Integrity
Watch the whole sitting back, with the evidence on the same timeline
Most assessment platforms hand a reviewer a risk score and a list of flags. This one replays the candidate’s entire sitting keystroke by keystroke, draws the findings on the same axis as the typing, and shows the suspicions the system examined and dropped.
- Playback speed
- 0.5x – 8x
- A two-thousand-event sitting at 1x is several hours. 8x is not a luxury.
- Dead time
- Skippable
- Gaps with no typing are computed on the server and jumped while playing.
- Marker positions
- 3 kinds
- Exact, nearest, and outside the recorded typing — which is listed separately, not pinned.
- Score changed by replay
- Zero
- The screen adds no judgement of its own. It shows; it does not grade.
What is being replayed
Not a video. Every change the candidate made to the editor was logged as a range replacement — the coordinates of the text that went away and the text that took its place — so the server holds a complete, ordered account of how each answer was built. Replay reconstructs the document from that stream, which means the player can stop at any keystroke and show the code exactly as it stood at that instant.
Every question's stream is merged into one chronological timeline, and that merge is what a per-answer player cannot give you. A question boundary is a tab switch: the lane strip under the scrubber is the map of the candidate moving between questions, with the density of edits drawn per question, and it makes visible the thing a reviewer is really trying to establish — the order in which everything happened.
Which question was open
A lane per question, so the moment a candidate abandons one and starts another is a shape rather than a timestamp to be cross-referenced.
How the answer was built
Per-question edit bars show where the work actually happened. An answer that grew steadily and one that appeared in three bursts look nothing alike.
The findings, in time order
A tab switch is one dot, a paste is one dot, and a large insertion the server derived is one dot — all on the same axis as the keystrokes.
The code at that keystroke
Select any marker and the editor pane shows the document as it stood at that exact event, not the final submission.
Moving through two hours without watching two hours
A reviewer with a shortlist of twelve is not going to sit through twelve sittings. The controls are the difference between the evidence being used and the evidence being ignored.
- Speed from 0.5x to 8x. Half speed exists for inspecting a burst of activity closely; 8x exists because a two-thousand-event sitting at real time is not something anyone will do twice. The player advances in fixed ticks with speed-scaled batches rather than a speed-scaled timer, so 8x is genuinely 8x instead of whatever the browser's minimum timer allows.
- Skip dead time. The stretches where nobody typed are computed on the server and jumped over while playing, so a forty-minute thinking pause does not cost forty minutes of review.
- Jump to a finding. Every marker is a seek target. In practice most reviews are three clicks: open the worst finding, watch the thirty seconds around it, look at the code that was in the editor before and after.
- The clip, where one exists. Findings that had a recorder running carry a short clip of the screen at that moment, linked from the marker. Findings derived after the fact on the server do not, and the interface says so instead of showing an empty player.
Where a marker is allowed to sit
This sounds like a detail and it is the whole credibility of the screen. A timeline that places evidence confidently in the wrong place is worse than one that admits it does not know.
The edit stream runs on the candidate's clock. Some evidence exists only on the server's. Lining the two up is an estimate, and the interface is built to say which kind of estimate each marker is.
Exact and nearest are drawn on the bar
The server returns the edit event a finding belongs to, along with how it matched. Where a finding lands on or beside a real keystroke, the marker sits there.
Outside the recorded typing is listed separately
A tab switch during the multiple-choice section genuinely has no keystroke to sit on. Pinning it to the nearest one would be a lie that looks like an answer, so those findings are listed below the timeline instead of drawn on it.
An inferred position is drawn hollow
Where the server had to shift a marker to reconcile two clocks, the dot is hollow and its detail panel says how the offset was measured and how tight the alignment was.
Nothing on this screen scores anything
Severity, weight and the integrity number are resolved server-side from your organisation’s own rules, exactly as they are everywhere else in the product. The replay is a viewer.
The findings that were cleared
This is the part that does not exist anywhere else, and it is the reason this page exists.
When the insertion analysis finds a large block of text that arrived at once, it runs a set of provenance guards before it raises anything: was this a reformat of what was already there, was it the candidate's own text being moved back, was it already present elsewhere in the same document, did it come off the question's own prompt or starter code, or was it the candidate's own typing carried over from another question in the same sitting.
When a guard accounts for the insertion, the obvious engineering choice is to discard it — it is not a finding, so why keep it. This product keeps it. The clearance becomes an item on the timeline in its own right, drawn as a hollow green marker, badged “No finding was made”, naming the guard that accounted for it and carrying the sentence that explains why.
The same principle runs through answer similarity: a match ruled out because the tokens came from the question's own template, or because that run of tokens turns up across a large fraction of unrelated answers in your corpus, is a recorded verdict rather than a deletion. Read how those guards work on proctoring and integrity evidence.
Why this changes the decision, not just the interface
Three things become possible once evidence is watchable rather than scored.
- You can disagree with the tool. A score you cannot inspect has to be either believed or ignored, and in practice teams ignore it — which means the proctoring was theatre. A finding you can watch is one you can overrule with a reason.
- You can talk to the candidate about it. "Our system flagged you" is not a conversation. "At 47 minutes, 312 characters appeared in your editor in one event — walk me through what you were doing" is. Most of the time the answer is good, and you have learned something either way.
- You can defend the decision later. Where a hiring decision is challenged, the record is a reconstruction of what happened rather than a number nobody can explain. It also makes the opposite case: the clearances are contemporaneous evidence that the system examined a suspicion and rejected it.
What replay does not tell you
- It replays the code editor. Multiple-choice answers, uploads and spreadsheet work do not produce a keystroke stream of the same fidelity, so the timeline is thinner there and findings in those sections are usually the ones listed as outside the recorded typing.
- It cannot show a stretch that was never recorded. Where the recording has a gap the timeline says so, and that gap is worth nothing against the candidate — a crash and a deliberate escape look identical from here.
- It shows what arrived and when, not why. A block of code that a guard did not account for is a question for the candidate, not an answer about them.
- It is a review tool, not a monitor. Nobody watches a sitting live from this screen; it exists for the review that happens afterwards.
Elsewhere on this site
Screening developers
Answers that execute against test cases in 22 languages, with the console output the candidate actually saw.
Take-home projects
A multi-file repository against a deadline, reviewed as a diff — and an honest account of what a take-home cannot prove.
Campus and volume hiring
A roster, an assessment window, and one live console that answers the only question that matters on the morning: how many turned up.
Remote and distributed hiring
One-way video screening, live interviews in the browser, and scheduling that works in the candidate’s clock rather than yours.
The only way to judge this is to watch one
Build a paper, sit it yourself as a candidate, paste something in halfway through, and then open the replay. Ten minutes end to end, and it tells you more than any feature list on any vendor’s site, including this one.