The trial should reveal how the mentor teaches and what the learner can do—not merely produce a demonstration.
Explore the comparison in this briefing
Session
1: Establish the starting point
- Suggested activity
- The learner explains an existing small task or attempts a new one. The mentor teaches one bounded concept and helps apply it.
- What the household should observe
- Does the mentor diagnose the difficulty, explain clearly and leave meaningful decisions with the child?
Compare each option against the same criteria. The complete table and qualifications remain in the briefing.
There is a research reason to separate assisted performance from independent learning. A 2025 high-school mathematics experiment found that unrestricted GPT-style assistance improved practice performance but reduced later unaided performance; a more carefully constrained tutor largely mitigated that harm. This was mathematics research, not a trial of private coding mentors, so it supports a cautious assessment principle rather than a promised outcome for this service. (pubmed.ncbi.nlm.nih.gov)
A suggested two-session trial
| Session | Suggested activity | What the household should observe |
|---|---|---|
| 1: Establish the starting point | The learner explains an existing small task or attempts a new one. The mentor teaches one bounded concept and helps apply it. | Does the mentor diagnose the difficulty, explain clearly and leave meaningful decisions with the child? |
| 2: Check transfer and fit | Revisit the concept, change one requirement and investigate a manageable fault. | Can the learner use something from the first session? Does the mentor adjust help rather than take over? |
For a coding-ready teenager, ask the mentor to work through a small unfamiliar debugging example and explain how they would teach it. Use a sandbox without personal information or credentials. For a beginner, substitute a simple prediction or classification activity.
How can a non-technical parent check learning?
Use a short, friendly explain–change–debug conversation:
- Explain: “What does this part do, and why did you choose it?”
- Change: “What would you change if the requirement were slightly different?”
- Debug: “What would you check first if this stopped working?”
- Attribute: “Which parts came from you, your mentor, a tutorial or AI?”
Agree one small task without generative assistance, while allowing appropriate notes, accessibility tools and other agreed support. Do not turn the whole project into a memory test or infer dishonesty from a hesitant answer.
Record evidence using plain-language categories:
| Evidence level | What to write |
|---|---|
| Independent | The learner completed the agreed action and explained the decision. |
| With a prompt | The learner completed it after a question or small hint; record the help. |
| Demonstration only | The mentor or AI supplied the solution; the learner’s understanding is not yet established. |
| Not observed | The activity did not provide enough evidence to judge. |
This is a discussion rubric, not a psychometric score or an AI-authorship detector.
The trial decision checklist
- The learner wants to continue and can raise questions comfortably.
- The mentor responds constructively to confusion.
- There is at least one specific example of learning or a credible diagnosis of what needs work.
- The proposed project fits the observed starting point.
- Mentor and AI contributions are acknowledged.
- Actual contact, tool use and reporting match the agreed arrangements.
Stage exit: Approve a short extension, reset the scope or stop. Do not require every skill to be mastered in two sessions; require useful evidence and a credible next plan.