Math at Home

Nine units of Ontario Grade 3 mathematics, 629 pages, built for the adult at the kitchen table rather than for a teacher. Each unit is a self-contained offline app: eight to ten games, six interactive Learn pages, forty-odd printables with answer keys, talk cards, a challenge set, a progress store held on the device, a page written for grown-ups, a scripted first ten minutes — and a Check instrument with no score, one attempt, skill-tagged items and a report addressed to the adult. Unit 1, How Much Is 1000?, is free and open.

Home mathematics resources are designed for teachers and then handed to parents. The parent receives a pack of worksheets with no statement of what it is for, no way to judge whether the child is on track, and no guidance for the situation that most affects the outcome: the child is stuck and the adult does not know what to say next. Content supply is not the constraint — there is more free Grade 3 mathematics material available than any family can use. The constraint is that the supervising adult has no instructions. The product had to treat that adult as the primary user.

Parents and carers with no mathematics training and no lesson-planning time, and the children sitting beside them, usually tired, usually after school.

• Learning-experience design for a non-expert adult and a child, together
• Assessment design — the Check instrument and its adult-facing report
• Practice design across nine units — questions made fresh each round rather than drawn from a fixed list
• Specifying the shared game behaviour and the on-device progress record
• Print design, curriculum mapping and all instructional writing

What the design set out to achieve. These are objectives, not results.

• Let a child start without an adult present
• Let a child be wrong without it costing anything
• Give the adult a next action rather than a ranking
• Make practice non-repeating, so playing a game twice is practice and not recall
• Keep everything on the device — no accounts, no uploads, nothing about a child leaving the home

Each unit is a self-contained set of 67 to 72 offline pages, and all nine units share the same building blocks: the games, the practice questions, the progress record, the interactive demonstrations and the Check. There is no server and no account. Progress is kept on the device itself, which is a privacy decision as much as a practical one — there is no account to create, no address to collect and nothing to leak, because nothing is ever transmitted.

Each of these follows from a published position rather than from taste. The reasoning is set out in The Learning Conditions: a child must be able to start alone, be wrong at no cost, and have an adult nearby who knows what to say next.

Nothing in these units calls a model, and no child’s work is transmitted anywhere — the whole app runs with the network unplugged. In the build, the most valuable use of AI was adversarial rather than generative: building the practice questions and then running them at volume to find which questions they could produce and, more importantly, which they could not. AI also drafted parent-facing explanations, which were then cut hard, because generated prose over-explains to a reader who is standing up in a kitchen.

The Check is the clearest example in this portfolio of a decision that cannot be delegated. Removing the score, allowing one attempt and addressing the report to the adult are each defensible on formative-assessment grounds, and each of them is commercially the wrong instinct — stars, streaks and a percentage are what sells a learning product. Holding that line is judgement. So is the three-minute target, which came from watching where a child’s attention actually goes rather than from what fits neatly into a lesson.

Questions made fresh each round can be technically valid and pedagogically wrong, and they arrive that way silently. A subtraction that never happens to require regrouping on a sheet whose whole purpose is regrouping. A comparison question where both numbers come out the same. Measurement in decimal centimetres on a page that promises whole numbers — that one reached the child in 88% of rounds on a page that said it would not, and it was found by machine rather than by reading. That class of defect is what forced the QA suite into existence, and it is why the questions are now checked against the finished pages a child actually sees.

The iteration on this series is recorded in the checks that now run on it rather than in a changelog. Each defect found by running the questions at volume became a rule written back into the design and a check added against it, so the same failure could not return quietly in a later unit. The units built later inherit the rules that the earlier ones paid for.

The Check is the design decision I would defend hardest and the one with the least evidence behind it. It is reasoned from formative-assessment principles and from what a parent can act on; it is not measured, and I have no data on whether families use it or what happens when they do. Unit 1 being free at least means the reasoning can be checked against the artefact instead of taken on trust. If I were to add measurement, it would have to be measurement that does not turn the Check back into a score, which is a harder design problem than it sounds.

• Formative assessment design
• Learning-experience design for non-expert adults
• Non-repeating practice design
• No-account, on-device data design
• Print and worksheet design
• Ontario mathematics curriculum mapping
• Accessibility