How Amal Works

Amal is a cycle, not a single feature. A reading test on its own is just a test, and a recommendation on its own is just a guess. What makes Amal a decision system is that every stage feeds the next, and the teacher decides at every stage. This page walks the full cycle from setup to re-measurement, naming what really happens at each step and linking to the deep page for it.

1. Provision the school

Before any child reads, the school is set up: the organization, its schools, its classes, the teachers, and the students. A teacher reaches a class only through an explicit teaching assignment, because a teacher in these schools often teaches more than one class, and access has to follow the real assignments rather than a single owner field. Roles decide who can see and do what.

Deep page: organizations and classes and roles and access.

2. Diagnose with a short Arabic reading check

The child takes a short native-Arabic diagnostic. It is adaptive: after each answer, the platform updates an internal reading-ability estimate for the skill being tested and chooses the next task near the child’s current level, so the check stays short and stays at the right difficulty. The estimate is internal only. The child never sees a score, only tasks and supportive feedback. For the youngest children the check is audio-first.

Deep page: diagnostic and practice.

3. Decide a status per skill and per area

The platform turns raw evidence into an honest status for each reading skill, then aggregates skills into six reading areas: reading foundations, accuracy and fluency, reading comprehension, listening comprehension, language for comprehension, and writing and spelling. The decision runs through deterministic rules, so the same evidence always produces the same status. When the evidence is split or too thin, the platform stops and marks the skill “not enough data yet” rather than guessing, and everything downstream waits on that skill.

Deep pages: skills taxonomy and skill-status engine.

4. Compose a learning profile

The per-skill statuses are composed into one clear profile per child for the period: the child’s main need described in growth language, plus modifiers and finer flags that capture secondary patterns. The profile is the bridge from “here is the measurement” to “here is what to work on.”

Deep page: student profiles.

From the profile, the platform recommends one support plan from a fixed, curated catalog. A plan is not a single drill: it is built around one main skill, plus one or two linked skills, plus a strength the child already has that serves as a way into the harder work. The recommendation is deterministic, so the same profile always yields the same plan, with no language model in this path. The teacher reviews it and decides.

Deep pages: intervention plans and content and items.

6. Group and activate

Teaching thirty children one at a time does not scale. The platform groups children who can be taught together: those who share the same plan, the same main skill, the same level of support, and a compatible way of working. Children with incomplete data or a sharp recent drop are deliberately left out of groups, because they need individual attention. The teacher reviews the grouping and the level of support, then activates, for one child or for a group. Nothing activates on its own.

Deep page: grouping and activation.

7. Practice and serve tasks

Once a plan is active, the child practices. The platform selects stored reading tasks matched to the child’s current level of support and serves them online, on paper, or in a live class session. As the child answers, the platform scores each answer and updates the internal ability estimate, so practice keeps feeding the measurement.

Deep page: task delivery.

8. Monitor with repeated checks

Short checks, repeated over time, show whether the support is actually helping. Reading speed and accuracy are checked through teacher-scored fluency reads. Comprehension, vocabulary, and other areas use quick checks, rubrics, and teacher observations. Five global safety rules wrap every monitoring decision, including the rule that no status ever changes on a single data point and the rule that a sharp drop raises a review for a human, never an automatic failure.

Deep pages: progress monitoring and fluency checks.

9. Review and move the level of support when justified

A response layer reads the monitoring signal and decides whether to continue, adjust, or change the level of support a child receives. Before it can call a plan “not working,” it asks whether the plan was actually delivered: if delivery was below a set bar, the conclusion is about delivery, not about the child, and no escalation follows. This single check blocks the most common false move in a response-to-intervention system.

Deep pages: response decisions and evidence and review.

10. Report to teacher, parent, and school leader

The picture reaches people in the language each one needs. The teacher sees status per skill and per reading area. The parent receives a short summary in everyday, encouraging language, with no scores and no labels. The school leader sees reading health across classes and schools, with no public ranking of children. There is never a single overall percentage on any of these.

Deep pages: teacher reports, parent reports, and school and network reports.

11. Re-measure, and repeat

At the next review window, the diagnostic runs again and statuses update. The record becomes a film rather than a snapshot: you can see whether a child’s reading is moving, where, and how fast. The cycle repeats, and each turn is grounded in fresh measurement.

The teacher decides, throughout

Every stage above produces a proposal. The status, the profile, the plan, the grouping, the level of support, and the response decision are all suggestions the teacher can accept, change, or set aside. The platform proposes and explains why. The teacher decides. That balance holds through the entire cycle, and it is backstopped by a language-safety layer and by determinism that lets any past decision be reviewed exactly as it was made.

Deep pages: language safety and quality and determinism.