Fluency Checks

Reading fluency is one of the clearest early signals of whether a child’s support is working. A child who is stuck on decoding will be slow and inaccurate; a child gaining ground will read more smoothly and make fewer errors across comparable sessions. Amal tracks this trajectory over time using a structured system of teacher-scored readings with no speech recognition or audio analysis of any kind.

What is measured

Two quantities matter in a fluency check.

Words correct per minute counts the number of words a child reads correctly in a timed 60-second oral reading. It reflects both speed and accuracy together: a child who reads fast but makes many errors ends up with a lower count than a child who reads more carefully.

Accuracy measures the proportion of words read correctly out of the total attempted, expressed as a percentage. A child might maintain adequate accuracy while still reading slowly, or might read quickly but with a high error rate. The two numbers together give a more complete picture than either alone.

Beyond these two quantities, the platform also tracks which Arabic linguistic features are producing errors, organized into a feature catalog built with Arabic in mind. This allows the teacher to see whether, for example, a child is consistently making errors on a particular class of words, separate from the general fluency trend.

How a session works

The teacher opens a fluency session for a specific child and passage, then listens to the child read aloud for 60 seconds. During the read, the teacher taps to mark each error as it occurs. When the time is up, the session closes and the platform computes words correct per minute and accuracy from the teacher’s taps and the word count.

No recording is made that is analyzed by any automated system. No speech recognition runs. The audio is kept as an archive record only. The teacher’s own judgment, expressed through the error taps, is the sole input. This design holds for the current phase and is not subject to change as a function of available technology: it is a deliberate choice that keeps human judgment at the center of the assessment.

The aim line

A single fluency session tells you where a child is at one moment. What a teacher needs to know is whether the child is on a trajectory toward their goal. The aim line makes that trajectory visible.

Before the aim line activates, the platform collects three baseline readings. These three sessions establish the starting point. The baseline is the median of those three readings, which reduces the influence of a particularly good or bad day on the starting estimate. Until three readings exist, no aim line is drawn and no trend signal is produced. One or two readings, however clear they seem, cannot alone define a trend.

Once the baseline is established, the aim line describes the expected growth path from the baseline toward the child’s fluency goal by a target date. Each subsequent reading is plotted against this line.

The four-point rule and trend signals

After four or more comparable readings have been recorded, the platform applies the four-point rule to determine whether the child’s trajectory is running above or below the aim line. Four points are required, not three or two, because a smaller set is too easily distorted by a single session.

A comparable reading is one taken under conditions consistent with the other readings in the window. If a reading was taken under different conditions, it is retained in the record but excluded from the trend calculation and flagged as such.

The four-point rule produces one of four signals:

A “below the aim line” signal is not an automatic problem and is not an automatic prompt to change the plan. It is information for the teacher, combined with other evidence, feeding into the broader picture of whether support is working. The fidelity layer (whether the plan was actually delivered at the required rate) and the tier-review process both stand between a trend signal and any change in a child’s support level. See Support Tiers and RTI Decisions for how the trend signal feeds those layers.

This handoff is now live. A “below the aim line” trend is read by the support-tier layer as a “not progressing” signal and can raise a student-level alert for the teacher. A “not enough points yet” trend is treated as a reason to wait, not a reason to act: it holds rather than raising a false alarm. An “above the aim line” trend, and a “mixed” trend, raise nothing; the mixed case is handled conservatively on purpose, because a mixed picture is ambiguous rather than a clear sign of trouble. In every case the trend reaches the teacher through an alert, never as a direct change to a child’s support level.

What happens when grade norms are missing

Fluency levels that are considered on-track or below-track differ by grade. For grades where the partner norms have not yet been finalized, the platform continues to score and track readings normally. Words correct per minute and accuracy are computed and plotted. What is not produced is a risk judgment against a missing benchmark. The status for that child on the norm dimension is shown as “not assessed” rather than inventing a cut score.

A missing norm is never treated as zero, and it never makes a child appear to be at risk when the data to support that conclusion does not exist. The fail-safe direction is always to withhold a judgment rather than to fabricate one.

The relationship between fluency checks and broader monitoring

Fluency checks and the broader progress monitoring system are separate processes and accept separate evidence. A fluency reading cannot be submitted through the general monitoring channel, and non-fluency evidence (comprehension checks, vocabulary probes, rubric observations) cannot be submitted through the fluency channel. This separation exists because the two processes use different analytical rules, different comparability tests, and different success criteria.

The fluency layer emits a raw trend signal: it reports what it sees in the data without applying any further gate. The tier-review layer applies the fidelity test and composes a teacher-facing message from the signal. The distinction matters because a “below the aim line” reading combined with a low delivery rate (the teacher rarely ran the sessions as planned) means something very different from the same reading combined with a high delivery rate (the teacher ran every session and the child still did not gain ground). The fluency layer does not make that distinction itself; the tier-review layer does, using the delivery rate from the fidelity tracker.

Arabic feature tracking

Each passage used in a fluency session is tagged with the Arabic linguistic features it contains: characteristics of letters, vowel patterns, word structures, and similar properties relevant to Arabic reading development. When a teacher marks an error, the platform records which feature categories the errored words belong to.

Over repeated sessions, this produces a picture of where errors are concentrating. A child who consistently errors on one category of features may need attention on that specific dimension that the general fluency trend does not capture. Feature tagging is maintained by the content team in coordination with the partner; the feature catalog reflects the linguistic scope of the current phase, and some features are not yet active because they fall outside the current instructional scope.

What the teacher sees

The teacher sees the trend response for a child: the time series of readings, the aim line, the current trend signal, and the feature breakdown. Alerts that require the teacher’s attention are surfaced separately; a raw trend signal is information, not an action item. When a signal does require action, the teacher receives it through the alert layer with a composed message that includes the fidelity context. The raw signal itself is not shown to children or parents at any point.