You test them on the meeting itself. For every request, MetaPret builds a short interpretation test from that engagement's field, in both directions of the language pair. Candidates hear each passage once and interpret it on the spot. The recordings are scored against a published 15-point rubric, and only candidates confirmed on that test are matched.
As of
Why a résumé is not enough
Most interpreters are still matched by network or price, and their skill shows only after the meeting has started. A résumé says where someone has worked. It does not show whether they can carry this speaker, in this field, in this room. MetaPret checks that directly, before the match exists.
How the test works
- The request comes in. Language pair, field, city, dates, and the format of the meeting.
- MetaPret builds a test for that request. The passages are written in the field of the engagement, so an investor call gets investor-call language and a factory audit gets audit language. Every test covers both directions of the pair, for example English into Korean and Korean into English. Passages read the way people actually speak in that room, including the self-corrections and idioms an interpreter meets live.
- Candidates take it under live conditions. A microphone check runs first. Each passage plays once, with no replay, and recording starts automatically when playback ends. Each answer can run up to five minutes.
- The recordings are scored and confirmed. An AI monitor drafts an itemized list of deductions for every recording, each one quoting the source line and what the candidate said, and flags items that need a human check. An expert evaluator listens to both directions before a score can be confirmed. The AI does not decide pass or fail, and nobody confirms their own test.
Open requests and their tests are posted on the MetaPret board.
The 15-point rubric
Each recording is scored on three axes, five points each.
- Content fidelity. Did the meaning and structure of the source come through in full? Points go for reversed meaning, a wrong figure, date or name, a dropped main point, a distorted scope or tense, and claims the speaker never made.
- Target-language fluency. Precise equivalents, in concise and natural target-language usage. Points go for a wrong field term, grammar errors, word-for-word calques, source words left untranslated, and unfinished sentences.
- Delivery. Pace, tone, and clarity for the room. Start time, length against the source, and long pauses are measured from the recording itself. Pronunciation and intonation are judged by a person listening, not by the AI.
What the client sees
On a private comparison page, the client sees every candidate whose test was confirmed for that request, under a pseudonym, with the three axis scores and the total out of 15. Real names and contact details appear after the client picks one. Individual scores are not published anywhere else.
What the interpreter gets
After the test deadline, interpreters whose test was confirmed get a private results page with written feedback in five blocks: Strengths, Omissions, Mistranslations, Delivery, and Suggestions. Each point quotes the source line and what the interpreter said. The page also holds their own recording. Scores, rank, and the number of applicants are not shown to interpreters.
What decides the match
The test is not the only gate. MetaPret assigns in this order: eligibility to work at the location, availability on the dates, then the interpretation test. A strong test on one request does not carry over to the next. The next request gets its own test.
Frequently asked questions
Q: Who builds the test?
A: MetaPret builds a new test for each request, from the field, language pair, and setting the client describes. The passages are not reused across requests.
Q: Can a candidate replay the audio or prepare the answer?
A: No. Each passage plays once, with no replay, and recording starts automatically when playback ends. A microphone check runs before the test starts.
Q: Who scores the recordings?
A: Every recording is scored against the published 15-point rubric. An AI monitor drafts itemized deductions and flags items for review, and an expert evaluator listens to both directions before confirming the score. Nobody confirms their own test.
Q: Does the highest score always get the match?
A: No. MetaPret assigns in this order: eligibility to work at the location, availability on the dates, then the interpretation test.
Q: Can I see the scores as a client?
A: Yes, for your own request. Your private comparison page shows each confirmed candidate's three axis scores and total out of 15, under a pseudonym. Names and contact details appear after you pick one.
Q: What does the interpreter get from the test?
A: Interpreters whose test was confirmed get written feedback after the deadline, in five blocks: Strengths, Omissions, Mistranslations, Delivery, and Suggestions, plus their own recording. Scores and rank are not shown to them.