Last updated: 2026-07-12
Contents
Luminid uses an automated system to score simulation responses against documented criteria. It reads the work you submit, checks it against criteria written and reviewed by people, and quotes the exact lines from your own response that support the result. It evaluates the work, not the person. People, not the system, make hiring decisions. You can request human review of any automated score.
This page is the full public statement of where automated scoring is used on Luminid, what it evaluates, what it never evaluates, and how to exercise your rights. It is written to be read, not skimmed past. If anything here is unclear, write to hello@luminid.org and we will answer plainly.
Automated scoring is used in one place: scoring the responses you submit in a Luminid simulation. That covers written responses and, for voice simulations, the transcript of what you said. Deterministic formats such as multiple choice are scored by fixed answer keys, not by AI.
For live voice simulations, an AI voice service conducts the call in real time and transcribes your spoken answers. The transcript is then scored the same way a written response is scored. You are told this before the call starts and you choose whether the audio is recorded.
When you apply to a job on Luminid, an automated calculation also combines your skills and any verification results into a match estimate that is shown to the employer to help them prioritize review. It is one input among many, it is not shown as a verdict on you, and no application is accepted or rejected by it.
Automated scoring is not used anywhere else. Your profile is not secretly scored. Your messages are not scored. Practice results are always labeled practice and are never a verification.
The system evaluates the substance of your response against documented, skill-specific criteria that were written and reviewed by people before the simulation went live. Every scored result must quote the exact evidence from your own response that supports it. If the system cannot point to evidence, it cannot award the point.
Scoring is against a fixed threshold, the same bar for everyone. You are never compared to other people, ranked in a cohort, or graded on a curve. The system never reads other people's responses or scores when scoring yours.
What a verification means is deliberately narrow: "Demonstrated this skill in a work sample, scored against reviewed criteria, at a fixed threshold, with quoted evidence." Nothing more. It is not a statement that you are qualified or vetted for any role, and it does not predict how you will perform in a job.
The system is instructed, in writing, to score meaning and task accomplishment only. It never evaluates:
If the meaning of your answer is clear, non-native phrasing is not penalized. The scoring instructions that enforce this are kept under version control alongside each set of criteria.
Luminid does not make hiring decisions, does not recommend rejecting anyone, and does not auto-reject anyone. Employers on Luminid see your verified evidence, your quoted work, and your application. Our terms require employers to apply meaningful human review to every hiring decision and prohibit them from rejecting, advancing, or making offers based solely on an automated score.
A verification opens a door to human attention. What happens after that is decided by people.
You can request that a person review any automated score you receive. To do that:
A person reviews your original response against the same documented criteria the system used. The review can confirm the score or change it, in either direction, and you receive the outcome with the reasoning. Requesting a review costs nothing, is never held against you, and does not remove an existing verification while the review is in progress.
You also have the right to express your point of view about a result and to contest it. Both go through the same channel.
Before any voice simulation, you are shown a consent screen. Recording of the call audio can be turned off before the call begins, and choosing not to record does not block you from taking the simulation.
Luminid never creates or stores a voiceprint. Your voice is not used to identify you, and no biometric identifier is derived from it. The artifact that is scored and retained is the transcript. If you consented to audio recording, you can request deletion of the audio at any time without losing the verification, because the verification rests on the transcript and its quoted evidence.
You are told that automated scoring will be used before you start, not after. Every AI-scored simulation shows a disclosure before you begin, in plain language, covering what will be evaluated, how, and your right to human review. You acknowledge it before starting. Deterministic simulations, which are scored by fixed answer keys, state that instead.
Luminid's published scoring methodology was withdrawn on 7 September 2026 and is being rewritten; luminid.org/methodology says what was withdrawn and why.
Write to hello@luminid.org with the subject line "AI disclosure question". We answer within five business days.
Related pages: your data rights at luminid.org/legal/data-rights, and our privacy policy at luminid.org/privacy.