What online proctoring can honestly claim

An exam platform that watches a candidate’s browser can tell you a small number of things very reliably, and a much larger number of things not at all. Selling the second set as the first is the industry’s standard mistake, and it produces records that fall apart the moment anybody contests them.

Here is what we think a browser can honestly report.

What is observable

Presence. Whether a candidate’s page is holding a live connection, and for how long. This is a real signal. A socket that closes is a socket that closed, and the gap is measurable to the second.

Focus. Whether the exam page is the foreground document. The browser will tell you when it stops being so, and the transition is unambiguous.

Reconnection. How many times a connection dropped and came back. Useful precisely because it distinguishes a bad line from a closed laptop: three five-second gaps in an hour is a train, one forty-minute gap is not.

Timing. When an attempt started, when each answer was last written, and when it was submitted. All of it server-side, none of it dependent on the candidate’s clock.

What is not

Who is at the keyboard. A signed-in session proves that somebody had the credentials. It does not prove who typed. No amount of browser telemetry changes this, and a platform claiming otherwise is claiming something it cannot support.

What is on the second screen. A browser cannot see the rest of the machine, let alone the rest of the desk. Focus loss tells you the candidate looked at something else; it does not tell you what.

Intent. A tab switch is a tab switch. It might be a search engine, a notification, or a hand slipping onto the wrong key. The platform’s job is to record that it happened, with a timestamp, and to leave the interpretation to the person who knows the candidate and the exam.

Why the distinction is the product

Because the record has to survive being questioned.

A system that reports “candidate lost focus for 8 seconds at 14:32, reconnected twice, submitted at 15:47” is making four claims it can substantiate. A system that reports “suspicious behaviour detected: 87% confidence” is making one claim it cannot, and the first time a student’s parent asks what the 13% was, the whole record becomes unusable — including the parts that were true.

So Scholasticus records the observable facts against the attempt, timestamped, and shows them to the invigilator while the exam is running so they can act on a problem rather than adjudicate it three weeks later. There is no score, no confidence figure and no verdict, because none of those would be ours to give.

The part that is not proctoring at all

The strongest guarantee we can offer is not about watching. It is about who was allowed to start.

A workspace decides its own sign-in requirements: which identity provider, and what extra credentials a member must present before an account is accepted — a student number, a cohort, an access code the institution issued. That check happens before the exam opens, and it is the one part of candidate identity that has a real answer.

Everything after it is observation, and observation is worth exactly what it can be shown to be.