Integrity report
How to read the AI-assisted integrity and code quality report, including every signal, score, and recommendation bucket.
Integrity report
After a candidate submits, the platform analyzes their work and produces a report. It appears on the candidate detail page once analysis completes. This page explains what each part means.
Top-level scores
| Score | What it measures |
|---|---|
| Composite | Overall assessment score combining integrity and quality |
| Integrity aggregate | How trustworthy the submission is |
| Quality aggregate | How good the code is |
| Consistency | How consistent the work is across problems |
| Recommendation | A hire recommendation bucket |
Scores are color-coded by threshold, exactly as in the product: above 0.7 is
success, above 0.4 is
warning, and at or below 0.4 is
destructive.
The recommendation uses six buckets, ordered from strongest to weakest. Each maps to the same color the product uses on candidate rows and the dashboard chart:
Strong Hire Hire Lean Hire Lean No Hire No Hire Strong No HireIf an integrity guard was triggered, a red Integrity Flag is shown alongside the scores.
Integrity signals (anti-cheat)
These look for signs a candidate may have had inappropriate help. Each is shown with when it happened, so you can correlate it with the video and code timeline.
| Signal | Meaning |
|---|---|
| Natural Evolution | How much the code developed naturally through normal editing |
| Bulk Paste Safety | Inverse of the probability that large blocks were pasted |
| Original Code | Inverse of the probability that the code was transcribed or copied |
| Logical Coherence | Inverse of the probability of sudden logical leaps inconsistent with the session |
Signals are leads, not verdicts
A single paste event or a brief tab switch is not proof of cheating. Candidates look things up. Use the video and code replay to judge the context.
Quality signals
Quality is summarized as a radar with eight axes:
- Code quality
- Maintainability
- Readability
- Efficiency
- Modularity
- Code maturity
- Debugging efficiency
- Refactor quality
Each axis is graded 0.0 to 1.0. A missing axis (one the platform could not
assess) is shown as absent rather than zero, so a candidate whose recursion was never
tested does not plot a "0" for code maturity.
Development metrics
For each problem the report includes behavioral metrics:
| Metric | What it is |
|---|---|
| Time to solve | From first keystroke to final submission, in seconds |
| Iteration count | Number of distinct edit bursts |
| Judge attempts | How many times code was run against test cases |
| Lines of code | Final line count |
| Deletion ratio | Fraction of written characters that were later deleted |
| Characters per minute | Typing cadence |
These help distinguish steady work from abrupt insertion.
Per-problem and cross-problem
The report breaks down each problem individually and also surfaces cross-problem patterns: repeated weaknesses, near-miss flags, and improvement suggestions. An executive summary ties it together in plain language.
Timeline pins
The report includes timeline pins for each problem. Each pin marks a notable moment:
| Pin type | What it marks |
|---|---|
| SUSPICIOUS PASTE Suspicious paste | A bulk paste flagged by the integrity model |
| BUG FIX Bug fix | A moment of debugging that produced progress |
| ACCEPTANCE Acceptance | A test that went green, or an insight moment |
| Other types | Scaffold, refactor, struggle, and more |
Clicking a pin seeks the video and loads the code snapshot at that moment. See Video and code replay for the synced timeline.
AI assists, you decide
The report gives scores and a recommendation, but the final hiring decision is yours. Every signal is reviewable alongside the actual video and code, and the screening verdict is recorded separately from the model output.