Algorithm Transparency - Daily Maths Review

Short Version

Teachers stay in control

Suggested reviews are draft plans. Teachers can save, change, ignore or launch them, and the system stores the reason and score when a suggested review is saved or launched.

Parents can see the logic

Home recommendations are based on the child's maths progress, placement results when available, recent accuracy and topics that need another look.

No hidden AI ranking: recommendations do not use advertising data, payment data, unrelated browsing behaviour, or hidden machine-learning profiles.

School assessment reports also use a published, versioned evidence-band policy. It describes the selected classroom sample and never creates an automatic report-card grade or mastery claim.

What Data Is Used

Student work

Activity completed
Settings used, such as denominator 8
Questions attempted and correct answers
Recent scored attempts
Correct streak for auto-marked answers

Teacher setup

Assigned work
Class and student progress summaries
Recent suggested-review history
Saved or launched suggestion reasons

Placement

Area checked, such as number or operations
Estimated level
Next step
Recommended activity for that step

Progress is tracked at activity level and, where the activity supports it, activity-plus-settings level. For example, fractions practice can be summarised as fractionsAndDecimals/barFraction/denominator=8, so a broad topic does not hide a specific setting that needs work.

How Mastery Is Worked Out

Mastery is current, not lifetime-only. The system keeps lifetime totals for history, but current mastery gives the strongest signal to the most recent scored questions.

Recent window

The main mastery calculation uses the last 30 scored questions for that activity when they exist. Older questions remain in the history but do not keep dragging the mastery label down forever.

Enough evidence

A student needs at least 10 recent scored questions before an 85% or higher recent accuracy can count as percentage-based mastered.

Mastery level	How it is interpreted
Support required	Current accuracy is less than 50%.
Partially mastered	Current accuracy is 50% to less than 85%, or 85%+ with too few recent scored questions.
Mastered	Current accuracy is 85%+ with enough evidence, or the student has 12 correct auto-scored answers in a row.

Example: if a student gets 97 wrong and then the last 3 right, they are not marked mastered. If the last 30 scored questions are 27 wrong and 3 right, the current accuracy is 10%, so it is still support required. If the student then gets 12 auto-scored answers right in a row, the streak shortcut can count that activity as mastered.

How Curriculum Outcome Evidence Is Interpreted

School assessment reports bring together trusted evidence for each curriculum outcome. They preserve whether evidence was server-validated or teacher-finalised, its source, assessment purpose and version, curriculum edition and mapping version, date, question and session counts, attempt context, support or accommodations, teacher judgement and known limitations.

Evidence from changed curriculum editions stays separate. Work from the latest 42 days is marked current for planning; older work remains visible as historical evidence.

Evidence band	Versioned classroom-evidence rule
Insufficient evidence	Fewer than 4 distinct scored questions; fewer than 3 students for a class interpretation; or important conflict, partial mapping, or missing version context.
Support required in this sample	Enough valid evidence and accuracy below 50%.
Developing in this sample	Enough valid evidence and accuracy from 50% to below 80%.
Secure in this sample	Enough valid evidence and accuracy of at least 80%.

Classroom evidence, not certainty: these labels describe the selected sample only. They do not diagnose a student, prove mastery, create an automatic report-card grade, or replace teacher professional judgement. Materially conflicting sources are shown as a limitation instead of being hidden.

Teacher Suggested Reviews

Teacher review suggestions use only dated evidence from the latest 42 days. Older and undated work stays in progress history, but it does not decide today's review draft.

score = current support + developing need + due refresh + accuracy gap + evidence + class reach + precision - recent repeat penalty

Factor	What it means
Current support	Adds 42 points for each student whose current evidence shows support is required.
Developing need	Adds 24 points for each student whose current evidence shows the target is developing.
Due refresh	Adds 12 points for each student whose saved review date, or transparent fallback interval, is due. Mastered work is excluded unless it is due for a light refresh.
Accuracy gap	Lower current accuracy increases the score by up to 34 points.
Evidence	More current scored questions give the system more confidence that the signal is real, up to 14 points.
Class reach	A target affecting more students in the class receives more weight, up to 20 points.
Precision	Adds 8 points for an exact DMR target or 5 points for an exact activity setting.
Recent repeat penalty	An activity suggested recently loses 32 points so the same item does not keep coming back too often.

The planner prefers exact DMR target evidence, then activity-plus-settings evidence, then activity-level evidence. It ranks the highest current needs directly and chooses up to five activities, with at most one exact target or setting from each activity.

The teacher-facing panel explains the current evidence window, students affected, exact focus, score factors and plain-language reason. Where saved data includes exact activity options, those settings are preserved in the editable Review Builder draft.

Eligible Live Review responses contribute at a lower 0.5 formative source weight. A malformed, stale or too-small Live Review sample is ignored for automatic planning, and Live Review never creates a mastery or formal assessment claim.

Live Review Follow-up Suggestions

A completed Live Review report may prepare an editable Review Builder draft from the exact recent activity settings. It uses the same latest 42-day window as current-needs planning and explains the participation, response sample, accuracy, unanswered work, recorded support, confidence, source weight and latest evidence behind each activity.

Tracked class evidence

Only current-roster, server-validated tracked responses can contribute as named formative evidence. They use the 0.5 source weight and require at least 5 recent responses from 5 students, plus either 2 questions or 8 participants.

Quick join stays aggregate

Quick join can provide only a whole-room anonymous signal. It uses a more cautious 0.25 signal weight, requires at least 8 aggregate responses and never becomes named student progress.

Teacher judgement stays final

Small, incomplete, old, excluded or conflicting samples do not create an automatic draft. Every generated plan can be edited or ignored and is not a mastery result, formal assessment or report-card grade.

Assigned Worksheet Evidence In Current Needs

Tracked Worksheets are labelled assigned practice. For each exact activity setting, the planner uses work from the latest 42 days and needs at least 5 server-validated first responses across at least 2 students before it can suggest follow-up.

First response drives the signal

The first scored response contributes to the current accuracy denominator. A later correct response, retry or hint records useful eventual-success and support context, but it does not become another first-response question.

Practice never becomes mastery by itself

A Worksheet suggestion is an editable teaching starting point, not an assessment grade or curriculum mastery claim. Small samples stay visible as insufficient and materially conflicting sources suppress automatic suggestions.

Mappings must be exact

Curriculum codes are used only when the immutable Worksheet item has a reviewed item-level mapping. Broad activity-family mappings are shown as broad and do not create an outcome claim.

The planner still prefers an exact DMR target over activity settings and activity-level evidence. Formal School assessment evidence remains in its separate reporting and teaching-action workflow. Supported Worksheet needs may prepare an editable Worksheet, Activity Assigner draft or Review Builder plan; these handoffs carry safe activity settings and evidence context, not student responses.

Student And Home Practice Suggestions

Student recommendations use the same current mastery thinking, then add two extra signals: placement and new-topic exploration.

Placement boost

If a placement result points to a starting point or next step, that activity receives a boost and the reason says that placement suggested it.

New-topic exploration

Activities with no saved practice can still appear with an exploration score of 28, so a student is not locked into only topics they have already tried.

Student signal	How it affects the recommendation
Progress score	Uses the same current-mastery score as the teacher tool, but can include mastered activities as low-priority spaced review. The DMR target-level model is designed to start with close review intervals, such as 1 day, 2 days and 4 days, then grow the interval after repeated secure evidence.
Placement result	Placement adds 72 points for support required, 46 for partially mastered and 18 for mastered next-step review. Combined placement boosts are capped at 140.
New topic	No saved practice gives the activity an exploration score of 28.

Signed-in students can receive progress-based suggestions. Guest practice uses selected year level, curriculum scope and light shuffling because there is no saved student progress to read.

Limits And Human Control

Not a diagnosis

Mastery labels are planning signals. They should be read with classroom observation, student confidence and the type of question attempted.

Self-check is weaker evidence

Drawing, visual or self-check questions can be recorded, but the 12-in-a-row shortcut only uses answers the system can auto-score confidently.

Adults can override

Teachers and parents can choose other work. Recommendations are there to explain a likely next step, not to force one.

Audit trail

When a teacher saves or launches a suggested review, Daily Maths Review stores the recommendation score and reason so the choice can be reviewed later.