agent teacher · docs
Fraction pack review and pilot
Status: prepared. Human expert review and a real-user pilot have not been conducted by this task. This protocol is not a participation record or readiness certificate.
Content appraisal
The inspected pack is fractions-vi-0.1.0. Pin its SHA-256 from the implementation manifest before reusing this review. This is an AI-assisted arithmetic/script review, not independent human expert approval.
| Case | Mathematical check | Remaining teaching judgment |
|---|---|---|
| fraction-magnitude | 1/8 < 1/5; 1/7 > 1/9; 3/4 > 2/3; 2/3 > 3/8. The wrong opening answer is intentional learner input. | Can probes distinguish an overgeneralized whole-number rule from incomplete recall? |
| fraction-correct-reason | 8/9 > 7/8 since 64 > 63; complements are 1/9 and 1/8. 3/4 = 21/28 > 16/28 = 4/7; 3/4 > 2/3 > 4/7. | Is the 2/3 benchmark clear to the intended novice? Does feedback allow a valid procedural explanation? |
| fraction-language | 1/8 < 1/5 and 1/6 < 1/4. Although 1/2 > 1/3 as numbers, cake portions cannot be ordered without whole-cake sizes. | Does short sign instruction fit the evidence without unnecessarily reteaching the concept? |
The authored transfer keys are probe, model and units. They answer the information need within the options. The distractors are conspicuously poor, so high scores could reflect recognizing desirable wording. A separately reviewed follow-up should include several defensible alternatives before drawing conclusions about transfer.
No arithmetic correction was required by this inspection. Documentation preserves pack bytes and review labels. General research abstracts do not establish these exercises’ effectiveness; scholarly source appraisal remains open.
Named expert review
A mathematics educator or teacher educator should review the exact pack revision. Record concerns by case/branch, rationale, acceptable alternatives and whether a transfer key needs changing.
| Review field | Entry |
|---|---|
| Reviewer name and relevant experience | Pending |
| Date, pack version and SHA-256 | Pending |
| Mathematics and representations | Pending |
| Plausibility of student replies | Pending |
| Acceptable alternative teaching moves | Pending |
| Transfer ambiguity and wording cues | Pending |
| Vietnamese wording and accessibility | Pending |
| Decision, corrections and disagreements | Pending |
A changed pack needs a new version/fingerprint and review record. Export existing sessions before selecting a new pack; the current runtime refuses replay against changed content.
Pilot protocol
Recruit consenting adult teachers, tutors or mentors through a separate authorized action. Explain that replies/feedback are authored, free text is not semantically assessed and the workflow does not certify competence. Avoid identifiable student work. Agree on access and deletion before collecting data; use a separate session directory.
Ask each participant to complete a case, explain what feedback means, reopen the saved session, try a different case and export a reflection. Observe completion, confusion, relevance, error recovery and changes in the next intended teaching move. Participants may stop at any time.
Use a separately reviewed unseen scenario for follow-up. Independent educators should assess actions/reasons without knowing who used Agent Teacher, recording valid alternatives and disagreements. Simulator scores, software tests and a small usability pilot cannot establish learning gains.
| Observation record | Entry |
|---|---|
| Participant code and consent | Pending |
| Task, case, fingerprint and session ID | Pending |
| Completion and assistance required | Pending |
| Interpretation of authored feedback | Pending |
| Teaching move before/after practice | Pending |
| Unseen-scenario response and review | Pending |
| Product change supported by observations | Pending |
Choose a next workflow after reviewing observations. Lesson planning, transcript coaching, portfolios, agent adapters and additional domains remain options. Keep usability, teaching-action quality and actual learner outcomes distinct.