Use fast estimates and human feedback for different jobs

The useful answer is not “which is better?”

Automated feedback and a human tutor observe different evidence. A quick estimate can respond immediately to a writing sample or recorded answer and consistently flag visible patterns. A tutor can ask why you made a choice, adjust the task, challenge you in live conversation and notice non-verbal confusion. Neither can guarantee your test result.

Choose feedback according to the next decision. If you need to know whether your response omitted a required detail, a quick check may be enough. If you repeatedly freeze when interrupted, practise with a person.

What exists in this preview

The repository includes a free mini practice set and free instant marking endpoints for writing and speaking. The response is designed to give an estimated CEFR/NCLC range, criterion comments, corrections and a suggested next drill. Speaking can include a transcript. The system uses two model assessments and may return a range when they disagree. Daily caps protect the free service.

The product contract also describes in-depth writing and speaking reports for paid access, a submission queue, progress tracking, full mocks and other account features. Those are product plans unless they are visibly available in the current preview. This guide does not present planned features as currently usable.

What a quick estimate can do

1. Check visible task completion, approximate length and basic organisation.
2. Point to sentences that appear to contain grammar or clarity problems.
3. Suggest a more controlled version and explain the change.
4. Identify recurring patterns across one response.
5. Offer a likely range and a next drill quickly enough for a practice loop.

This is valuable when repeated: attempt, inspect, repair, retry on a fresh prompt. The output should lead to an action, not sit as a decorative score.

What it cannot establish

1. It cannot issue a TEF Canada or TCF Canada result.
2. It cannot reproduce secure test conditions, calibrated forms or the providers’ full assessment process.
3. It cannot guarantee that a different prompt, assessor or day will produce the same level.
4. It may misread an unusual but valid sentence or miss an error.
5. Speech transcription may be wrong because of audio quality, accent, background noise or overlap.
6. It cannot fully simulate a live interlocutor who changes direction or presses an objection.

That is why every estimate must retain the label: Practice estimate, not an official score.

Quick estimate vs in-depth report vs tutor

Need	Quick estimate	In-depth report	Human tutor
Immediate task-completion check	Useful	Useful with more detail	Useful, but requires scheduling
Sentence-level patterns	Selected examples	Broader evidence and revision path	Can explain interactively
Live objection handling	Limited	Limited if asynchronous	Strong fit
Pronunciation diagnosis	Constrained by recording/transcript	Can review more evidence, still recording-based	Can ask for repetition and model distinctions
Availability	Immediate, subject to caps	Planned turnaround for paid reports	Depends on booking
Test result	No	No	No

How to audit automated feedback

1. Return to the prompt. Did the feedback understand the required role and purpose?
2. For every correction, compare the original and revision. Does the revision preserve your intended meaning?
3. Verify grammar claims in a trusted dictionary or grammar reference when uncertain.
4. Accept the smallest useful correction. Do not replace your entire voice with a polished model paragraph.
5. Write a new sentence using the corrected pattern.

Feedback audit

Your sentence: “Bien que le service est pratique, il reste cher.” A correction proposes soit after bien que. The useful lesson is the subjunctive pattern. Write two fresh examples: “Bien que le trajet soit court…” and “Bien que les horaires soient limités…” Then use one in a new task.

When a tutor is the better next step

Consider a skilled tutor when the same problem survives several self-correction cycles; when you need live speaking interaction; when pronunciation causes frequent misunderstandings; when you do not understand why a correction matters; or when anxiety changes your performance in front of another person.

Bring evidence to the lesson: two timed responses, the prompt, feedback you received and a short list of questions. Ask the tutor to observe one priority, such as whether examples actually support claims. A focused session is easier to transfer into practice than a broad request to “improve my French.”

When the quick estimate is enough

Use it for frequent low-stakes iteration: testing whether your outline produces a complete response, spotting repeated agreement errors, comparing two introductions or deciding the next drill. It is also useful before a tutor session because it clears routine issues and preserves human time for interaction and explanation.

A combined feedback cycle

1. Write or record under the real time limit.
2. Self-assess purpose, structure and one language concern.
3. Request a quick estimate and audit the comments.
4. Revise one section, then use the pattern in a new prompt.
5. After three or four attempts, take the recurring issue to a tutor or experienced teacher if it remains.
6. Run a fresh timed response without aids to measure transfer.

Privacy and practical care

Do not include passports, immigration file numbers, addresses, employer secrets or another person’s personal data in practice text or audio. Use fictional details. The free speaking contract says audio is transcribed and discarded rather than stored, but you should still submit only what is needed for the exercise. Read the privacy page before using any service.

Interpreting a range

If two assessments produce adjacent levels, a range is more honest than false precision. Treat the lower bound as the risk to investigate. Look at the criterion evidence: perhaps organisation is stable while grammar control varies. Do not average the labels or round upward. Gather another fresh response and examine whether the same range repeats.

Questions to ask any feedback provider

1. What exact task and criteria are being considered?
2. What evidence from my response supports each comment?
3. Is the result a score, an estimate or general language feedback?
4. What should I practise next, and how will I know it improved?
5. How are text and audio handled?

Sources and product basis

1. CCI Paris Île-de-France: understanding TEF results, calibrated comprehension and two-person assessment for expression.
2. France Éducation international: TCF Canada, test structure.
3. PrepFranco privacy notice, current service data handling.

External sources and repository capabilities checked 26 October 2026.

Try one low-stakes feedback loop

Submit an original practice response, inspect the evidence and choose one next drill. Keep the result in its proper role.

Try free practice

Writing review methodSpeaking review method