How to Interpret Paste and Tab-Switch Signals During a Live Coding Interview
What these signals are (and aren't)
During a live coding round, the platform records two streams that show up later in the integrity report:
- Paste events — every paste into the editor, with size in characters and timestamp.
- Tab/focus changes — every time the browser tab loses or regains focus, with duration.
They are raw observations, not accusations. A paste event might be a candidate copying their own scratch from a notepad. A tab switch might be them checking the time, the docs, or — yes — an LLM. The signal is in the shape, not the presence.
This is different from keystroke-biometric profiling, which models a candidate's typing rhythm over time. Paste and tab events don't build a profile; they're a real-time event stream.
Reading the paste signal
A useful taxonomy:
Small paste, plausible source. 5–50 characters. Variable names, constants, a regex. Almost certainly the candidate copying from elsewhere on the same page or from their scratch. Note it; don't react.
Medium paste, mid-solution. 100–400 characters at minute 6. Could be the candidate's own work from a side editor; could be a snippet pasted in from documentation. Usually fine — but worth a follow-up: "I noticed a paste — was that from docs or your own scratch?" A real candidate explains. An LLM-paster will hedge.
Large paste, suspicious timing. 400+ characters that appears around the moment the candidate "finishes" the problem, or that doesn't match what they were just typing. This is the strong signal. Don't accuse — ask in the moment: "That looks like a chunk landed at once — walk me through it." A real author can; a paster can't.
No paste at all. Equally informative. Code that nobody pasted is more likely original.
Reading the tab-switch signal
A single tab switch, under 5 seconds. They glanced at the clock, a notification, their own notes. Almost never a real cheating signal.
Two or three tab switches, evenly spaced. They're probably checking the time periodically, or there's a notification spamming them. Worth flagging in the report, not worth responding to live.
One long tab switch (30+ seconds) right before a major code chunk. This is the strong shape. Pair it with a paste event in the same window and the picture sharpens.
Repeated short switches in the second half of the interview. Often a sign the candidate is stuck and is checking docs / chat / an LLM tab. Not necessarily cheating depending on your rules, but a useful signal that the round may have left their comfort zone — which is itself a finding.
Responding in the moment
The temptation is to "save up" the signals for the debrief and accuse the candidate later. Don't. The right move is to ask in real time, neutrally:
- "That looked like a paste — where did it come from?"
- "Quick check — were you looking up docs just now?"
- "Take a second; tell me what you're doing."
A candidate doing fine answers easily and keeps moving. A candidate cheating goes vague, deflects, or contradicts the timeline. The interview shifts back into a real conversation, the signal gets resolved in the room, and you don't end up litigating the integrity report a week later.
Tell candidates the rules up front: "You can use docs in another tab. You cannot use an LLM. The platform records pastes and tab switches; I'll ask if I see anything." Now no one can claim they didn't know.
What goes in the scorecard
Two fields, separate from the rubric axes:
- Signals observed — list them concretely, with timestamps. "One 380-char paste at 8:12; one 42-second tab switch at 8:08."
- Resolution — what you asked, what the candidate said, your final read. "Asked at 8:13; candidate said it was their scratch from notepad. Plausible given the function shape. No further concern."
This trail is what turns "I had a bad feeling" into a defensible decision.
How ClarityHire surfaces these signals live
The interviewer view of the live coding room shows paste events and tab-focus changes in a side panel as they happen, with timestamps and durations. The post-round integrity report renders the same events on a unified timeline alongside keystroke rate, face presence, and code-coherence flags. No event auto-rejects a candidate — that decision stays with the interviewer.
TL;DR
Paste size, paste timing, and tab-switch duration are the three dimensions to read. Small / steady = ignore. Large / suspiciously-timed = ask, in the moment, neutrally. Record what fired and how you resolved it. The signal is shape, not presence, and the conversation in the room beats any retroactive analysis.