Diagnostic & Practice

Amal measures each child’s Arabic reading skills through two connected activities: a short adaptive diagnostic and a practice queue that follows it. The diagnostic produces an evidence-based picture of where the child stands on each skill. The practice queue gives the child targeted, level-matched work right away. Both run entirely within the platform and adapt in real time to what the child shows.

The adaptive diagnostic

What it is and what it is not

The diagnostic is not a fixed test where every child receives the same questions in the same order. It is an adaptive assessment: the platform selects each question based on the child’s current estimated ability level, so easier items are not wasted on a child who is clearly fluent, and very hard items are not piled on a child who is still building foundational skills.

A typical diagnostic session covers a focused set of skills relevant to the child’s grade and the current assessment window (beginning of year, middle of year, or end of year). The number of items per skill is enough to produce a reliable estimate but not so many that the session becomes exhausting.

The 15-minute active time cap

A diagnostic session runs for at most 15 minutes of active time. Active time means time the child is actually working on questions. Pauses and audio replays are subtracted from the clock, so a child who needs to re-listen to an audio question is not penalized. When the 15-minute active limit is reached, the session closes automatically and the ability estimate from the items answered up to that point is used.

How the platform picks the next question

After each answer, the platform updates its estimate of the child’s ability and selects the next item using the following principle: choose the question that will give the most new information about the child’s level right now. In practice, this means selecting items whose difficulty is close to the child’s current estimated ability, within a window of roughly half a logit in either direction. Questions that are far too easy or far too hard provide little new information and are skipped in favor of better-matched ones.

Items the child has already answered in the same sitting are never repeated, regardless of how they were answered.

How the ability estimate updates

Each answer shifts the ability estimate up or down. A correct answer raises it; an incorrect answer lowers it. The size of the shift depends on how well the item matched the child’s level. If the item was close to the child’s estimated ability, the answer carries more information and the update is larger. If the item was very far above or below the child’s level, the update is deliberately smaller, because an unlucky error on an unusually hard item should not dramatically change the picture, and a correct answer on a very easy item adds little.

This dampening means the estimate tracks the child’s actual ability rather than bouncing with every individual answer.

Who determines whether an answer is correct

The answer key for every item lives on the platform’s server, not on the child’s device. When a child submits an answer, the server compares the selected option against the stored key and makes the correctness determination there. The child’s device transmits only the selected option. A child or anyone with access to the device cannot send a fake “correct” signal to inflate a score. This is a core integrity guarantee.

Audio-first for Grade 1

Children in Grade 1 receive audio-delivered questions. The platform determines grade from the child’s own profile record, not from a class-level label, to ensure each child is assessed correctly even in mixed or multi-grade settings.

If no audio item is available for a particular skill, the session reports that condition rather than silently substituting a text item. A substitution would change what is being measured without signaling that the measurement type changed. Transparency about the item type is part of measurement integrity.

Fluency is teacher-scored, not platform-scored

Oral reading fluency, which captures how many words a child reads accurately and smoothly in one minute from a physical passage, is scored by the teacher, not by an automated system. No voice recognition is used. The teacher listens to the child read aloud, marks errors on a printed passage, and enters the result into the platform. This is intentional: teacher observation captures dimensions of reading behavior that an automated counter cannot reliably detect in Arabic, and it keeps the teacher at the center of the assessment process.

Fluency scores enter the platform through the teacher’s direct input and then feed into the same interpretation pipeline as all other evidence.

Resuming after an interruption

School environments are unpredictable. Devices lose connection, browser tabs get closed, and device locks engage mid-session. The platform is designed so that none of these events causes a child to lose their session.

When a diagnostic starts, the platform stores a signed token in the browser’s local storage. If the session is interrupted for any reason, the child can return to the platform within 24 hours and the session will restore exactly where it left off, at the same question, with the same accumulated ability estimate.

The resume is not based on the token alone. The server re-checks that the token matches the session record and that the session belongs to that specific child before restoring it. A token from one child’s browser cannot be used to resume a different child’s session.

Practice after the diagnostic

From diagnostic results to a practice queue

Once a diagnostic is complete, the platform builds a practice queue from the skills that showed weakness or warrant monitoring. The queue is seeded from the same ability estimate the diagnostic produced, so the child begins practice at a level that matches where they ended the diagnostic, not at a generic starting point.

In the current phase, each queue focuses on one skill at a time, starting with the weakest-eligible skill. When that skill’s queue is exhausted, the child can move to the next. This focused approach avoids spreading practice too thin across many skills at once.

Adaptive item selection in practice

Practice uses the same adaptive item-selection logic as the diagnostic: the platform picks the item that gives the most information about the child’s current ability within a difficulty window near their estimated level. Items already answered in the same queue are not repeated.

Block-by-block difficulty routing

Practice is organized into blocks of 8 to 10 items. When a block is complete, the platform evaluates how the child did across those items and adjusts the difficulty level for the next block:

There are five difficulty levels: very easy, easy, medium, hard, and very hard. A child who consistently gets more than 80% of items right in a block will gradually move toward harder items. A child who is struggling will work at easier items until accuracy improves.

If a child is already at the very easiest difficulty level and still scores below 60% in a block, the platform recognizes that a prerequisite skill may need attention first and routes accordingly.

Difficulty level persists across sessions

The difficulty band a child reaches in practice carries over between sessions. A child who returns the next day, or after a weekend, begins practice from the difficulty level they were working at when they last stopped. They do not restart from the beginning. This means sustained effort compounds: every block a child completes moves them forward, and progress is not erased between logins.

What the child never sees

The child sees questions, feedback on individual answers, and progress indicators. The ability estimate, the difficulty band label, the skill status, and any profile or intervention information never appear in the child’s view. The platform deliberately separates the measurement layer from the child-facing experience.