An error becomes a practice signal.
The animation explains the architecture; no live actions are executed.
Errors relative to attempts guide word selection. The diagram does not use the visitor’s test history.
- History
-
- Comparable tests
Use eligible word tests with the same language and layout. A result from a different task should not automatically guide the same exercise.
- Errors / attempts
-
- Per key and pair
Use relative error burden, recency and sufficient attempts. Many errors on a frequent key do not automatically indicate high relative weakness.
- Target patterns
-
- Up to three
Once there are enough attempts and errors, I give a small selection of weak patterns extra weight. This explicit selection rule guides practice words; it does not predict how quickly someone learns.
- General words
-
- Retain variety
General words remain in the mix. If evidence is too thin, practice starts without designating weak patterns.
- Existing corpus
-
- No invented words
Selection uses existing language data. Weighting changes exposure probability, not the origin or meaning of the words.
- Practise again
-
- Feedback on input
New input creates new signals. Subsequent selection can change; this map promises no fixed learning rate.
Connections
- History → Errors / attempts
- Errors / attempts → Target patterns
- Errors / attempts → General words
- Target patterns → Existing corpus
- General words → Existing corpus
- Existing corpus → Practise again
- Practise again → Errors / attempts
How to read this
- Start with relevant local test data.
- Assess errors relative to attempts, with sufficient evidence.
- Extra attention to weak patterns stays combined with general practice.
- Select from existing words, not AI-invented text.
- Input informs the next assessment. The exercise remains adaptive, not certified.