Demo Candidate
Senior Backend Engineer · eng/L5 · candidate@example.test
aaaa1111-0001-0000-0000-000000000001
May 21, 2026, 2:27 PM
Competency rollup
| Competency | Score | |||
|---|---|---|---|---|
Code quality c.code_quality | 3.60 | 0.10 | 4 | 4 |
Debugging fluency c.debugging_fluency | 3.60 | 0.25 | 4 | 4 |
Modalities
| # | Modality | Driver | Status | Duration |
|---|---|---|---|---|
| 0 | live_code | live_code@0.1.0 | completed | 2 s |
| 1 | live_code | live_code@0.1.0 | completed | 2 s |
| 2 | live_code | live_code@0.1.0 | completed | 2 s |
| 3 | live_code | live_code@0.1.0 | completed | 2 s |
Transcript
Round 1 — The Liar Filter. 10 minutes, python. Solve this: Write `most_common(words: list[str]) -> str | None` that returns the most-frequent word. Ties: return the lexicographically smallest. Empty list returns None.
```python def most_common(words: list[str]) -> str | None: pass ```
def most_common(words: list[str]) -> str | None: if not words: return None counts = {} for word in words: counts[word] = counts.get(word, 0) + 1 best = None for word, count in counts.items(): if best is None or count > counts[best] or (count == counts[best] and word < best): best = word return best
Round 1 — The Liar Filter. 10 minutes, python. Solve this: Write `most_common(words: list[str]) -> str | None` that returns the most-frequent word. Ties: return the lexicographically smallest. Empty list returns None.
```python def most_common(words: list[str]) -> str | None: pass ```
def most_common(words: list[str]) -> str | None: if not words: return None counts = {} for word in words: counts[word] = counts.get(word, 0) + 1 best = None for word, count in counts.items(): if best is None or count > counts[best] or (count == counts[best] and word < best): best = word return best
Round 1 — The Liar Filter. 10 minutes, python. Solve this: Write `most_common(words: list[str]) -> str | None` that returns the most-frequent word. Ties: return the lexicographically smallest. Empty list returns None.
```python def most_common(words: list[str]) -> str | None: pass ```
def most_common(words: list[str]) -> str | None: if not words: return None counts = {} for word in words: counts[word] = counts.get(word, 0) + 1 best = None for word, count in counts.items(): if best is None or count > counts[best] or (count == counts[best] and word < best): best = word return best
Round 1 — The Liar Filter. 10 minutes, python. Solve this: Write `most_common(words: list[str]) -> str | None` that returns the most-frequent word. Ties: return the lexicographically smallest. Empty list returns None.
```python def most_common(words: list[str]) -> str | None: pass ```
def most_common(words: list[str]) -> str | None: if not words: return None counts = {} for word in words: counts[word] = counts.get(word, 0) + 1 best = None for word, count in counts.items(): if best is None or count > counts[best] or (count == counts[best] and word < best): best = word return best
Integrity monitoring
Multiple high-severity window-focus violations combined with a mid-interview multiple-faces detection suggest substantial dishonesty risk; human review of video at flagged timestamps is essential.
- 6:46:48Multiple faces detected (2 in frame, 1 selfie) with low similarity (0.24). This is a direct indicator of potential identity mismatch or unauthorized person in frame during interview.
- 9:42:52 – 9:43:59Cluster of 10 rapid window-focus losses in under 2 minutes, escalating from LOW to HIGH severity. Includes fullscreen exits, tab switches, and window blurs—pattern consistent with switching to external resources or unauthorized applications.
- 25:00:37 – 25:00:41Second cluster of 4 focus-loss events (LOW to HIGH) in rapid succession, suggesting deliberate window switching during interview.
- 32:51:11 – 32:51:15Third cluster of 4 focus-loss events (LOW to HIGH), with tab switch at end. Late-interview distraction or resource access.
The multiple-faces detection at 6:46:48 is the highest-priority flag and warrants immediate video inspection to confirm identity. The three distinct clusters of rapid, HIGH-severity window-focus events (especially 9:42–9:43) strongly suggest deliberate switching away from the interview window, likely to consult external materials. Together, these patterns elevate confidence in dishonesty risk. Recommend human verification of video frames at all flagged timestamps.