Allocate 40 points to learning evidence and teaching quality, 15 to technical competence, and the remaining 45 to fit, evaluation, reporting and delivery. This deliberately makes a technically impressive but ineffective teacher difficult to recommend.
Compare the evidence behind the offer.
Clear the five gates first. Rate only defensible evidence from 0 to 4; use U when it is not established.
The supplied weights total 100. Edited weights are proportionally normalised to 100; minimum criterion ratings and gates still apply.
| Gate or criterion | Weight | |||
|---|---|---|---|---|
| 01 / Non-negotiable gates · no weighted points | ||||
G1. Named adult and suitability checksIdentity matched to the assigned mentor; relevant experience; references from comparable teaching; screening provider, date, scope and limitations; who reviews concerns and substitutes. Hold: screening or identity cannot yet be established. Fail: material misrepresentation or refusal to meet the household’s agreed suitability requirements. | — | |||
G2. Contact boundaries and escalationParent-visible approved communication; no disappearing or secret channels; agreed contact hours; sessions observable or interruptible by a responsible adult; a concern-reporting route that does not depend solely on the mentor. Requests for secrecy, unauthorized private contact, resistance to reasonable oversight, or no workable escalation arrangement. For a solo mentor, agree a separate adult route and an immediate pause procedure. | — | |||
G3. Tool and account eligibilityExact products and features, account type, intended user, applicable age terms, parental permissions and a dated check of official documentation. False ages, shared credentials where prohibited, or treating parental consent as an override of a product’s minimum age. | — | |||
G4. Data, recording and publicityData-flow inventory; purpose and recipients; retention/deletion arrangements; training-use settings; separate decisions about recordings, transcription, AI meeting assistants and public portfolios. Unexplained uploads, mandatory publicity without a learning need, hidden recording/transcription, or no answer about who receives child data. Use a no-recording alternative if recording cannot be appropriately authorized. | — | |||
G5. Project permissions and deploymentApproved dataset and services; no confidential employer material; defined access permissions; adult-controlled spending; ownership/export arrangements; private demonstration as the default; separate release approval. Exposed credentials, uncontrolled spending, unauthorized external actions, personal-data collection without review, or public deployment before approval. A non-deployed project can pass by explicitly excluding those activities. | — | |||
| 02 / Teaching, learning and delivery evidence | ||||
1. Learner understanding and ownershipWeak · 0–1: Only the finished project is shown; the learner cannot explain an agreed component; mentor and AI contributions are obscured. Adequate · 2: The learner explains a familiar example and makes a change with substantial guidance; contributions are recorded. Strong · 3–4: The learner explains choices, changes a bounded requirement and diagnoses a simple fault. A 4 requires evidence of retention or transfer at a later session, not just an immediate repeat. | ||||
2. Teaching and feedbackWeak · 0–1: The mentor lectures, takes over, supplies answers immediately or responds poorly to confusion. Adequate · 2: Explanations are understandable and feedback is constructive, but adaptation or checking of understanding is inconsistent. Strong · 3–4: The mentor diagnoses misconceptions, gives proportionate hints, allows thinking time and adjusts the task. Strong practice is repeated rather than performed only in a showcase lesson. | ||||
3. Relevant technical competenceWeak · 0–1: The mentor cannot explain their own example, relies on unexplained generated code or makes unsupported capability claims. Adequate · 2: The mentor can explain and modify the proposed stack, with identifiable limits or specialist gaps. Strong · 3–4: The mentor debugs an unfamiliar, appropriate example; explains data, model and application boundaries; tests fixes; and acknowledges uncertainty. Depth matches the project rather than the résumé. | ||||
4. Readiness, scope and progressionWeak · 0–1: The same ambitious project is offered regardless of starting skills; prerequisites and between-session work are undefined. Adequate · 2: A baseline and milestones exist, but scope reduction, dependencies or the learner-owned core need clarification. Strong · 3–4: The plan identifies prerequisites, a small achievable core, observable learning milestones, optional extensions and a fallback if progress is slower than expected. | ||||
5. Learner rapport, agency and accessibilityWeak · 0–1: The learner is reluctant, sidelined or pressured; reasonable communication or accessibility needs are dismissed. Adequate · 2: The learner is comfortable and participates, but has little meaningful choice or ownership. Strong · 3–4: The learner helps choose goals, feels able to ask questions and make mistakes, and wants to continue. The mentor adapts pace and participation without turning enrichment into an examination. | ||||
6. Evaluation and responsible technical judgmentWeak · 0–1: Success means one impressive demonstration; limitations, failed cases and release decisions are ignored. Adequate · 2: The plan includes basic tests and limitations, but lacks a clear comparison, failure analysis or release rationale. Strong · 3–4: The learner and mentor compare alternatives, test unseen or difficult cases, document failures and explain when not to use AI or publish the project. Any actual unsafe practice triggers a gate review, not merely a low score. | ||||
7. Evidence-based parent reportingWeak · 0–1: Updates describe enthusiasm, hours or features without showing what the learner understands. Adequate · 2: Updates connect activities to goals but inconsistently distinguish independent work from assistance. Strong · 3–4: Reports show the baseline, new capability, assistance received, learner/mentor/AI contributions, unresolved difficulty and the next learning target. Evidence can be reviewed without routine recording. | ||||
8. Continuity and household coordinationWeak · 0–1: The assigned mentor is unclear; scheduling, substitution and handover arrangements are vague. Adequate · 2: A named mentor, schedule and contact exist, but absence or replacement arrangements are incomplete. Strong · 3–4: Availability, response expectations, holiday arrangements and substitutions are agreed; transferable notes and project files make an orderly handover possible. An independent mentor may satisfy this through a clear exit plan rather than an agency bench. | ||||
9. Commercial clarity and total commitmentWeak · 0–1: The quote obscures lesson length, preparation charges, expiry, recurring payments or cancellation exposure. Adequate · 2: Core prices are clear, with a small number of unresolved extras or exit terms. Strong · 3–4: The household can calculate scheduled teaching hours, expected expenditure, cancellation exposure and exit cost. Included support, tools, usage charges and ownership terms are explicit. | ||||
Below 65 · 65–74 · 75–84 · 85–100
- Evidence coverage
- 0.0%
- Unresolved weight
- 100.0
- Possible scoring range
- 0.0–100.0
Hold the affected activity. Missing gate evidence is not clearance.
Below 65 · 65–74 · 75–84 · 85–100
- Evidence coverage
- 0.0%
- Unresolved weight
- 100.0
- Possible scoring range
- 0.0–100.0
Hold the affected activity. Missing gate evidence is not clearance.
Below 65 · 65–74 · 75–84 · 85–100
- Evidence coverage
- 0.0%
- Unresolved weight
- 100.0
- Possible scoring range
- 0.0–100.0
Hold the affected activity. Missing gate evidence is not clearance.
A complete longer-commitment assessment requires all nine criteria established, none below 2, and criteria 1, 2, 3 and 5 at least 3. The weights and cutoffs are editorial decision rules. Do not enter child identifiers or confidential material. Nothing is saved unless you explicitly choose to save it.
The distinction between successful output and learning has research support. A randomized study published in PNAS found that unrestricted generative-AI assistance could improve supported mathematics performance while harming subsequent unassisted performance; instructional guardrails substantially mitigated that harm. The study concerned high-school mathematics in Turkey—not US private AI mentorship—so it supports checking independent understanding, not predicting outcomes for this market. (pnas.org)
The 100-point rubric
The descriptions below are proposed assessment anchors. Score against the agreed learning goal: an AI-literacy mentor for a beginner should not be assessed as though the household were commissioning a production machine-learning system.
| Criterion | Weight | Weak: typically 0–1 | Adequate: typically 2 | Strong: typically 3–4 |
|---|---|---|---|---|
| 1. Learner understanding and ownership | 22 | Only the finished project is shown; the learner cannot explain an agreed component; mentor and AI contributions are obscured. | The learner explains a familiar example and makes a change with substantial guidance; contributions are recorded. | The learner explains choices, changes a bounded requirement and diagnoses a simple fault. A 4 requires evidence of retention or transfer at a later session, not just an immediate repeat. |
| 2. Teaching and feedback | 18 | The mentor lectures, takes over, supplies answers immediately or responds poorly to confusion. | Explanations are understandable and feedback is constructive, but adaptation or checking of understanding is inconsistent. | The mentor diagnoses misconceptions, gives proportionate hints, allows thinking time and adjusts the task. Strong practice is repeated rather than performed only in a showcase lesson. |
| 3. Relevant technical competence | 15 | The mentor cannot explain their own example, relies on unexplained generated code or makes unsupported capability claims. | The mentor can explain and modify the proposed stack, with identifiable limits or specialist gaps. | The mentor debugs an unfamiliar, appropriate example; explains data, model and application boundaries; tests fixes; and acknowledges uncertainty. Depth matches the project rather than the résumé. |
| 4. Readiness, scope and progression | 10 | The same ambitious project is offered regardless of starting skills; prerequisites and between-session work are undefined. | A baseline and milestones exist, but scope reduction, dependencies or the learner-owned core need clarification. | The plan identifies prerequisites, a small achievable core, observable learning milestones, optional extensions and a fallback if progress is slower than expected. |
| 5. Learner rapport, agency and accessibility | 10 | The learner is reluctant, sidelined or pressured; reasonable communication or accessibility needs are dismissed. | The learner is comfortable and participates, but has little meaningful choice or ownership. | The learner helps choose goals, feels able to ask questions and make mistakes, and wants to continue. The mentor adapts pace and participation without turning enrichment into an examination. |
| 6. Evaluation and responsible technical judgment | 8 | Success means one impressive demonstration; limitations, failed cases and release decisions are ignored. | The plan includes basic tests and limitations, but lacks a clear comparison, failure analysis or release rationale. | The learner and mentor compare alternatives, test unseen or difficult cases, document failures and explain when not to use AI or publish the project. Any actual unsafe practice triggers a gate review, not merely a low score. |
| 7. Evidence-based parent reporting | 7 | Updates describe enthusiasm, hours or features without showing what the learner understands. | Updates connect activities to goals but inconsistently distinguish independent work from assistance. | Reports show the baseline, new capability, assistance received, learner/mentor/AI contributions, unresolved difficulty and the next learning target. Evidence can be reviewed without routine recording. |
| 8. Continuity and household coordination | 5 | The assigned mentor is unclear; scheduling, substitution and handover arrangements are vague. | A named mentor, schedule and contact exist, but absence or replacement arrangements are incomplete. | Availability, response expectations, holiday arrangements and substitutions are agreed; transferable notes and project files make an orderly handover possible. An independent mentor may satisfy this through a clear exit plan rather than an agency bench. |
| 9. Commercial clarity and total commitment | 5 | The quote obscures lesson length, preparation charges, expiry, recurring payments or cancellation exposure. | Core prices are clear, with a small number of unresolved extras or exit terms. | The household can calculate scheduled teaching hours, expected expenditure, cancellation exposure and exit cost. Included support, tools, usage charges and ownership terms are explicit. |
| Total | 100 |
Why evaluation is separate from coding: UNESCO’s student AI framework includes human-centered thinking, ethics, techniques and applications, and system design—not only tool operation. NIST’s generative-AI guidance also emphasizes empirical testing and warns against extrapolating capabilities from narrow demonstrations. These are useful reference points, not endorsements of this rubric. (unesco.org)