Clip detail
“Shrinking - TV drama”
Curated window: 0:01 – 1:01 (60s). Plays the same window each time — replays return to 0:01.
Plays only the curated 60-second window. · Embedded via the YouTube Player API · © original creators.
Each star is one emotion. Star size and brightness scale with how many readers chose it. The most-picked reading glows in peach; the AI's reading in mint.
Full distribution
How readers split across the emotions
- Embarrassed / awkward5 · 83% (4p)
- Affectionate / tender2 · 33% (1p)
- Happy / amused1 · 17% (0p)
- Anxious / nervous1 · 17% (0p)
Modal reading (by mentions) · AI's readingSolid bar segment = readers who picked this as primary. Lighter extension = readers who tagged it as part of a Mixed reading.
AI's reading
What one AI saw on this clip
PrimaryHappy / amused
SecondarySad / sorry, Affectionate / tender
CuesFacial expression, Tone of voice, What they said, The situation
Why
The man seems gently amused by the woman's awkward reaction to his sad news, a reaction he explicitly predicted. His soft, knowing smile and light tone suggest he is trying to put her at ease, finding a bit of wry humor in a difficult social situation he has likely navigated before.
AI's own caveat
His smile could also be interpreted as a mask for deep sadness, a practiced coping mechanism rather than genuine amusement in the moment.
The AI's reading is one interpretation among many. We expose it as a single point of comparison against the human spread — not as an answer key. The rationale + caveat above are the AI's own words, reviewed and approved by a human admin before publishing.
Which signals readers used
Cue distribution
Multi-select per response — totals can exceed the reader count.
- What they said5 · 83%
- Tone of voice3 · 50%
- The situation3 · 50%
- Facial expression2 · 33%
- Body language / posture2 · 33%
- Timing / pauses1 · 17%
The finding
On this clip, 83% of readers chose Embarrassed / awkward — but the rest split across 3 other emotions. The same moment, read in many ways at once. That spread is not a problem. That spread is the data.
Methodology + transparency
All counts above include only responses kept in analysis. Internal-test profiles are excluded. Free-text responses are not surfaced on this page; they pass through human moderation first and only feature on the dataset's curated quote page (forthcoming) once enough have been approved.
Want the full anonymized dataset? Email us.