Loading your Fluent workspace…
Feedback you can use honestly
Keep useful observations. Withhold stronger claims when the evidence is missing or fails.
No independent human-rated qualification is registered for the current text checks or audio measurements. No CELPIP level, exam readiness or official result is inferred from them.
That means unassessed, not zero. A target entered in the score guide is your hypothetical input, not an estimate we have earned.
| Skill | Evidence retained | CELPIP level |
|---|---|---|
| Writing | Text excerpts, word count and limited pattern checks | Unassessed |
| Speaking | Recording duration and signal measurements; no pronunciation or language rating | Unassessed |
| Listening and reading | Answers checked against original practice item keys; raw accuracy only | Unassessed |
Feedback quality and full component estimates are evaluated separately. Two qualified independent raters grade each held-out sample blindly; a separate rater adjudicates the reference range. Development learners and prompt families stay outside the sealed evaluation.
A score claim needs complete component coverage, permitted evaluation data, assessment specialist review and evidence for the exact exam edition, skill, mode, form, rubric, prompt, model and processing versions. A material change requires evaluation again.
Policy human-evidence.v1 is a Navisa research policy, not an exam-provider standard or proof that a scale is valid.
Reports retain abstentions and confidence intervals. Model self-confidence, familiar retries, developer fixtures and transcript-only pronunciation claims cannot substitute for this evidence.
Registered human-rated evaluations: 0. No rater agreement, accuracy rate or score range is claimed.
Review one supported observation, practise the suggested focus, then try an unseen task. If a finding misreads your response, use the challenge control on your feedback; a saved challenge does not imply a human review has occurred.