Assessment design
What a Proctoring Flag Actually Triggers
A flag routes your session to a person, and on its own it is not a decision. CodeSignal says human reviewers validate flagged activity and final decisions [1]. HackerRank tells recruiters to review flagged attempts by hand and puts its AI plagiarism model's precision at 85% [3]. Those are vendor descriptions of their own products, not audits, and what an employer does with a flag is the employer's call. Write to the recruiter and ask whether a person reviewed the session.
The takeThe flag is not the hard part. The silence after it is: the vendor pages stop at the point where a human takes over, and the candidate fills the gap with the worst version available. A company that switches proctoring on should be able to say, before the assessment starts, who reads a flag and what a flag can and cannot do to an application. That sentence costs nothing to write. Its absence is what turns a routine review into days of dread.
Where Olive fits
Open a role and see what the work shows
If the assessment you're sent is Olive, using the AI assistant is the point: it's a 40-to-60-minute assignment on your own clock, with screen capture limited to the assessment tab and no camera in the product. A person writes or edits each of six findings in words, and you're granted the identical report the employer gets, free.
Assess your shortlistWhat Does a Flag Actually Trigger?
A review by a person, in every vendor description cited here. CodeSignal's cheating-and-fraud page says sessions are evaluated by automated analysis with human reviewers validating flagged activity and the final decisions, and that proctoring results reach the hiring team within an hour 1. The page does not say what the hiring team is supposed to do with them, which is the part that decides your application.
Three other platforms describe the same shape, each about its own product.
- HackerRank highlights flagged attempts in the candidate report and lets recruiters replay keystrokes to see why a flag fired 3.
- Vervoe says its flags exist for review, and that they are designed not to penalise a candidate who used ChatGPT or the web to research an answer 2.
- Greenhouse, writing about identity checks rather than proctoring, says a failed verification does not automatically disqualify a candidate and instead indicates the need to investigate further 7.
Each of those is a company describing its own product, which is not an audit of it. The vendor builds what produces a flag. The employer decides what a flag costs you, and that second decision sits on none of these pages.
Nothing in the documentation binds an employer either way, so the same flag can be a non-event in one process and the end of another.
A flag is an input, and an input is not a finding.
Why Would a Flag Be Wrong?
Because a flag can be raised in error, and one maker publishes how often its own model does it. HackerRank's knowledge base states that its AI plagiarism model has an overall precision of 85%, meaning flagged sessions are correct 85% of the time, and that human oversight therefore remains crucial 3. Roughly one flagged session in seven is wrong by the vendor's own figure. No test set, no base rate and no recall number sits beside it on the page.
The signals feeding a flag are ordinary behaviour, which is where the misreads start. HackerRank's model looks at code-writing patterns, time taken, copy-paste activity and tab-switching 3. Vervoe compares the first 500 words of every text answer across candidates and flags duplicates, and flags a location that appears to change mid-assessment 2.
A paste from notes the rules allowed, or a phone hotspot swapping in for a dropped connection, can look to a system like something else.
The system is not reading your intent.
The anxiety is documented too. Across four waves of survey data covering 1,760 students in 105 courses at one Dutch university faculty, proctoring had no measurable effect on self-reported temptation to cheat, perceived exam difficulty or performance, while students reported higher test anxiety 5. Courses chose whether to proctor, so this is observational, and students sitting course exams are not job candidates.
The same study found the anxiety fell hardest on students without a dedicated study space, without reliable technology, or under financial stress 5. Sixteen interviews with disabled students who sit proctored exams describe fear of being misrepresented as cheating, and extra cognitive load during the test itself 6.
Sixteen interviews establish that this happens rather than how often.
What Can You Ask For After a Flag?
A human review, an explanation, and correction of inaccurate personal information. HackerRank's Candidate AI Notice lists all three as things a candidate may request, the explanation once an adverse decision is made, along with an alternative selection process or a reasonable accommodation 4. The notice points those requests at the employer rather than the platform, and it does not say the employer has to grant them. Ask anyway, in writing.
Keep the message short and factual. Four things belong in it:
- one sentence on what you were doing when the warning appeared, with no defence attached
- whether a person has reviewed the session, which is the question each vendor page above implies has an answer
- what the flag was, if the recruiter is able to say
- an explanation of how the flag was used, if a decision has already been made
Two jurisdictions put part of that list into law. A candidate living in New York City must be told at least ten business days before an automated employment decision tool is used. The same sentence of the law requires that notice to allow a request for an alternative selection process or accommodation 8. That law took effect on January 1, 2023, and it obliges an employer to allow the request, not to grant it.
In the EU, GDPR Article 22(3) covers decisions based solely on automated processing that carry legal or similarly significant effects, taken on contract or explicit-consent grounds. It requires human intervention, a chance to state your point of view, and a route to contest 9. Whether a person's review takes a flag outside that definition depends on how much the review actually decided, which is another reason to establish whether yours was reviewed.
Where neither applies and no other law gives you a right, the ask rests on the employer's own policy.
What Are Employers Told About Proctoring?
That a flag is weak evidence, at least in the guidance written for them here. The employer-side piece on proctoring versus an AI-open assessment argues that watching a candidate mainly raises the cost of hiding an assistant. A task built on material no model has seen does more to keep an assessment honest. That is the argument being made to the people reading your flag.
The companion piece on what an employer owes a candidate the AI screened out sets the bar at three things: notice that the tool was used, an accurate account of what it evaluated, and a person who can look again and correct wrong data.
Neither piece tells you what the employer holding your flag believes. Both argue against letting an automated output end an application on its own, which makes asking about yours an ordinary question.
If the flag has already been put to you as an accusation, what to say when you're wrongly accused of using AI covers the reply: ask what the finding was and what produced it, in writing, before defending anything.
Send a short message, ask whether a person reviewed the session, and keep applying while you wait for an answer.
A flag nobody asks about stays whatever the reviewer assumed.
Common questions
Does a proctoring flag mean I've been rejected?
No. The published vendor descriptions named here treat a flag as an input to a human review. CodeSignal says reviewers validate flagged activity and the final decisions. Greenhouse says a failed identity verification does not automatically disqualify a candidate and instead indicates the need to investigate further. Those are companies describing their own products, not audits, and the employer still decides what a flag costs you. Nothing on any of those pages obliges an employer either way.
How accurate is AI plagiarism detection on a coding test?
HackerRank publishes an overall precision of 85% for its AI plagiarism model, meaning flagged sessions are correct 85% of the time, and says human oversight therefore remains crucial. Roughly one flagged session in seven is wrong by that number. The page carries no test set, no base rate and no recall figure, so it says nothing about how many sessions go unflagged, and the feature covers coding questions only. It is the vendor's own measurement rather than an independent one.
Should I email the recruiter, or wait to be told?
Write, and keep it short. One sentence on what you were doing when the warning appeared, then the question that matters: has a person reviewed the session. HackerRank's candidate notice lists human review, an explanation of how AI was used in an adverse decision, and correction of inaccurate information as things a candidate may request. It sends those requests to the employer rather than the platform. Nothing obliges an employer to answer outside a jurisdiction that writes the right into law.
What behaviour actually produces a flag?
It varies by platform, and each one publishes only part of it. HackerRank's model looks at code-writing patterns, time taken, copy-paste activity and tab-switching. Vervoe compares the first 500 words of every text answer across candidates, flags duplicates, and flags a location that appears to change during an assessment. CodeSignal's proctoring, where an employer switches it on, records video, audio and screen activity for the whole session and matches a government photo ID to the person on camera.
Does proctoring land harder on some candidates than others?
The published research says yes. A survey study covering 1,760 students in 105 courses at one Dutch university faculty found higher test anxiety under proctoring, with no measurable change in self-reported temptation to cheat, perceived difficulty or performance. The effect was worse for students without a dedicated study space, without reliable technology, or under financial stress. Sixteen interviews with disabled students who sit proctored exams describe fear of being misread as cheating and added cognitive load. Both are education studies, and neither measures a hiring process.
References
- 1. Cheating Detection & Fraud Prevention codesignal.com Supports the claim that a flagged session is validated by human reviewers who also make the final decisions, and that results reach the hiring team within an hour.
- 2. Vervoe's Anti-Cheating Features help.vervoe.com Supports the claim that Vervoe's flags are raised for review rather than penalty, and the description of the first-500-words duplicate comparison and the mid-assessment location flag.
- 3. AI Plagiarism Detection support.hackerrank.com Supports the 85% precision figure, the vendor's statement that human oversight remains crucial, the keystroke replay available to recruiters, and the list of signals the model analyses.
- 4. Candidate AI Notice hackerrank.com Supports the list of what a candidate may request after an AI-assisted decision: human review, an explanation, correction of inaccurate information, and an alternative process or accommodation.
- 5. The fear of big brother: The potential negative side-effects of proctored exams research.tue.nl Supports the claim that proctoring raised reported test anxiety without changing temptation to cheat, exam difficulty or performance, and that the anxiety fell hardest on students with the least stable conditions.
- 6. Surveillance and Disability in Online Proctored Exams: Student Perspectives and Design Implications arxiv.org Supports the claim that disabled students sitting proctored exams describe fear of being misrepresented as cheating and added cognitive load during the test.
- 7. Making hiring safer and more human: An update on Real Talent greenhouse.com Supports the claim that a failed identity verification does not automatically disqualify a candidate and instead indicates the need to investigate further.
- 8. Int. No. 1894-A (enacted as Local Law 144 of 2021), a Local Law to amend the administrative code of the city of New York, in relation to automated employment decision tools dfs.ny.gov Supports the New York City ten-business-day notice requirement and the statutory sentence requiring that notice to allow a request for an alternative selection process or accommodation.
- 9. Regulation (EU) 2016/679 (General Data Protection Regulation), Official Journal L 119, 4.5.2016 publications.europa.eu Supports the three Article 22(3) safeguards for a solely automated decision: human intervention, a chance to state a point of view, and a route to contest.
9 sources, numbered by first appearance. How Olive sources claims
General guidance for hiring teams. What works at one company and one volume may not transfer to yours.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.