What Koda doesn't try to do.

Why a scope note.

Most product pages tell you everything a thing can do. We've found the opposite list is more useful — every “no” on it is a decision that lets us make the things we do say yes to actually good. If you read this and Koda turns out not to be the right tool for what you need, that is a clean outcome. A wrong fit is worse than a missing feature.

The frame: Koda watches a child do math on paper at the kitchen table, helps when help is wanted, stays quiet when it isn't, and gives a parent a weekly summary of what happened. Everything below is a thing we keep getting asked about and have decided, on purpose, not to ship.

It is not a classroom replacement.

Koda is not a substitute for school. We don't teach a sequenced curriculum from scratch, we don't track grade-level standards as a master list, and we don't certify mastery in a way that means anything to a teacher. What Koda does is sit next to homework and worksheets that already exist, give a child a patient second pair of eyes, and reinforce the habits — show your work, draw the picture, re-read the problem — that classroom teachers want kids to develop. The teacher does the teaching. Koda does the sitting-next-to.

Practically: if your child is being homeschooled and Koda is the only math input, Koda is the wrong tool. Pair it with an actual curriculum (Beast Academy, Singapore Math, Eureka, a workbook from the bookstore — any of these) and Koda becomes useful again. The worksheet is the spine. Koda is the spotter.

It is not a screener for learning differences.

Koda can be configured to support a child with dyscalculia, dysgraphia, or ADHD — different profiles get different defaults around hint timing, written-vs-spoken steps, focus-timer length, and break cadence. But those settings live on the parent profile because youtold us about them, not because Koda figured it out. We do not infer diagnoses from the data we see, we do not flag “possible dyscalculia” in any digest, and the persona never says anything like “your child shows signs of —”. Diagnostic claims about a child should come from clinicians who saw the child, not from a camera that watched a worksheet for 15 minutes.

What Koda willtell you, in the Friday digest, is concrete behavior: “this week your child spent more time on subtraction-with-regrouping problems than on anything else, and on three of them they re-tried after a slip — which is the pattern we look for.” That's a description, not a diagnosis. The next move is yours.

It observes to support you — not to surveil your kid.

The line we hold: observe to support, never to surveil. Today the cameras read the worksheet and recognize whose worksheet it is. On the roadmap, Koda will also notice how physical skills are developing — pencil grip and fine-motor control, whether posture holds, how focus and calm last across a session — and surface that to you, the parent, the way an attentive teacher mentions your kid faded after twenty minutes.

What it will never be is a number that ticks down on your child. A kid can fidget, slump, and stare at the ceiling for forty seconds without it costing them anything in the game, and none of it is shown to the child as a score or a judgment. We also don't do emotion-guessing — estimating frustration from a facial expression is unreliable in adults and worse in children, and we don't build on it. The developmental signal is parent-facing, runs on-device, is never sold or turned into a diagnosis, and gets a privacy and legal review before it ships. The frame-by-frame is in what the camera actually sees.

It is not a streak machine.

No streak counter, no daily-target badge, no “you broke your X-day streak” notification. The longer version is in why we don't use streaks; the short version is that streaks optimize for preservation, not practice, and we'd rather a kid take Saturday off and come back Sunday than do 90 seconds of trivial work to keep a number alive.

It is not a social product.

There is no leaderboard. There is no “friends” list. Kids can't see each other's XP or compete on totals. Two siblings using the same Koda have separate profiles that don't look at each other. We don't ship cross-household comparisons (“your child is in the 73rd percentile of 4th graders on Koda”) and we don't plan to. The math is between the kid, the paper, and the people in the room.

The closest thing to a social feature is the persona — the AI voice the child picked. It's warm, names the child, reacts to the work. But it is not pretending to be a friend, and it does not claim memories of things it didn't see.

It is not a voice clone of someone the child knows.

We've been asked whether parents could record themselves so the persona uses their voice. The answer is no, and not because the technology is hard — it's gotten easy, which is part of the problem. A tutor that sounds like a parent is in a position to deliver instructions, judgments, and small disappointments using a voice the child has learned to trust at a different level than a stranger's. That's a kind of weight we don't want a software product carrying. The persona voices Koda ships with are clearly synthetic and clearly not anyone the child knows. The line is deliberate.

It is not a chatbot.

Koda does not let a child open a chat window and ask anything. There is no free-text input, no conversation history the kid can browse, no “ask Koda about dinosaurs” surface. The persona talks about the math problem in front of it, and largely about that problem only. Two reasons. Safety: a chatbot for an 8-year-old has to handle every possible question an 8-year-old might ask, and the failure modes of generic chat have been documented enough times in the last few years that we don't feel like contributing another instance. Pedagogy: the kid we most want to help is the one who tends to drift off the worksheet. Giving them a chatbot would be handing them the drift on a silver tray.

It is not free of failure modes.

Some things we'd like to do, we can't reliably do yet. Handwriting recognition on the messy end of the distribution is imperfect — a child whose 4 looks like a 9 will sometimes get flagged as wrong, and the fix is more often a better camera angle than a better model. Hand-written word problems are harder for us than typed ones. The face-recognition handoff between siblings close in age sometimes asks for a confirm-tap because we'd rather ask than guess wrong. None of these are scope decisions — they're engineering work we're actively doing. We mention them because honest scope includes the gap between “what we've decided not to do” and “what we haven't finished doing yet.”

It is not the only math input your child should have.

We don't want a child whose only math contact is a piece of software. The part of math learning we cannot replicate — and don't try to — is the back-and-forth with another human who already knows the kid. The teacher who notices the kid was quiet today and asks why. The parent who, in the car, says “what do you think nine eights is” for no reason. The cousin who shows them a card trick that turns out to be multiplication. Koda is not those people. It's the patient pair of eyes at the kitchen table when nobody else has the bandwidth for the third worksheet of the night — a real role, and a useful one. It is not the whole job.

What this list isn't.

It isn't a list of things we've ruled out forever. It's a list of things we've ruled out for v1, and that we want parents to know about before they decide whether Koda belongs in their house. Some of these will stay ruled out — the voice-cloning line, for instance, isn't one we expect to move. Others — the OCR on messy handwriting, the persona's sense of conversational timing — will improve as we ship. When that happens we'll say so, by name, in a release note. We'd rather you find out from us than from a feature you didn't notice quietly turning on.

If after reading this Koda still sounds like the right thing, the waitlist is here. If it doesn't — that's also a result we're happy with. The related notes on why we don't use streaks, why we reward effort, and what the camera sees cover the rest of what stays in scope.