The Online Assessment Flag Reference, Each Logged Event Keyed to Its Reply
For any event a coding or skills test can log, you can look up what records it, what honest habit trips it, how to prevent it, and what to send if it is flagged.
Key takeaways
- Codility runs its similarity check only on submissions at or above the employer's passing score, or 40% if none is set, so only a strong honest attempt can be flagged for similarity.
- HackerRank's own support documentation says that solving a coding question in an external IDE and pasting it in is likely to be flagged, so write your code in the platform editor from the first line.
- HackerRank reported that MOSS similarity flags had false-positive rates as high as 70% in its own internal research, and its 93% accuracy claim for its AI detector comes with no published methodology.
- Assessment platforms act as data processors and redirect candidates to the employer, so the only appeal routes are the recruiter and a data access request to the employer as controller.
- CodeSignal proctoring data is reviewed only by CodeSignal and deleted within 15 days, so a data request about camera evidence has to go in fast.
- In Refolk's index there are 702 technical recruiters in the UK against 21,154 in the US, so a UK appeal often comes down to one person reading one short note.
This is a lookup table for anyone with an online coding or skills test link open. It covers each event an assessment platform can log: what records it, whether it is on by default, the honest habits that set it off, the setup step that prevents it, and what to send if it is raised. It is written for the candidate, not the hiring team, and it covers HackerRank, Codility and CodeSignal as far as their own documentation goes. Where a vendor has not published how something works, I say that instead of guessing.
What does an online assessment record about you?
An online assessment can record four layers of evidence. The first is what you do in the editor. The second is what your browser and machine do around it. The third is what the camera sees, if proctoring is on. The fourth is what automated analysis concludes about your code after you submit. None of it is shown to you. It all goes into a report for the employer.
Three terms come up throughout:
- Flag: an automated marker on the employer's report saying an event or pattern needs human review. It is not a verdict.
- Proctoring: monitoring of the test session itself, by webcam, ID check or device checks such as counting monitors.
- Similarity check: comparing your submitted code with other submissions and known solutions.
What an assessment can log, outermost first
- Post-submit analysisCode similarity, AI-assistance rating, composite risk score
- Camera and identityWebcam snapshots, phone or tablet in frame, photo ID match
- Browser and deviceTab or focus loss, monitor count
- EditorPaste events and content, task-text copies, keystroke timeline and playback
The three main platforms split these signals differently. This table comes from each vendor's support pages. Cells marked "secondary" come from third-party write-ups.
| Signal | HackerRank | Codility | CodeSignal |
|---|---|---|---|
| Code similarity | Every test (MOSS) | All customers, at or above the score gate | Inside Suspicion Score |
| Paste logging | On by default | Behavioural signal | Every test |
| Tab/focus loss | Disputed (off or on) | Behavioural signal | Inside Suspicion Score (secondary) |
| Webcam/ID | Per test | Optional ID verification | Shown in advance if required |
| Who sees camera data | Employer | Employer reviewer | CodeSignal only, deleted in 15 days |
Two gaps in what is published matter to you:
- The HackerRank tab default is disputed. A HackerRank-hosted article says copy/paste tracking is on by default for all tests and Tab Proctoring is off by default. Several candidate-facing posts say both are on. Your screen cannot show you how the employer configured the test.
- What the consent screen lists for non-camera signals is not public. CodeSignal says you will see in your Assessments section whether proctoring is required before you start, and a camera permission prompt is hard to miss. No platform documents whether the pre-test screen mentions paste or tab logging.
The event reference: one entry per logged event
Find the event your recruiter named, or the one you are worried about, and read only that entry. Each entry has the same six fields: what records it, the default, what the consent screen tells you, the honest triggers, the setup step that stops it, and what to say if it is raised.
Tab or focus loss
- Records it: the browser, which reports when the test window loses focus. HackerRank calls this Tab Proctoring. Codility lists it as a behavioural signal. Secondary write-ups say CodeSignal folds it into its Suspicion Score, a composite risk rating.
- Default: disputed on HackerRank (see above). Otherwise it depends on how the employer set up the test.
- Consent screen: not established publicly.
- Honest triggers: opening documentation in another tab, switching to a calculator, clicking anything outside the window. No vendor documents whether an OS notification or a password manager pop-up counts as focus loss.
- Stop it: use one browser window with one tab, silence notifications, and quit everything else before you start.
- If raised: say what you opened and why. If the rules allowed documentation, quote the rule back.
Paste
- Records it: the editor's clipboard log. HackerRank records what you pasted and when. CodeSignal logs paste on every test. Codility treats it as a behavioural signal.
- Default: on by default for all HackerRank tests, according to the HackerRank-hosted article.
- Consent screen: not established publicly.
- Honest triggers: solving the task in your own IDE and pasting the answer in. HackerRank's AI Plagiarism Detection documentation says this behaviour is likely to be flagged. Pasting your own boilerplate or test input looks the same in the log.
- Stop it: write in the platform editor from the first line.
- If raised: ask what was pasted and at what time. The log will show your own code arriving in one block. Offer to walk through it live.
Copying the task text
- Records it: the editor or page. A secondary source says CodeSignal watches for a pattern: the description is copied, then a quiet gap, then a paste or a straight-line retype. No vendor documents this first-hand.
- Default: not documented.
- Consent screen: not established publicly.
- Honest triggers: copying the prompt into your own notes, into a translation tool, or into a second window so you can read it side by side.
- Stop it: read the task in the page. Take notes on paper or as comments inside the editor.
- If raised: say where the copied text went and why.
Typing pattern
- Records it: the keystroke timeline, which reviewers can replay. The CodeSignal AI-assist pattern above, ending in a straight-line retype, comes from a secondary source only.
- Default: playback appears wherever paste and editor events are logged. Exact defaults are not published.
- Consent screen: not established publicly.
- Honest triggers: typing out a solution you worked through on paper, or typing a memorised standard algorithm in one clean pass.
- Stop it: build the solution in the editor in small steps. Keep your paper working.
- If raised: offer the paper notes and a live walkthrough. A person who wrote the code can extend it on the spot.
Idle gap
- Records it: the timeline. A gap is only documented as part of the CodeSignal pattern above (secondary source). No vendor documents an idle gap as a flag on its own.
- Default: not documented.
- Consent screen: not established publicly.
- Honest triggers: thinking, working on paper, rereading the prompt, a short break.
- Stop it: you should not try to. If you are worried, sketch your thinking as comments in the editor.
- If raised: say what you were doing during the gap. Paper notes back this up.
Fast finish
- Records it: none of the vendor documents in my sources describe finishing quickly as a flag on its own. That is not established publicly.
- Default: not documented.
- Consent screen: not established publicly.
- Honest triggers: you have seen this task type before, perhaps in another employer's test. That case does have a documented risk: a similarity match against your own earlier submission.
- Stop it: nothing to stop. Do not pad your time.
- If raised: say you have solved similar problems, name where, and offer a walkthrough.
Code similarity
- Records it: analysis after you submit.
- HackerRank runs MOSS (Measure of Software Similarity, a standard code-comparison tool) on every test. It also offers an optional AI Plagiarism Detection model that rates attempts "High" or "Medium" and works on coding questions only.
- Codility's Similarity Check compares against over 12 million assessments and against leaked solutions found on the web. Codility states that it does not use AI to decide whether a solution is flagged. It excludes project, bugfix, MCQ, essay, upload, SQL and VS Code tasks.
- CodeSignal includes similarity in its Suspicion Score.
- Default: MOSS on every HackerRank test. Codility runs Similarity Check for all customers. HackerRank's AI detection is optional per test, but it is on by default when Proctor Mode is on.
- Consent screen: not established publicly.
- Honest triggers: very simple tasks, textbook algorithms, and your own earlier submission to another employer. Codility documents all three.
- Stop it: you cannot fully prevent it. Write in your own style, and keep a dated list of every employer that has sent you a test.
- If raised: ask which task matched, and whether the matching submission was "another candidate" from a different employer. If it was you, name that employer and date.
Extra monitor
- Records it: HackerRank Proctor Mode, which keeps checking that only one monitor is connected.
- Default: only when the employer turns on Proctor Mode, which is set per test.
- Consent screen: the invite and pre-test screens show whether proctoring is on.
- Honest triggers: a laptop docked to an external screen counts as two monitors, even with the lid closed.
- Stop it: undock and use one display.
- If raised: say it was a docked laptop and that nothing was open on the second screen.
Webcam snapshot
- Records it: HackerRank Image Proctoring, plus Image Analysis, which is on by default when Proctor Mode is on. HackerRank says Proctor Mode flags phones and tablets in the webcam feed even when they are only partly visible. CodeSignal proctoring is reviewed by CodeSignal, not passed to the employer, and deleted within 15 days.
- Default: set per test.
- Consent screen: yes, in practice. The camera permission prompt tells you, and CodeSignal shows the proctoring requirement in advance.
- Honest triggers: a phone lying on the desk at the edge of the frame, leaving the frame, someone walking behind you.
- Stop it: put the phone in another room, sit centred, and clear the space behind you.
- If raised: explain what was in frame. On CodeSignal, any data request has to go in quickly.
Face or ID mismatch
- Records it: a photo ID check, which is optional per test on HackerRank (according to a third-party summary), and Codility's optional identity verification.
- Default: off unless the employer turns it on.
- Consent screen: yes. You are asked to show ID.
- Honest triggers: no vendor in my sources publishes what causes a false mismatch.
- Stop it: sit facing a light, have the ID ready, and use the same name on the test as on your application.
- If raised: offer to verify again on a live video call. See the biometric law section below.
Banned app found on the device
- Records it: not established publicly. None of the vendor documents in my sources describe scanning your device for installed apps. What is documented is the one-monitor check and phone or tablet detection on camera.
- Default: not documented.
- Consent screen: the stated rules list which tools are not allowed. Read them.
- Honest triggers: unknown, because the mechanism is not documented.
- Stop it: quit any tool the rules disallow before you start.
- If raised: ask what was detected and how it was detected. You are entitled to know what data the claim rests on.
How do assessment flags go wrong? Failure modes and false positives
Most flags on honest candidates come from about eight predictable causes, and the vendors admit several of them. Detector accuracy figures do not help you here. A headline accuracy rate cannot tell you, or a reviewer, whether one particular flag is right.
- Paste flag. Pasting your own IDE draft or boilerplate looks like outside help.
- What it proves: code arrived in a block.
- How it lies: the log cannot say whose code it was.
- Check: the paste content and timestamps, which HackerRank records.
- Similarity on easy tasks. Textbook solutions match by coincidence, and Codility says so.
- What it proves: two files look alike.
- How it lies: short, standard tasks have few reasonable answers. Codility says SQL tasks can have 20 to 50% identical solutions by pure coincidence, which is why it leaves SQL out of the check.
- Check: which task matched.
- Same candidate, several employers. Your own earlier submission matches you.
- What it proves: nothing about outside help.
- Check: whether the match is "another candidate" from a different employer, then name that employer in your note.
- Multi-monitor. A docked laptop counts as two monitors.
- Check: whether Proctor Mode was on. The invite says so.
- Retyping from paper. A straight-line retype after a quiet gap matches the CodeSignal AI-assist pattern as a secondary source describes it.
- Check: the playback. Keep your paper notes as evidence.
- Tab default confusion. Candidates assume tab switching is not logged because they read a source saying it is off by default. Sources conflict. Assume it is logged.
- The vendor as appeal route. Writing to HackerRank goes nowhere. Its data processing addendum says it will not respond to a data subject request itself, except to confirm receipt or send you back to the employer.
- Accuracy claims. HackerRank claims a 93% accuracy rate for AI plagiarism detection in a blog post that gives no methodology and no false-positive rate. A secondary source cites "85% precision" from a support doc, and I could not verify it. The same vendor reported that MOSS false-positive rates reached as high as 70% in its own internal research. Neither figure tells anyone whether your flag is correct.
There is a pattern under all of this. Codility only checks submissions at or above the passing score, or 40% on the task if no passing score is set. HackerRank's backend avoids flagging questions where the candidate scored zero. Review effort goes to scores that could advance.
A similarity flag can only land on a score worth reviewing, so it lands on candidates who did well.
Some of these failures you can prevent and some you can only explain. Sort the event in your report by those two questions before you write anything.
Triage a flag by prevention and evidence
Step-by-step: from test link to a closed flag
The procedure has eight steps. The first three are yours before and during the test. The next two belong to the platform and the employer, and the last three are your response path if a flag surfaces. The vendors advise human review before any rejection, but whether a given employer actually does it is not published, so the steps assume you may need to push.
From invite to resolution
- Read the invite and pre-test screensSpend about five minutes noting whether proctoring is required, whether webcam and ID checks are asked for, and which tools the rules allow.
- Set up a one-screen, quiet environmentUndock to one display, silence notifications, close tools the rules do not allow, and put your phone out of camera view.
- Take the test inside the editorType your code in the platform editor, do not paste from your own IDE, do not copy the task text out of the page, and keep your paper notes.
- Let the automated checks runAfter you submit, similarity and integrity analysis runs for minutes to hours, and Codility only checks submissions at or above the passing score.
- Expect a human review of any flagThe vendors advise the reviewer to confirm or dismiss a flag, contact you, or hold a live interview, but how long this takes is not published.
- Write to the recruiter within daysName the innocent cause, offer to walk through your solution live, and ask which event was flagged.
- Send a data access request to the employerAsk the employer, as controller, for the report, paste log, playback and images, within the one-month GDPR or 45-day CCPA deadline.
- Escalate to the regulator if ignoredIf the deadline passes with no answer, complain to the ICO in the UK or the relevant regulator where you live.
Sources disagree on what happens between steps 5 and 6. Codility tells reviewers: "If you have doubts, contact the candidate, or run an interview in report mode and ask them to explain their approach." HackerRank recommends that hiring teams "refrain from auto-rejecting a candidate based on the plagiarism flag" but leaves the policy to the employer. So you may never be told. A silent rejection after a strong score is a reason to ask.
What to send if a flag is raised
Send a short note to the recruiter first. If the note does not settle it, send a data access request to the employer. Do not write to the platform: no vendor publishes a candidate dispute route, and HackerRank's own terms send you back to the employer.
The recruiter note
Keep it to one innocent cause, one offer and one question. A reviewer can confirm or dismiss a flag, so make dismissing it easy. A live walkthrough lands best when it is anchored in work you have actually shipped. The projects that Refolk pulls from your own history into a tailored resume are the ones to offer as proof that you write this kind of code.
Subject: [Role] assessment - context on the integrity flag Hi [Name], Thank you for letting me know the [platform] report for my [role] assessment carried a flag. I want to give you the cause directly. [Paste:] I drafted the function in my own editor and pasted it in at about [time]. I did not realise paste was logged, and the pasted code is my own. [Similarity:] I took a [platform] assessment for [other employer] on [date], and the matching code is likely my own earlier submission. [Monitor:] My laptop was docked to an external screen, which counted as a second monitor. Nothing was open on it. [Retype:] I worked the solution out on paper and then typed it in. I still have the notes. I am happy to walk you or an engineer through my solution live, explain the approach, and extend it in front of you. Could you tell me which event was flagged, and whether the reviewer has confirmed or dismissed it? Thanks, [Name]
Keep only the one cause sentence that applies and delete the others. Send within days of hearing about the flag.
Who reads it
The person reading your note is usually a technical recruiter, and in the UK there are not many of them. When a single recruiter at a concentrated employer handles a role, that person's judgment decides the outcome, so precision matters more than length.
| Market / title | People | Top employers in sample |
|---|---|---|
| US Technical Recruiter | 21,154 | K2 Partnering Solutions, Experis, Google, Apple, Snowflake |
| UK Technical Recruiter | 702 | Monzo, Meta, Capgemini, HSBC |
| UK Data Protection Officer | 738 | NHS, PrivaLex Advisory, Trainline |
These counts are from Refolk's index. The employer lists come from samples of 25 people, so treat them as indicative only. There is roughly one data protection officer for every technical recruiter in the UK, so a named privacy contact usually exists for your data request.
The data access request
A data access request (DSAR) asks a controller for the personal data it holds about you. Send it to the employer's privacy contact or data protection officer, not the platform. Send it quickly: CodeSignal deletes proctoring data within 15 days, and HackerRank anonymises data within 14 days once an employer asks for bulk deletion.
Subject: Data subject access request - [platform] assessment for [role] To: [Data Protection Officer / privacy contact], [Employer] I am requesting access to my personal data under [the UK GDPR / the GDPR / the CCPA]. You are the controller for the [platform] assessment I took on [date] for the [role] position, using the email [email]. Please provide: 1. The full assessment report, including any integrity, plagiarism or proctoring flags and risk ratings. 2. The paste log and the code playback or keystroke timeline. 3. Details of any similarity match, including which submission my code was compared with (redacted as needed). 4. Any webcam snapshots, ID images and image-analysis results. 5. Any record of how a reviewer handled a flag, and whether it affected the hiring decision. 6. Your retention period for this data and the recipients you shared it with. Please preserve this data while my request is open, as I understand some platform data is deleted on short schedules. I expect a response within the statutory deadline [one month under the GDPR / 45 days under the CCPA]. [Name]
Pick the law that applies to you, fill in the dates, and send it to the employer's privacy contact. Keep a dated copy.
What does the law give you? Data access and biometric rules
The employer, as controller, owes you access to your assessment data within statutory deadlines: one month under the GDPR and 45 days under the CCPA. Read those deadlines in the statutes themselves before you rely on them. Biometric laws in a few US states add rules for face and ID checks, but whether they cover behaviour-only camera flags has not been tested.
Access rights and the exam exemption
HackerRank's privacy policy names the employer as controller. Its data processing addendum says HackerRank will not answer a request itself beyond confirming receipt or directing you to the employer. So your request goes to the employer.
The UK has an exam exemption in the ICO's guidance on subject access requests. It says candidates do not have the right to a copy of their answers to exam questions. Marking information can be delayed until five months after the request or 40 days after results, whichever is earlier. The ICO adds that if the request is made after results are announced, the organisation cannot rely on that exemption. Whether any of this covers an employer's coding test is not established publicly. Ask for the flags, logs and images regardless. They are not your answers.
Biometric laws
| Law | Who can sue | Penalty | Notice form |
|---|---|---|---|
| Illinois BIPA | Individuals | $1,000 negligent / $5,000 intentional | Written release, e-signature allowed |
| Texas CUBI | Attorney general | Up to $25,000 per violation | Informed consent before capture |
| Washington RCW 19.375 | Attorney general | Up to $7,500 per violation | "Context dependent" |
- Illinois: Public Act 103-0769 (SB 2979) took effect on August 2, 2024. It treats repeated collections from the same person by the same method as a single violation, with at most one recovery.
- Texas: the attorney general enforces CUBI. In 2024 Meta settled with Texas for $1.4 billion.
- Washington: the law only applies to organisations that "enroll" a biometric, meaning they turn it into a stored template.
How to keep this reference current before each test
Vendor defaults change, and each employer configures each test, so check the setting for every test you take instead of relying on what you remember. Before each test, reread the invite and the pre-test screens. When a flag surfaces, go back to the vendor's own documents rather than third-party posts.
The documents to re-check, by name:
- HackerRank's Proctor Mode and AI Plagiarism Detection support articles, for which features switch on together.
- HackerRank's GDPR FAQ and its Data Processing Addendum, for retention and who answers requests.
- Codility's Assessment Integrity Features and Similarity Check articles, for the score gate and excluded task types.
- CodeSignal's proctoring support article, for what is required and who reviews the data.
- The ICO's guidance on subject access request exemptions, if you are in the UK.
Tests cost time, and a flag dispute costs more. Before you spend an evening on a take-home, Refolk can score how well you actually fit the posting, so the tests you sit are for roles worth defending.
Before you click start, and after you submit
- I have noted whether this test requires proctoring, webcam snapshots or a photo ID check.
- Exactly one monitor is connected and the laptop is undocked.
- Notifications are silenced and every tool the rules disallow is closed.
- My phone is out of the room, not just face down on the desk.
- I will write all code in the platform editor and paste nothing from my own IDE.
- I will read the task in the page and keep notes on paper or as editor comments.
- My dated list of every employer that has sent me a test is up to date.
- If I am told about a flag, I will write to the recruiter within days and offer a live walkthrough.
- If the note does not resolve it, I will send a data access request to the employer, not the platform, and date my copy.
Questions job seekers ask
Does an online assessment detect tab switching?
Assume it does. The browser can report when the test window loses focus, and HackerRank, Codility and CodeSignal all have some version of this signal. Sources disagree on whether HackerRank turns tab tracking on by default: one HackerRank-hosted article says it is off, while several candidate-facing posts say it is on. You cannot see the setting from your side, so keep to one window for the whole test and treat any switch away as recorded.
Can I paste code in an online assessment?
You can, but it will probably be logged and may be flagged. HackerRank tracks copy and paste by default on every test, and its own support documentation says that writing a solution in an external IDE and pasting it in is likely to be flagged. The clipboard log cannot tell that the code was yours. Write in the platform editor from the start, and keep paper notes if you think better offline.
My coding test was flagged for plagiarism but I did not cheat. Is that a false positive?
It may be, and the vendors admit how it happens. Codility says very simple tasks produce matches by coincidence, and that someone who sits tests for several companies may submit similar code. HackerRank reported MOSS false-positive rates as high as 70% in its own research. Ask the recruiter which task matched and whether the match came from another employer's test. If it was your own earlier submission, name that employer and give the date.
How do I appeal an online assessment flag?
Write to the recruiter, not the platform. No vendor publishes a dispute route for candidates. HackerRank's data processing addendum says it will direct you back to the employer. Send a short note that names the innocent cause, offers a live walkthrough of your solution, and asks whether the reviewer confirmed or dismissed the flag. If that goes nowhere, send a data access request to the employer, who is the data controller.
How do I request my coding assessment data?
Send a data access request to the employer that invited you, because it is the data controller and the platform is only its processor. HackerRank's FAQ says candidates cannot download test data from HackerRank directly. Ask for the full report, any flags, the paste log, the playback and any webcam or ID images. The statutory deadline is one month under the GDPR and 45 days under the CCPA. Send it quickly, because some proctoring data is deleted within 15 days.
Can I see whether my assessment was flagged?
Not from the platform. The flag sits in the employer's report, and none of the three major vendors shows integrity flags to candidates. You learn about a flag only if the recruiter tells you or if you get the report through a data access request. On CodeSignal, camera-based proctoring data is reviewed by CodeSignal, is not passed to the employer, and is deleted within 15 days. That limits what anyone can show you later.
Put this to work
Paste your career in once. Every application after that is written for you.
Drop a resume or a LinkedIn URL. I rank the live openings against it, rewrite the resume and write a cover letter for the best of them, and fill in the employer's form when you press the button. You read, you decide what goes out.
01Drop your resume
A PDF or a LinkedIn URL. About a minute, once.
02I rank the openings
Every weekday morning, the live catalog scored against your history. Up to 20 worth your time, not two hundred links.
03Each one is written up
Resume rewritten for the posting, a cover letter, a fit score. Press send, or let me fill in the form.
- New matches ranked and written before you are up.
- Every bullet stays inside what your history supports. Nothing invented.
- Queued, submitted, interviewing, offer: one screen, not a spreadsheet.
Free, no card. Up to 25 applications sent a week, and nothing is sent until you say so.