The weekly summary says 92%. Last week it said 74%. Nothing about your child's reading changed in the seven days in between. What changed is that this week's sessions were mostly words she first met a fortnight ago, and last week's were nearly all new. Same child, same effort, eighteen percentage points apart, and the number by itself tells you none of that.
A dashboard compresses a few hundred small events into half a dozen figures, and every figure answers a question you cannot see on the screen. Some reward close attention. Others exist because they were easy to measure, and reading those as if they mean something is how families end up worrying about the wrong thing.
Accuracy means nothing until you know what was on the screen
The split that matters more than any other is new material versus review. Two weeks both showing 90% can mean opposite things.
- 90% or above on review: expected, and not news. Review is supposed to be easy. The signal here only appears when review accuracy falls below roughly 80% — that means earlier material never consolidated and is quietly leaking.
- 90% or above on new material: the new material is not new enough. Difficulty is lagging behind the child, usually because a level-up rule is too conservative. Worth nudging up manually if the app allows it.
- Roughly 60% to 75% on new material: normally where learning is actually happening. It looks worse on a chart and feels worse to read. It is fine.
- Near 50% on two-option items, or near 25% on four-option items: that is chance. Nothing is being learned. The child is clicking.
If your dashboard gives one blended accuracy number with no breakdown, treat it as an attendance figure rather than a learning figure. Where a breakdown does exist — by topic, or by first attempt versus retry — that is where the information lives. First-attempt accuracy is far more informative than final accuracy, because on any system that lets a child retry until correct, final accuracy converges on 100% for everybody and stops distinguishing anyone from anyone.
Time on task is the weakest number on the page
It is the easiest thing to log and the least trustworthy to read. Eighteen recorded minutes might contain a trip to the bathroom with the tab left open, four minutes of animated reward sequences, three items answered by an older sibling, and a parent reading questions aloud. It might equally be eighteen minutes of real thought. The number cannot tell you which, and it will never admit it.
Better numbers, if the dashboard exposes them: items attempted, median seconds per item split by new versus review, and the spread of those times. Some rough guides that hold up in practice. On new material, a steady two to four seconds per item means clicking rather than thinking, especially when accuracy is sitting near chance. Thirty to sixty seconds could be careful work or complete paralysis, and the accuracy on those slow items is what tells you which. The most encouraging pattern is an uneven one: quick on the familiar, visibly slower on the new. Uniform pace across easy and hard items usually means the child has stopped distinguishing between them.
One more thing about time: more is not better. Accuracy drifting down while time on task drifts up is one of the few genuinely bad shapes on a dashboard. It almost always means grinding — the child is putting in more and getting less, which is a reason to shorten sessions rather than praise the hours.
Retention after a gap is the number worth trusting
Performance at the end of a practice session is inflated by definition. Everything is warm, the format is familiar, the last four questions primed the fifth. The real question is whether any of it survives a gap.
A workable standard for home use: if a skill survives seven days untouched, it is probably learned. If it survives thirty, it is reasonably durable. If it collapses after two days, the session produced familiarity rather than learning, and repeating the same session will mostly produce more familiarity.
If your dashboard reports anything shaped like "returned to this topic after N days, X% correct", read that panel before anything else on the page. It is the most honest thing there. Many dashboards do not report it at all, in which case you can run the check yourself in about ninety seconds: on a Saturday, pick three things from three or four weeks ago and ask them cold, in your own words, away from the app. The result is often sobering and always more useful than the weekly percentage.
This is also why long streaks flatter. Daily practice with immediate re-testing produces excellent session numbers and mediocre memory. Spacing — practise a topic, leave it, return a week later — produces worse-looking sessions and better retention. A dip as a topic comes back after a break is the system working, not failing.
What the chart shapes mean
| Shape | Usually means | Reasonable response |
|---|---|---|
| Steady climb, then flat near 100% | Either genuine mastery, or difficulty stopped rising | Check what the last fortnight actually contained. If it is the same level, raise it. |
| Sawtooth — dips when a topic starts, recovers within it | Healthy. New material is meant to hurt briefly. | Nothing. This is the shape you want. |
| Flat around chance level | Guessing | Stop and drop back. Either the level is too high or the instructions are not landing. |
| Accuracy sliding down while minutes climb | Fatigue, or growing aversion to the subject | Halve the session length. Change the time of day before changing the content. |
| Very high accuracy, very short sessions | The child is picking the comfortable topics | Choose the topic yourself for a week. |
| Sudden jump with no change in difficulty | Usually a change in who is sitting beside them | Not a problem, but read later weeks knowing it. |
| Long gap, then a return well below the old level | Ordinary forgetting | Expect two or three sessions to recover, and it often finishes above where it was. |
Three signals that beat any chart
Ask which one was hardest. A child who answers specifically — "the ones with the silent e", "when the number had a zero in the middle" — is monitoring their own learning. "All of it", "none of it" or a shrug usually means the session was undemanding, or they were not really present in it.
Ask them to explain it to you, badly. Explanation exposes memorisation quickly. A child who can only reproduce the exact wording the app used has stored a sentence rather than an idea. A child who explains it wrong but in their own words is closer to understanding than the first one.
Change the surface. Same skill, different clothes. If the app asked seven times six with counters on a grid, you ask about six bags with seven apples each on the way to the shops. That is transfer, and transfer is both the thing you actually want and the thing no dashboard can see.
What dashboards cannot measure, and how to talk about them
The absent columns are worth listing: whether your child enjoyed it, whether they would choose it unprompted, comprehension beyond what can be auto-marked, handwriting, spoken fluency, and the quality of a wrong answer. That last one costs more than people realise. A nearly-right answer reveals a child almost there; a lucky guess reveals nothing; the log records both identically.
It is also worth holding the whole thing at the right altitude. A dashboard is a record of activity inside one product. It is not a school report and it is not a diagnostic assessment. If you see something persistent and lopsided — decoding stuck for months while everything else moves, or a consistent collapse on anything involving following a sequence of instructions — that belongs in a conversation with your child's teacher or a qualified professional, not in a settings menu. A tool such as Kid Genius World can show you where practice went and where it stalled; it cannot tell you why the pattern is there.
Finally, how you say it out loud. Below about eight, do not present percentages as a score. Translate into content: "the 'ch' ones went well — 'th' is the one to come back to." From nine or so the numbers are usually fine, and some children enjoy them, but watch what the number starts doing. A child who avoids the harder topic to protect a percentage has learned something real, just not the intended thing.
If you only look at two things each week, make them first-attempt accuracy on new material and how a topic performs when it comes back after a gap. Those two carry most of the signal. Everything else on the page is context.