Find a book for this child

Not for an age bracket. For this child.

Age brackets like 1–3 or 3–5 exist because a person cannot hold more than a few buckets in mind at once. That was our limitation, never a fact about children, who develop at wildly different rates within any bracket. A machine has no such limitation. In this standard every skill carries its own age range and its own difficulty rather than a bracket membership, which means a child can be described by what they actually do.

So the questions below ask what the child does rather than how old they are. Answer what you can and skip the rest. Two children of the same age who answer differently will be shown different books, which is the entire point.

A word of caution before you start. Nothing here assesses a child. The questions exist to point at the right shelf, and the answer is about books. Children develop at very different rates and nearly all of that variation is ordinary; a child who does none of the later things on this page is not behind. If you have a real worry about your child, a pediatrician is the right address, not a book finder.

Why every position carries its reason

Every line above can be traced to a rule at /api/matching. Disagree with a result and you can point at the rule that produced it. A recommendation you can argue with is worth more than one you have to trust.

This version is a step, not a destination

Asking about the child instead of the calendar is an improvement over a bracket, and it is still a rough instrument. Four questions and a fixed rule set cannot describe a child, and this version does not pretend to. It is a transitional stage: better than a bracket, far short of what the problem deserves.

The destination is different. A system able to reason properly about a child — and such systems are arriving — will not need a form with four questions. What it will need is something to reason from: a body of book properties that were measured rather than asserted, published openly, and not written by whoever profits from the sale. That layer does not currently exist, which is the actual gap. A model asked which book suits a child can read a sales rank, a star average and a printed age label, and nothing that describes the book. However well it reasons, it cannot reason its way to data that was never recorded.

This standard exists to make the third option available, and it is offered to be taken apart and rebuilt. It is published under Creative Commons Attribution 4.0 and deposited with a permanent DOI precisely so that someone can supersede it. A better standard replacing this one would be the outcome we want, not a defeat. What matters is that a standard exists at all.

The same service for machines

A recommendation system can call /api/recommend with an age band and a list of skills and receive this identical result as JSON, explanations included. It is the same code path: no separate version of this site exists for crawlers, and nothing is shown to a machine that is not shown to a reader.