← All case studies
Live forecast · prediction locked to disk before a single comment existed to read

We predict comment
sections before they exist.

The Sidemen dropped a video with F1 star Lando Norris. The same day, before reading a single comment, our panel of 16,191 real fans wrote its forecast of the comment section to disk: how many would love the guest, how many would complain, what they'd ask for next. Then we read all 1,275 comments and graded ourselves.

01 — the receipt

Locked at 23:08. Graded by midnight.

Four measurable predictions about a comment section that didn't exist yet, each a share of what fans would actually post. Grey is what really happened, orange is what we called in advance:

what fans posted our locked forecast
Praise Lando as the guest
11 · 12
Negative about the video
19 · 13
Ask for more F1 guests
4 · 8
Bring up KSI leaving
8 · 21
5.9
points — average distance from reality, across every measure, fully blind. The guest-reaction call landed 1.1 points from the truth.
02 — the one that matters

It reads the cut like a fan. And catches the flop.

A month earlier the Sidemen uploaded a video that looked like a guaranteed banger on paper — big-concept challenge, full cast, main channel. The comment section turned on it: one team stopped trying, one member got left waiting, and fans were furious. Could anything have seen that coming?

Panel shown the concept
78%

approval predicted. The premise is genuinely great — every concept-level read greenlights it.

Panel shown the cut
34%

red flag. The engine audits the footage like a viewer: who drives, who coasts, who gets a rough deal.

What fans really said
46%

approval across 5,071 judged comments — one of the channel's worst-received videos in months.

What our audit of the cut flagged, before any fan saw it: “the outside teams frequently stall… members waiting 10–20 minutes between clues, largely in isolation, sometimes ignored… a clear disparity in urgency.”
What fans posted hours later: “This had so much potential… too bad one team didn't bother trying at all. Ruined the video.”

Same complaint, word for word — except one of them was written before upload, while there was still time to recut, retitle, or hold the video. That's the product: the comment section, delivered early enough to act on.

03 — how it works

Real fans, distilled. Then pointed at your next upload.

No demographics, no guesswork. The panel is built from what this exact audience has actually written — hundreds of thousands of their real comments, distilled into 249 living segments. Show it a concept, a cut, or a brand integration and every segment reacts in character. Their answers are calibrated against the channel's own measured history, so the output isn't a vibe — it's a number with a track record.

Test the video before you upload it.

Every upload is a bet — on a concept, a guest, a sponsor, an edit. Right now the first feedback arrives when it's too late to change anything. Our panel gives you the comment section a day early: within striking distance on the calls that landed, and loud about the one that wouldn't. The fans already know. Ask them first.

Blind test: “GUESS THE F1 DRIVER ft. LANDO NORRIS” (published 2026-07-20); forecast locked to disk 23:08 UTC same day, before comment fetch (data/validation/sidemen_live/eventcast_o6oq3pcaKeU.json); key = 1,275 comments individually stance-judged. Flop-catch: “SIDEMEN ABANDONED IN CONTAINERS”, 5,071 comments judged; concept-brief vs cut-informed runs both on record. Panel: 249 personas distilled from 16,191 real Sidemen commenters. Bars scaled ×2.5 for legibility; printed numbers are the real percentages.