The RTI Loop

Amal is a loop, not a feature. The product wedge is the integrated cycle (measure, decide, support, monitor, re-measure), not any single screen in it (Vision rule V-1). A diagnostic on its own is a test, and a recommendation on its own is a guess. Amal is a decision system because every hop feeds the next and the teacher decides at every step.

This page describes the cycle from start to finish and names the subsystem responsible for each hop, so you know which module owns what when you integrate.

The cycle, hop by hop

1. Provision. Organizations, schools, classes, and users are created first. Identity comes from the auth module, tenancy from the org/school/class tree, and people and rosters from the users module. A teacher reaches a class only through the ClassTeacher junction, never a teacherId on the class.
2. Diagnose and score θ. A student takes a short adaptive diagnostic. The diagnostic session module serves items, and the engine scores each answer and chains the ability estimate θ (theta). θ is internal only and is never shown to a student or family.
3. Decide skill status. The skill-status engine turns raw evidence into an honest per-skill and per-domain status through five deterministic layers, and rolls up to six macro-domain tiles. When evidence is split or thin it emits do_not_decide_yet and everything downstream waits.
4. Resolve a student profile. Statuses compose into exactly one primary educational profile per student per window (plus modifiers and sub-flags) that describes the child’s main need in growth language.
5. Recommend an intervention bundle. A deterministic profile→bundle map picks one of 19 named bundles, built as an anchor skill plus one or two supporting threads plus an asset bridge.
6. Smart-cluster. Students who share a bundle, anchor, scaffold tier, and delivery mode are grouped so they can be taught together. Students with insufficient data or an acute-regression alert are excluded.
7. Teacher activation. The teacher reviews the recommendation and the grouping, sets scaffolding, and activates, alone or for a cluster. Nothing activates without the teacher (V-12).
8. Task delivery and practice. Stored bank items matched to the student’s scaffold tier are selected and served (online, on paper, or in live sync mode). The server grades answers and re-chains θ.
9. Progress monitoring. Short repeated probes track whether the support is actually working. Fluency and accuracy use CBM/ORF, and everything else uses quick checks, rubrics, and observations. Five global safety rules wrap every decision.
10. RTI decision and tiers. The RTI layer reads the monitoring signal and the fidelity gate to decide: continue, adjust, or move tier. It never escalates on a plan that was simply not delivered.
11. Report. Per-skill and per-macro-domain status flows to teacher reports and a growth-language parent PDF. There is never a single overall percentage.
12. Re-measure. The next window re-runs the diagnostic and statuses update, so the record becomes a film, not a snapshot. The loop repeats.

Hop → subsystem → guide

Where to go next

• Measurement & Determinism: θ, δ, Rasch, and why the same input always replays to the same output.
• The Decision Engine: the five transformation layers and the statuses they produce.
• Intervention Design: bundles, the Threaded Path Model, scaffold tiers, and clustering.
• Safety Rules: the five GSR safety rules, the fidelity gate, and the teacher-decides guarantee.
• Glossary: every term and ID family in one place.