the crop's own days-to-maturity before the question is even sent, and hands the AI a settled verdict — yes, no, or risky — which it leads with and then explains. When the answer is no it says what you can do instead, like potting the slips up for next spring. On that exact question the honest answer came back 4 times in 15 before and 14 times in 15 after. One limit we would rather state than hide: the AI still sometimes writes its own countdown to frost inside the paragraph and gets it wrong — 69 days where the true figure is 61. The verdict is computed and reliable; a day count in the prose is not, and your Today screen has the real number. That half is not finished.
-
Buttons you can actually hit
We have had a number written down since the app had its first screen — a control should be at least 44 points, which is roughly the pad of an adult thumb. Nothing checked it. Measuring the real app on a phone-sized screen found 66 controls smaller than our own rule, including the Start button on your daily list at 21 points tall and the two attachment buttons in chat at 36 by 36. Twenty-four of them were not oversights. They were written with a smaller number typed in, in the same file as the rule that says 44. Those are fixed, and more usefully, controls now start at that size by default and a check refuses to let a smaller one be written. The daily list, the hardening card and the add-a-plant screen are now clear. Some links are still short and they are next — a link behaves differently from a button and needs a different fix, which is the honest reason they are not done yet rather than an oversight.
-
Fill in a plant you already have, by saying it
This morning you could tell the chat which kind a plant was. Now you can tell it the day it went in, how many you put in and which bed they are in — and all of it at once. Say "I put six Red Russian kale in the back bed on Saturday" about kale already in your garden and you get one card listing exactly what it is about to write, with one button. Before today the app could record all four while you were ADDING a plant and none of them on a plant it already had, so the only way to fill a gap was to go into the plant and tap it in. The date is the one worth having most: every job and every ready date is counted from it, so a plant without one gets no dates at all. Three things it deliberately does. It fills blanks only, field by field — say four things about a plant that already knows three and only the empty one is written, so the card can never quietly correct you. It shows the date as a real date before you save, so "Saturday" reads as 9 August 2026 rather than being taken on trust. And a phrase that names no day, like "in the spring", writes nothing rather than guessing.
-
Hardening off has a way in now
Somebody reported that they could not work out how to reach the hardening off page. They were right, and it was worse than hard to find: the only link to it anywhere in the app was a card on Today that shows up while a run is under way. No run, no card, no route in — so between runs the page, and the reminder times you had already set, could not be reached at all. Profile now has a Hardening off card with two links: one to the page that tracks your runs, and one to the reminder times, which reopens the short setup so you can change when you are nudged to put trays out and bring them in. The same links sit on the hardening page itself, including when nothing is running — an empty screen that only explains itself is a dead end. The setup was already skippable, already remembered your answers, and already enrolled later plants without asking again. What it did not have was a door.
-
Hardening off moved to the top, and the buttons now do the thing
A hardening week now has its own block near the top of Today, above your garden tasks, with a purple edge so you can find it without reading. It is there because it is the one job in the app where being an evening late can cost you the plant. The bigger change is what the buttons do. "Out they go" used to send you to another screen to press it again; it now records the tray as out, on the home screen, with an Undo underneath. The offer of a bring-in alert still appears — after the tray is written down, so dismissing it loses nothing. That offer was the whole reason the button was a link, and the honest fix was to carry it here rather than drop it. Three things had been scattered by sorting one week across four levels of urgency: a tray already outside, today's step, and "the Sungold is hardened off" — the end of a good week, which came out filed under No rush. They are one week and now they sit together, on Today and on the week screen both. Two quieter fixes. When the day is done the row says so with a tick and stops asking, instead of disappearing or growing a button. And if you open the app late enough that today's step would run past dark, it now tells you the tray goes out tomorrow morning — that sentence was being worked out correctly and shown to nobody, so from about five in the afternoon the only hardening thing on the screen was a link about repairing past days.
-
How much do you want to gamble?
Your last frost date is an average, and an average hides the thing that actually bites you. The same 800-foot square in Utah runs from early April to early May depending on the year — thirty-four days of spread in one spot, standing still. Every gardening app in the world prints the middle of that and says nothing, and so did we. Setup now asks one question instead: play it safe, middle of the road, or push my luck. Play it safe counts against the date the frost is done nine years out of ten and costs you about two weeks of season. Middle of the road is the date on the back of the seed packet, and it is what you get if you never touch it — nobody’s dates moved because we shipped this. Push my luck takes the early end, and says out loud that most years you will be covering things at least once. Pick one and watch the growing-day count and both dates move on the screen in front of you. From then on it is not a note in a help page: every date in the app counts against the end you chose — the season chart, the plant pages, the home screen counter, what the app says when you ask it a question. It also knows where it cannot offer the bet. Along the Gulf Coast, in coastal California and across the southern tier there are winters that never freeze at all, so the records genuinely cannot produce an early date. Houston gets two buttons and a sentence explaining why, rather than a third that quietly hands back the same date as the middle one.
-
It can tell you your plant is fine — and say what to look for when it is not sure
Photograph a healthy plant and the app used to find something wrong with it anyway. A good sunflower came back with a fungal leaf spot. A healthy corn plant was said to be losing out to a nearby tree. Three of thirteen real garden photographs, all wrong in the same direction, and not one of them hedged — which is the worst shape a wrong answer takes, because nothing tells you it apart from the ones that are right. The cause was not carelessness. It had no way to say nothing was wrong: the vocabulary existed in our own data and the instructions never mentioned it, so a clean leaf arrived as a puzzle to be solved. It can say a plant is fine now, and say what you are actually looking at — the lower leaves yellowing because that is what a tomato does to its oldest leaves all season. Two things keep that from becoming a blind spot. Normal covers color and age, not a change of shape: puckered, thickened or distorted growth still gets looked at properly, which is what peach leaf curl looks like and exactly what a softer app would miss. And where a plant is growing is not an illness — corn planted under a tree in long grass really is competing for light, and you should be told that, but it is an observation about the spot rather than a verdict on the plant. The third change is the one we care most about. When it genuinely is not sure, it no longer asks you to go and take a better photo — that is the app asking you to do work for it. It names what it might be, tells you the signs on your own plant that would confirm or rule it out, and says what to do if it turns out to be that. "This might be peach leaf curl — watch the new growth as it comes in, puckered and thickened and often reddish rather than just pale, worst after a cool wet spring." Something you can act on this evening, rather than a shrug. Across the same thirteen photographs, confidently wrong answers went from three to none.
-
It works out which plant you meant, even when you do not say
Telling the chat about a plant you already have — the kind, the day it went in, how many, which bed — only worked if the crop got named along the way, because that word is how the app finds the plant. Say "it is Red Russian" with no "kale" in the sentence and you got a friendly answer and no button. Measured against the live model, that was happening about twice in every eleven tries: a correct answer, and nothing to press. Now the app does the part it can do without asking. If you named no crop and exactly one plant in your garden is missing everything you just mentioned, that is the only plant the sentence could be about, so the card appears. Two facts are treated as one question — a plant missing the kind and a different plant missing the date are two candidates, not one, and it will not write half your sentence onto the wrong plant. If more than one plant could be meant, or none, it stays quiet exactly as before. It can only ever turn a dead end into a button; it can never turn one button into a different one.
-
Light or dark, and the app now opens light
GardenTrack has been dark since the first screen, which reads well on a phone in the evening and badly on a bright morning in a garden. There is now a switch on your profile, and the app opens light for anyone new. Two things about it worth knowing. It is remembered on the device you set it on rather than on your account, so a phone you use outside and a laptop you use at the kitchen table can look different — which is usually what you want, because the reason to pick one is the light you are standing in. And it is applied before the first pixel is drawn, so opening the app does not flash the wrong color at you for a fraction of a second the way a lot of apps do. Every screen has been checked in both, one text color at a time, against the contrast standard for readable text.
-
It stops offering to back things up when you have no signal to do it with
If you were not signed in, the line at the bottom of the screen offered to back your changes up whether or not you had a connection — and with no connection that button could not do anything, because signing in works by emailing you a code. It now says the true thing instead: your changes are saved on this device and safe there, and signing in to back them up needs a connection. The count itself was never wrong. Those changes really are on your phone and really are not backed up anywhere else, online or off. It was the invitation that could not be honored. When your signal comes back the offer returns on its own, without you reloading anything.
-
A bed you described reads the same on both screens
Describe a bed and the review row said "raised bed · 4x8" while the card one tap underneath said "4 ft × 8 ft" — the same sentence, spelled two ways, one screen apart. The row was reading your words back and the card was showing what had actually been understood, and side by side that looks like two facts rather than one. The row now says what will be saved, however you typed it: "4x8", "4 by 8" and "4 × 8" all read the same. One case deliberately still shows your exact words — a size we cannot read, like "4 by 8 meters". We only record beds in feet and will not convert a number we might be guessing at, so the boxes stay empty, and seeing what you actually said is the only thing that explains why.
-
One plant, not "1 plants" — and a question that was using your own words back at you
A bed with a single seedling in it read "1 plants" on the plant page. Signing in with one unsaved change offered to back up "1 changes". Neither is hard to write correctly; both were written by somebody looking at a screen with three of the thing on it, which is the only state anyone ever tests. There is now one small helper for counting things and the three places that got it wrong use it. The bigger one: the app asked "which kind of may tomato?" It was building the question out of YOUR name for the plant rather than the crop's name. Call something "May tomato" and that is what came back; a plant already showing its variety got "which kind of tomato · sungold?", which asks which kind of a thing that is already a kind. It only went wrong for a crop we do not carry, because everything else had a real crop name to use. There is no honest way to name a crop we do not hold, so the question stops trying — it reads "which kind is it?" now. It was reported on the plant page. We found the second one, on the garden list, by going to look for it. And on the add-a-plant screen, the gray heading above the crop list — "ready to plant this week" — was being laid out by a rule written for the rows underneath it, so its explanatory half sat in a second column and wrapped badly on a phone. It is one sentence again.
-
Past conversations you could not read
Open the list of your past chats and the words were not there. Not missing — invisible, pale text on a pale panel, the whole list a blank rectangle. It lasted a few hours and it is fixed. The cause is a small thing worth explaining, because it is the kind of fault that waits. That panel had always named the wrong color for its background — it asked for the color we use for text. That did nothing at all for as long as the text color had no name in the app, because the line carried a spare instruction to fall back on, and the spare was correct. This week we started giving our colors proper names, so that the app can offer a light theme as well as the dark one. The moment the text color had a name, the panel got what it had been asking for all along, the spare instruction stopped applying, and it painted itself in the color of its own writing. A mistake that had been sitting there harmlessly became visible the day something else went right. We checked whether anywhere else in the app does the same thing. Nothing does.
-
Photos you never sent no longer back themselves up
Earlier today we fixed a photo that was uploaded after you removed it from a question. This is the other half, and it is the commoner one: attach a photo, then close the app or wander off to another screen and never send it. The picture stayed on the phone, still queued, and went up to your account at the next bar of signal — a photo you never sent, from a question you never asked. The app now clears those when it opens, and it is deliberately fussy about which ones. It will only take a photo that is more than a day old, is not part of any journal entry, and is not attached to any question — including one still waiting to send. Anything in progress is left alone, and it runs on opening rather than on leaving a screen, because leaving a screen is exactly what somebody does halfway through asking something. It also keeps a record of what it removed and when, so that if a picture is ever missing we can tell you what happened to it rather than guess. A deletion nobody wrote down is impossible to tell apart from a bug.
-
Pinning your garden no longer disappears when you hide the setup list
Hiding the setup list used to hide all four items, and one of them should never have been in that group. Everything else on it costs you something you can see if you skip it. Pinning your garden costs you something invisible: every journal entry records roughly where the plant should have been on that date, worked out from your frost dates — so with no location, every entry you write is saved with that part permanently blank, because we never recalculate a stage once it is written. Now hiding the list hides the other three and leaves this one as a single quiet line. Nothing is blocked and there is no pop-up; it simply does not go away until the app knows where you garden. Pin it and the line goes for good, and the entries that were already blank get filled in at that moment — each marked as worked out later, so you can tell them from a stage recorded on the day.
-
Plants go into the bed you just described, before we ask its size
Say "I built a new raised bed and put a row of lettuce, spinach and kale in it" and you used to get three loose plants and a Save button — no bed. The bed only appeared a turn later, once you had answered a question about how big it was, and only then did the plants gather under it. Take the button in between and you got three plants belonging nowhere, plus a bed arriving afterwards that had to be joined up by hand. The grouping was something we already knew; the dimensions were a detail we did not, and there was no reason to hold the first hostage to the second. Now the plants land in the bed straight away and the size question is just a question. Two things it deliberately will not do: if you name a different bed for a plant, that wins, because you telling us beats us guessing — and if you describe two new beds in one message it leaves the plants unfiled and asks which, since a plant in the wrong bed is worse than one not yet in a bed. It looks finished.
-
It no longer offers to record a pest it was only reminding you about
Ask when things will be ready to pick and the answer might mention the codling moth you logged last week — which is the whole point of it knowing your garden. What should not happen is a row of buttons underneath offering to record a codling moth, as if it had just been found. Your journal is given to the AI so it can answer from your own history, so naming something out of it is quoting you back to yourself, and saving that would file the same sighting twice on a day it never happened. The buttons now only appear for something it can tell from the message you just sent or the photo on it. Spot a genuinely new pest in a new photo and the button is still there.
-
Starting your garden over replays the setup
The new first-run walkthrough had a rule that sounded sensible and was wrong: it only ran for people who were not signed in. Signed-out was standing in for "new", and the two are not the same thing. Anyone with an account who wiped their garden and started again got dropped onto an empty list with nothing to guide them — which is the exact moment the walkthrough exists for. What it asks now is whether the garden has been set up, not whether you have an account. So restarting from Profile with Clear everything puts you back at screen one, the same as a brand-new gardener. Keep my location still does not, and should not: it leaves your frost dates in place, so there is nothing left to ask. Skipping the walkthrough once no longer blocks it either — a skip in March is not a decision about a garden you start over in July. One nuance worth stating, because it is the sort of thing that goes wrong quietly: if you are signed in on a new phone, the app waits until it has heard from the server before deciding your garden is empty. An empty phone and an empty garden look identical for the first second or two, and walking someone through setting up a garden they already have would be worse than saying nothing.
-
Say which kind it is, instead of going into the plant to fill it in
Say "my kale is Red Russian" about a plant already in your garden and you used to get a friendly reply and nothing else — no button, no record. The same sentence about a plant we had never seen produced a card. The app could record a variety while you were ADDING a plant, where it rode along as one detail of the new row, and had no way to record one on a plant it already held; the only route was to leave the conversation, find the plant and tap into it, which is the tapping the chat exists to save you. Now a button appears offering to save it onto the plant you named. If the kind is one of ours it links to the real variety and your dates sharpen, exactly as picking it from the list would; if it is not, your words are kept and the app stops asking. Two things it deliberately will not do: it never overwrites a kind already recorded, because a passing sentence should not be able to replace something you chose by hand, and it never guesses at half a name — "Red" does not become "Red Russian". Fixed on the way past: five varieties are named exactly what their crop is called — Lovage, Mizuna, German Chamomile, Sweet Marjoram, French Tarragon — and picking one wrote a label that trailed off after the separator with nothing behind it. The kind you had just chosen simply was not there.
-
Search the plant database for a wax bean and you now get a wax bean
The plant database holds 3,174 named varieties and, until today, you could not reach one by typing its name. Search read crop names only. So "wax beans" found nothing at all — not because we do not hold it, but because no crop is called that: it is a variety sitting inside bush beans, and you had to already know that to find it. Which is backwards. The people most likely to type a variety name are the ones who know exactly what they are growing. Search now reads every variety name we hold, and the crop card tells you which variety you matched, so you know why it came up. Type zucchini, land on summer squash. Type cherokee purple or black krim, land on tomato. Three smaller things came out of testing it on real phrases rather than on the ones we invented. Plurals work, because a person types "wax beans" and every variety name in the database is singular. Word order does not matter, so "purple cherokee" finds it too. And if you type the name of a crop we hold, that crop comes first — before we fixed the ranking, searching "tomato" put Purple Coneflower at the top, on the strength of a variety called Tomato Soup. A real match. Never the one you meant. This came from a gardener who went looking for wax beans, gave up on the search, and found them by opening bush beans and scrolling.
-
Settings are switches now
The alerts control we shipped in the morning was two things pretending to be one: a line of text telling you the state, and a red link underneath to change it. Neither looked like something you could press. Two cards further down, another setting was a plain blue browser checkbox sitting on a screen with a full palette and three typefaces of our own. Both are now switches — one thing that shows you the state and is also how you change it. The rule behind it: anything you can turn on or off gets a switch. A link that says a verb is for things that are destructive or hard to undo, like deleting a plant or restarting a garden, where stopping to read the sentence is the whole point. A setting is not one of those. Three of them across the app, all changed. The alerts switch also stays visible when it cannot move — grayed out, with the reason underneath — because a control that disappears teaches you the feature is missing, and one that explains itself teaches you what to fix.
-
You can run the setup walkthrough again whenever you want
Skipping the setup walkthrough used to be permanent. That was on purpose — a skip should mean something, rather than the same screen nagging you every time you open the app — but it went too far. A decision you can never revisit is not really a decision, and one accidental tap on the X cost you the whole thing for good. Profile now has a card called Walk me through setup again. It runs the same five screens on any garden, in any state: skipped, half-finished, or three years old with two hundred plants in it. It deletes nothing. It updates your frost dates if you give it a location, adds a plant if you name one, and leaves the rest of your garden exactly as it was — and it puts you back on Profile when you are done, because that is where you started. It is deliberately not beside Restart my garden. That one destroys a garden and this one cannot, and putting a harmless control beside a destructive one is how people learn to skim the destructive one.
-
Signing in now tells you your unsaved work is coming with you
The sign-in screen has always been able to say "4 changes will back up as soon as you are in" — the sentence was written and the app knew the number. It just never appeared, because the count feeding it was left at zero and nobody came back to it. So the one screen where you are most likely to wonder whether your work survives said nothing about your work. It says it now, with the real number. We found it while checking for a much smaller thing: whether the app ever writes "1 changes" instead of "1 change". It does not anymore either.
-
Ten squares instead of a paragraph
The hardening card now opens with a row of ten small boxes, one per day of the week. Filled means the plant went out and came back. Hollow means that day went by with nothing logged. Half-filled is right now. Faint and dotted are the days that have not arrived, with a line between those and the rest. That line is the part that took the thinking. A day you missed and a day that has not happened are both "nothing recorded" — draw them the same way and on day one the row reads as nine failures, which is the exact opposite of the truth. The count beside it works the same way: how many days you managed out of the days SO FAR, not out of ten. When the week ends those are the same number and it says "6 of 10 days" on its own, which is the only moment that sentence is true. A percentage of the hours sits next to it, and the last night counts toward it. That step records no hours — you leave the tray out, so there is nothing to log coming in — and an earlier version of this arithmetic scored a perfect week at 90%. The explanations have not gone. The reasoning for each step is under "why this step" and the day-by-day record is one tap away. What changed is what the card leads with: the thing you came to find out is now answerable at a glance, and the words are the second thing rather than the first.
-
You could ask what was wrong with a photo. You could not read the answer.
Take a photo on a plant's page and GardenTrack offers to have a look at it. Tap yes and it asks, gets an answer back, and saves it. Then the offer disappears and nothing else happens — because the only screen that shows one of those answers was not linked to from anywhere in the app. The reply was sitting on your own phone and there was no route to it. The plant's page now carries a line for it, from the moment you ask: it says we are looking while we are, and becomes the way in once the answer is back. That also matters for a reply that arrives while the app is closed, which was the case most likely to be lost entirely. It leads somewhere worth going. That screen is where you tell us which parts of an answer are right, and a reading you confirm gets written into the plant's own record rather than staying a paragraph you have to remember. Asking in the chat gives you words; confirming turns them into something the app knows. We found this while auditing pages with no way in. It had been filed as a screen kept for a capability nobody used, which is nearly the opposite of what it was.
-
The app could open to a blank screen. It cannot now.
Two things had to be true at once. You had a plant whose crop name this copy of the app did not recognize, and the app had a weather forecast saved. Then every screen came up empty — not a broken card, the whole thing, including your daily list. How a plant ends up with a name the app does not know is the part worth explaining, because it is not exotic. Our catalog has "head lettuce" and "leaf lettuce" but nothing called just "lettuce"; it has "cucumbers" rather than "cucumber". If we rename a crop, or if you use GardenTrack on two devices running different versions, one of them can be holding a name the other has never heard of. Local-first means your phone keeps working on its own, and it also means two phones can disagree for a while. The weather half is why it stayed hidden from us. The code that crashed only runs when there is a forecast to read, and a test device with no saved weather never reaches it. Everything we do on this app we do with real data on a real screen, and this is one we could look straight at and not see. Fixed at the root rather than at the one place it happened, so a future bit of code asking the same question cannot bring it back. And your garden was never at risk — nothing was lost, the app just could not draw it.
-
The app now asks where you garden
Until today, opening GardenTrack for the first time asked you nothing at all. It went straight to an empty list. That sounds like restraint and it was actually the bug: almost every date in the app is worked out from your frost dates, so with nothing to go on it fell back to a typical zone 6b and labeled its own answers a guess. The first run is now five screens with one thing on each, and it asks two real things — where you garden, and one thing that is growing. The second screen hands the answer straight back: your last frost, your first frost, and the number of growing days between them, worked out for a square about 800 feet across around your garden rather than for your region. A ZIP center can sit four miles from a garden, eight in the country, and in hilly ground four miles is weeks of difference. Every screen has a skip and none of them dead-ends. It never asks for an account — the whole app works on your device without one — and it does not ask about notifications, because asking permission before you have been given anything is a poor trade. It also does not ask your hardiness zone, for the honest reason that we cannot work it out from your coordinates and would rather not ask for something we cannot check. That one stays in Profile. The same pass fixed the orange "1 thing today" bar on the home screen. It looked like one big button and only the word Start was actually tappable — about 2% of it — so tapping the obvious place did nothing. The whole bar is the button now.
-
Your weekly job list is written by the app now, not by the AI
Tap "What should I do this week?" and the list underneath comes straight from the app — the same task list your Today screen uses. The AI writes one short sentence above it saying which job matters most and why, and nothing else. It used to write the whole list, and we measured what that cost: in five rolls out of fifteen it invented a job that did not exist and named none of the three real ones. Not a wrong word — made-up work, on the button a new gardener taps first. We had already handed it the correct list; we were asking it to read that list back, and one time in three it wrote something else. We tried tightening the instructions first and that is what was already in place when we measured. So the app draws the list and the AI does the part that genuinely needs judgment: which one to start with, in your garden, this week. A list the app renders from its own records cannot be wrong one time in three, because nothing about it is a roll of the dice. One honest limit: that drawn list currently lives on the phone that asked, so opening the same conversation on a second device shows the sentence without the list under it. The list is not lost from the phone that has it, and putting it on both is on our list.