Where reports live
A report exists for every completed monitored interview. Open it with View report on the interview row, or with Open on a pipeline card. Reports are generated when the session ends, so a round that is still live does not have one yet.
Score and risk band
The headline is an integrity score out of 100 and a band. Every finding deducts points, and the deductions shown against the findings add up to the drop from 100.
| Score | Band | What it means |
|---|---|---|
| 90 to 100 | Monitored | No integrity concerns detected. |
| 75 to 89 | Low risk | Minor signals only. Worth a glance, likely benign. |
| 50 to 74 | Moderate risk | Multiple signals. Review manually before proceeding. |
| 25 to 49 | High risk | Serious concerns. Review, and consider a re-interview. |
| 0 to 24 | Critical risk | Critical signals. Review before any decision. |
What we found
Findings are grouped into three buckets, strongest first:
- Correlation: several independent observations that line up, for example an AI assistant in focus followed immediately by a large paste. This is the strongest kind of evidence, because it is hard to produce accidentally.
- Interaction: what happened between the candidate and their machine, such as paste bursts or focus switching into an assistant.
- Detection: something was present, for example a known AI tool running or a virtual camera being active.
Each finding has a headline, a sentence of explanation, and the points it cost. Several observations of the same underlying thing collapse into one finding, so a tool spotted through its window title, its process, and its network traffic counts once, not three times.
The wording deliberately describes what was observed, never what the candidate was doing. "A known AI assistant was running" is a fact. "The candidate cheated" is a conclusion, and the report does not make it for you.
Telemetry coverage
If the agent was blind for part of the session, the report says so above the findings, with the coverage percentage and the longest gap. A clean report at 100% coverage and a clean report at 60% coverage are not the same result, so this caveat is stated rather than buried.
Gaps usually mean the machine slept, the network dropped, or the agent was closed and reopened. Deliberate interference shows up separately as a finding about monitoring being interrupted.
Full evidence and the timeline
See full evidence expands the complete event timeline: every recorded observation with its type, tool, severity, start, end, and duration. Each row expands again to the raw JSON for that event. The session header line states the start and end times in UTC, the length, and whether the agent stayed connected the whole way through.
Downloads and PDF
From the report page or the ⋯ menu on the interview row:
- Download summary JSON: the score, band, findings, and session bounds.
- Download detailed JSON: everything above plus the full event evidence.
- Save as PDF: opens the report and prints it, for attaching to a hiring packet.
If the interview was reconnected, the downloads become combined files covering all sessions, and each individual session's report stays available on its own.
Verifying a report
Every report is cryptographically signed when it is generated. Anyone holding a downloaded file can check it:
- Open the Interviews page
Click Verify report in the top right.
- Upload the JSON file
Pick the summary or detailed file you downloaded.
- Read the result
A valid file confirms the contents match the signature. A failure means the file was altered after signing, was signed by a different InterviewWatch deployment, or is not a signed report at all.
This is what makes the report useful to someone outside your team: a candidate, a client, or an auditor can be handed the file and check it themselves.
Using a report in a hiring decision
- Read the findings before the score. A 62 driven by one long correlated sequence is a very different conversation from a 62 made of scattered minor signals.
- Check coverage before treating a clean report as an all clear.
- Pair the report with the interviewer's scorecard. The board deliberately calls it out when a panel wants to hire on a round whose integrity score is low.
- Give the candidate the chance to explain when something material shows up. Observations have innocent explanations more often than people expect.