How accurate are our score estimates?
Honest answer: we don't know yet, and we won't pretend to. Here is where our checks stand today, what we'll publish, and when.
Status: not yet calibrated
No teacher calibration has been published yet. Every Writing and Speaking score on OETAce is an AI estimate from our own conversion table, and every Listening and Reading score converts an exact raw mark with our own table, because OET doesn't publish one. None of these ranges has yet been checked against OET teachers' marks or against real exam results.
Where the evidence stands today
Candidates who told us their official scores.
Scores from OET's free official practice tests.
– letters and role-plays in the pool so far, with – teacher marks in total.
Live counts, refreshed every few minutes. No personal details.
What we'll publish
AI vs teacher agreement
Qualified OET teachers mark letters and role-play transcripts blind: they never see our score or each other's. For each criterion we'll publish how often the AI matches a teacher exactly and within one band, the average difference, and whether the AI marks higher or lower. We also publish teacher-to-teacher agreement, so you can see how much teachers differ from each other.
When: once a skill has at least 30 items marked by two teachers.
Over-estimates near B
The mistake that costs you most is being told you're at B when you're not. Among work that teachers place between 300 and 399, we'll publish how often our range reaches 350 or above when the teachers put it below 350.
When: with the first agreement table. If we're too high, we'll change our conversion table and say so here.
Estimates vs real results
When candidates report their official OET scores, we'll compare them with the estimates they had on OETAce beforehand: how often the real score fell inside, above or below our range, for each sub-test.
When: once a sub-test has at least 30 reported real results.
How each score is made today
- Listening and Reading. The raw mark is exact: your answers checked against the key. Converting it to the 0–500 scale uses our table, built around the widely reported "about 30 of 42 is a B".
- Writing and Speaking. AI marks each official criterion and quotes your own words as evidence. The points go through our conversion table into a range.
- The range. It runs down from our estimate, never up, and gets wider when we're less sure. It isn't a statistical confidence interval.
How you can help
- Report your scores. If you've sat the real OET or an official free sample test, add your scores on your readiness page. Scores only: we never ask about questions or content.
- Share a letter or role-play. On a Writing or Speaking result, you can choose to share an anonymised copy for teachers to mark blind. It's optional and doesn't change your score.