Interviews are a mirror of reaction. Paste a transcript — see who really said what.
How to read thisDrawn live from the transcript you paste, using the same speaker split as the decode. Each reed is one turn, placed where it starts in the conversation (by word count). Its height is the number of words in that turn, scaled to the longest turn. The speaker who says the most stands above the waterline; the next is reflected below (further speakers alternate). A ring at the tip marks a turn with a question. The colours and the sway are layout. Nothing is sent anywhere.
How auto-fetch works (and why paste is the safe path)
The Fetch button asks the backend to pull the video's captions — it only works on the deployed version, and YouTube
often blocks caption requests coming from servers, so it can legitimately fail. Pasting always works. Captions carry no
speaker names, so add a Name: in front of each person's lines for a full per-speaker breakdown.
From the video you can also: ••• / Show transcript → copy → paste above.
Same person, two interviews → what drifted
MirrorScope · everything here is computed locally in your browser from the words themselves — nothing is sent anywhere.
Counts and signature words are facts. Tone is a small word-lexicon approximation. The personality read is the
next (AI) layer — and will always be labelled inference, not a validated assessment.
Your decode library
This passcode is the only key to your library — anyone who knows it can see your saved videos, and it's how they sync across devices. Pick something only you know, and remember it (there's no reset).