How FluencyScope works

What FluencyScope is

FluencyScope is a screening tool that helps teachers measure oral reading fluency. A student reads a short passage aloud on their device; the app records the reading, transcribes it, and reports words correct per minute (WCPM), accuracy, and related measures.

Every result is advisory. FluencyScope does not diagnose reading disabilities, assign placements, or make decisions about students. It gives teachers information; teachers make the judgments.

How the fluency score is computed — without AI judgment

The core score is deterministic: the same recording always produces the same result, computed by fixed rules, not by an AI model's opinion.

1. The recording is transcribed to text by a speech-recognition service (Deepgram).
2. A rule-based algorithm aligns the transcript against the passage, word by word, and classifies each word as read correctly, substituted, omitted, self-corrected, or inserted. Self-corrections count as correct, following standard fluency-scoring practice.
3. WCPM and accuracy are simple arithmetic over those word-level results, and are compared against published, public-domain national norms (Hasbrouck–Tindal) to show an At / Below / Well-Below Benchmark band.

Where AI is used

A language model (Anthropic's Claude) is used for four clearly-marked, advisory features. It only ever receives structured numbers and text — never the student's audio:

• A 2–3 sentence written summary of the metrics for the teacher.
• An estimated 1–4 prosody rating (expression, phrasing, pace) based on timing patterns.
• Drafting comprehension questions for a passage.
• Suggesting a grade (correct / partial / incorrect) for a student's typed comprehension answers.

Everything the AI produces is labeled with an “AI” badge in the report and can be edited or replaced by the teacher. If the AI is unavailable, the app falls back to simpler rule-based output — it never blocks a report.

Teacher control

• Teachers can correct any word-level result — approve or reject a flagged error, or flag a missed one. Scores recompute from the teacher's corrections.
• Teachers can rewrite the AI summary, adjust prosody ratings, and re-grade comprehension answers.
• Corrections are kept as an audit trail alongside the original automated result, so it is always clear what was machine-scored and what a teacher decided.

Student data

• Students do not create accounts or passwords; they open a link from their teacher and read.
• Voice recordings are treated as potentially biometric data: access requires an authenticated teacher of the student's own school, and each school's data is isolated from every other school's.
• No student data is used to train any AI model.
• Teachers can delete a student's session, which removes the recording and its results.

Known limits

Speech recognition is imperfect. Background noise, quiet reading, and accent or dialect differences can cause transcription mistakes that surface as apparent reading errors. This is why every word-level result is reviewable and overridable by the teacher, why the report shows confidence information, and why FluencyScope should inform — never replace — a teacher's own listening and judgment.