Caveon® Observer Launch Event Recording

Observer works with updated versions of Firefox, Chrome, Safari, and Edge. So far, we’ve had good success across browsers. The biggest consideration right now is that the phone and computer need to be on the same network, though we are working to remove that requirement.

No. Exam delivery and Observer run on separate data feeds, so Observer’s calculations do not impact the responsiveness of the exam.

Yes, but it’s quite low. A speed of 3 MB per second is sufficient for all data feeds to run smoothly.

If the internet connection drops or becomes unstable, the exam is paused and the candidate is returned to the setup screen until the connection is restored and all required inputs are working again.

The second camera is optional. While it is a valuable data input, it is up to a testing program to determine if they want to use it. If the phone is not used as a data feed for Observer, then no access code is needed.

Yes. Observer can be configured to allow accessibility tools such as screen readers or similar programs to access the exam materials. This is part of Caveon’s commitment to accessibility.

Observer is designed to minimize disruption from false positives. For candidates with approved accommodations, test sessions can be configured in advance to account for their needs. For younger students or others who may fidget or show unusual patterns, you can adjust the security pre-set to take into account the behaviors that are expected. Keep in mind that it takes a very short amount of time (seconds rather than minutes) for an Observer to review most flags.

Observer supports flexibility in communication. Caveon’s Observers are currently US-based, but programs can provide their own observers or hire observers from outside organizations who speak the required languages, and localization can be configured to meet program needs. Full-scale user interface and chat bot localization are on the near-term roadmap and will be available soon. If Scorpion is used as the exam driver, the full exam delivery interface may already be localized.

Observer will support localization of instructions and agreements. This functionality is not available yet, but it is on the roadmap and coming soon.

The secondary camera is not required. Programs can configure Observer to use different data feeds and security measures depending on their audience.

Observer can be used with all test formats. It is designed to work particularly well with Randomly Parallel Tests and other secure test designs, which provide the advanced security and cost savings of Observer Plus.

No. Observer does not include scheduling functionality or test content. These functions remain the responsibility of the LMS or CMS. However, Observer doesn’t require advanced scheduling for test-takers.

Payment functionality is optional and designed to streamline the candidate experience. Payments are processed directly to the client organization; Caveon does not retain them.

Observer has been tested with LTI 1.3 compliant LMS quiz features such as those found in Canvas and Blackboard. Currently, for real-time data forensics to operate properly, Observer needs access to specific test data. This means that for now, Observer is only at maximum functionality with Caveon’s Scorpion test driver, but we are open to integrating with other drivers that can provide the necessary data in the future.

Observer can be configured to support open-book exams. Books could be uploaded as ebooks or pdfs to the test engine, or settings can be configured to not flag users looking at books in their workspace.

Yes. Observer can be configured to support oral exams with both a candidate and an evaluator present. The system can be set to allow two faces in view and conversation between them.

Observer favors an LTI 1.3 integration for consistency and security, but it can integrate with any LMS or CMS. LTI 1.3 usually requires the learner’s given name, family name, and email address. While some systems may use slightly different standards, these three fields are generally universal.

The risk score is calculated using a combination of:

• Test Design Data – for example, the structure and integrity of the test, such as the test design, item design, number of items, exposure rates, etc.

• Real-Time Testing Data – live data forensics from the test session, such as response times or answer patterns.

• Typical Proctoring Inputs – audiovisual monitoring, screen sharing, and device usage, such as detecting extra faces or unusual browser activity.

Using these signals, Observer’s AI-powered system calculates a patented risk score in real time. If a session’s score exceeds the organization’s chosen threshold, the session is flagged, and a human Observer reviews the evidence to decide whether intervention is needed. Programs determine which signals to include for their exams.

Each program decides the threshold that will trigger an Observer review. If the risk score for a test session rises above that threshold, the session is flagged and sent to a human Observer, who reviews the evidence and determines whether intervention is necessary. Please see the question above for details on what types of data are analyzed to determine the risk score.

An Observer will also be involved when a test-taker exceeds the warnings threshold (and what constitutes a warning is configurable by organization).

The third way for an Observer to be involved is when a candidate asks for help via the chatbot.

Observer is designed to be flexible. When all data sources are active and being used, it is well suited for high-stakes exams. However, thresholds and data feeds can be scaled down to meet the needs of lower-stakes exams, reducing both the intensity of monitoring and cost.

Bias in AI is a valid concern. Observer addresses this by using a “human-in-the-loop” approach. While AI can flag a potential risk, a human Observer reviews the flagged session to determine whether intervention is appropriate. If bias is suspected, the human Observer can dismiss the alert. This combination ensures fairness and avoids relying on AI judgment.

Yes. Observer’s detection thresholds are configurable. Programs can adjust the sensitivity of the system to align with the stakes of the exam, applying stricter or more flexible settings as needed. This flexibility makes Observer a strong option for a wide range of uses — from low-stakes practice exams to the most secure, high-stakes testing environments.

[As of August 22, 2025] Observer does not currently have this functionality, but is possible for the near future.

Observer leverages multiple data sources to achieve strong accuracy in flagging potential security breaches. While no system can be 100% perfect, we are confident in Observer’s performance and plan to continually improve as the system matures. For the highest level of security, we recommend using Observer Plus, which combines real-time monitoring with secure exam design to reduce risks even further.

Observer Plus is the future of testing, and where maximum security benefits and cost savings come into play. Observer Plus combines Observer’s patented test administration monitoring with secure test and item design to create virtually impenetrable test security. This dramatically reduces the need for human monitoring and allows programs to eliminate many of the most frustrationg aspects of test administration. This approach delivers greater efficiency, fewer interventions, and the highest level of test security.

That’s a great question. The best path depends on each program’s test design. In general, Observer Plus combines Observer with secure exam design approaches such as SmartItems or Randomly Parallel Tests (RPTs). Caveon can help programs identify changes that would maximize both cost savings and security. For programs interested in this, we recommend reaching out for a direct conversation.

Yes. If a test session does not trigger an alert, no human Observer is ever involved. A candidate completing their exam without issues would never know Observer was running.

Candidate identity is verified first through the client system login. Observer can also require a selfie check and additional biometric authentication if desired. Programs may also choose to include document upload and ID verification as part of the pre-exam setup.

Observer does not rely on blocking or restricting browser use. Instead, it monitors browser data and screen share to detect suspicious activity. This includes opening new tabs or windows, navigating away from the exam, or using screenshot and function key commands. When these activities occur, they are flagged and incorporated into the risk score, ensuring that suspicious browser activity is tracked and reviewed without the need for a lockdown browser.

This question was repeated frequently and deserves a frank discussion.

Any group claiming they can stop screen recording with their extension or app isn’t being realistic. By attempting to maintain blacklists of programs, services, and processes, they are desperately fighting a battle they already lost. The fact that exam content is exposed constantly – even from exams using these technologies – proves it. Even if we built an entire testing operating system so no unwanted processes could possibly be present, it would still get beat by a $25 hidden camera or monitor recorder.

That’s why we so strongly recommend secure delivery platforms that can use watermarking and, more importantly, truly secure test design such as RPT with Observer Plus. If your exam is designed in a secure way, item harvesting and pre-knowledge aren’t a true threat.

Proxy testers are a major concern for online exams. Our team has literally observed situations where someone was typing answers while their hands were clearly visible and not on the keyboard. Observer counters this with a multi-camera setup and by comparing mouse and keyboard activity against on-screen response data. This process is patented, and these checks make it extremely difficult for a proxy to go undetected.

Observer analyzes detailed mouse and keyboard activity in real time. Unless a proxy can perfectly replicate a candidate’s input patterns, discrepancies will be detected.

A VPN in and of itself isn’t always a security threat. For example, some candidates may be required to use a VPN to access secure systems. However, proxy testers and remote access tools can be detected and deterred through Observer’s use of secondary camera data, response timing, and other real-time analytics.

Yes. Observer can be configured so that a room scan is required before an exam begins, similar to other pre-exam steps like completing a payment or signing an NDA. But it is also worth discussing with stakeholders how effective a room scan is compared to the invasiveness of the test-taker experience. Room scans are the most legally challenged part of online test security.

Observer does not always require a live monitor. Programs can choose to allow test sessions to run uninterrupted and have them flagged for review later if the risk score exceeds the chosen threshold. This record-and-review option provides flexibility while still ensuring flagged sessions receive human review.

No. Observer represents a fundamental change in how exam security is managed.

Traditional live proctoring relies on a proctor watching someone take a test and trying to “see” if they are engaging in misconduct. Observer instead continuously analyzes data from three input streams: real-time testing data, test design data, and typical proctoring inputs (such as audio/video and screenshare). Most test sessions are completely normal, with no evidence of theft or cheating. Observer is designed with that in mind — it monitors the data and only flags a session when the calculated risk score exceeds the threshold set by the program. At that point, a human Observer reviews the flagged session and intervenes only when deemed necessary.

This approach avoids unnecessary interruptions and ensures test takers aren’t over-monitored while taking their exams.

Yes. Caveon offers support staff to manage Observer queues if a program chooses not to staff them internally. Programs may also choose to BYOP (bring your own person) which allows them to assign their own staff as Observers and realize additional cost savings.

Caveon’s monitors are currently based in the US, but if a program requires proctors in specific regions, we can build a proctor pool to meet those needs.

In testing so far, two Observers can monitor up to 100 test sessions at the same time. When a session is flagged, the average response time has been less than two minutes from when the alert enters the Observer queue.

The retention period for video clips and live feeds is determined by the program. OEach testing program acts as the data controller, meaning the program decides how long data is retained and when it is deleted. Programs can request bulk deletion at any time, and auto-deletion policies can be set to match program or regulatory needs. All test taker data, including videos, selfies, and other records, is stored on AWS in the United States by default. If required, data can be stored in other regions to meet specific legal or organizational requirements. Observer does not use candidate recordings to train its system.

Observer is a new product and has not yet gone through accreditation with ANSI, NCCA, or similar bodies. However, Caveon looks forward to supporting programs through those accreditation processes as needed.

Not at this time. Export functionality is on the roadmap and can be explored with clients as a potential feature addition.

For detailed pricing information, please contact us directly at the Observer demo link.

Observer is offered on a per-exam session pricing model rather than as software ownership. Programs pay per test administration.

Observer has been in beta with real test takers and we have collected feedback. When compared with other secure online testing experiences we have received very positive feedback.

Formal satisfaction studies have not yet been conducted. Unlike traditional tools that rely heavily on what a human proctor sees, Observer prioritizes the analysis of test data and real-time data forensics. As a result, most test takers experience a much less invasive testing process. Depending on the preferences of a program, certain unpopular steps like room scans, human check-in, or even video feeds altogether, may prove unnecessary for that exam — further improving the test taker experience. Caveon plans to conduct and publish formal test taker experience research in the future.

White papers and case studies are currently in development. We understand the importance of providing evidence of Observer’s efficacy and plan to publish supporting materials in the near future.

Observer pricing is designed to be much more affordable than that of in-person test centers. For detailed pricing information, please contact us directly at the Observer demo link.

After initial onboarding, setting up an individual test in Observer takes only a few minutes. The exact setup time may vary slightly depending on the rest of your technology stack, but the process is designed to be simple and efficient for program staff.

Exam takers are not required to run a connection test before their exam. However, Caveon can provide optional “System Check” links that allow candidates to try out the setup process and verify they can launch an exam through Observer. These checks are free of charge, though they require configuration with your LMS or platform.

Observer does not include built-in test content. Programs provide and manage their own tests, which can then be secured with Observer.

Caveon is actively seeking early adopters to help programs compare Observer’s security with traditional live proctoring approaches.

If you are interested in exploring this, please contact us directly to discuss partnership opportunities.

In addition, we intend to conduct secret shops and in-depth studies of the efficacy of Observer and to release the results of these studies publicly to demonstrate Observer’s effectiveness.