Comprehension (% of points)
% correct per question
Per question
| Q# | Question | Type | PIRLS | % correct | Mean time (s) | Answer |
|---|
Fluency × comprehension
Speed alone is not fluency: a fast reader with low comprehension may just be scanning. Profiles follow the two-track fluency + comprehension model.
Comprehension processes (PIRLS)
Item quality
Discrimination: how well the question separates strong from weak readers (low or negative = check the question).
| Q# | Question | % correct | Item quality |
|---|
Question map: difficulty × discrimination
Each dot is one question, colored by its PIRLS process. Good questions sit in the middle-difficulty, high-discrimination area; dots low on both axes are candidates to fix or retire in the test bank.
Class at a glance
Comprehension processes (PIRLS)
Processes per student (% of points)
Students (click a header to sort, click a row for details)
Reading metrics
What the children said
Answered voluntarily after the test — every question could be skipped. Results appear once at least six children in this class have answered, so no single child can be recognized.
Answers per question
Each bar is one question, from “not true at all” (red) to “very true” (green).
Every survey question is optional, so each question has its own n — the number of children who answered that particular question. It can be lower than the number of submitted surveys.
Opinion next to measured reading
How the children who agreed and those who disagreed actually scored on this test. Small groups are hidden.
| Question | Agreed (comprehension) | Disagreed (comprehension) | Reading of it |
|---|
In the children’s own words
Individual responses
Click a learner to see their answers. Deleting a survey never deletes the learner or their test data.
Local LLM analysis
Checking LLM status…
Leave blank for the standard report. The report is written in the language of the test.
Past reports
Export
Download section JSON Export CSV Survey CSV Download fine-tune JSONL
The section JSON is a full portable dump (test, students, answers, events, gaze).
The entire database is the single file data/eyerec.db — copy that file
to move or back up everything.
The survey CSV has one row per learner and one column per question, using the
permanent question ids — so files from different classes and dates line up.
The fine-tune export contains every saved (features → report) pair as chat-format
JSONL, ready for Unsloth fine-tuning. Run at least one AI analysis first.