Play-first Discovery

Overview

Let the trailer play first. Let the story follow.

Play-first Discovery is a TV experience where a recommended title starts with its trailer, not a wall of posters. Press up or down to change, like channel-surfing. When the trailer ends, or you press OK, the film or first episode begins.

That is the promo at the top of this page. It is tidy, and it is not quite honest. Two of its lines, “Not feeling it? Skip it.” and “It learns what you like and what you don't,” are claims I could not stand behind after five conversations. Below is the prototype, one feature at a time, with the field notes pinned where they belong.

How to read the notes: Subjects A to E are the five people I talked to. Both margin notes are mine. At the time is what I wrote down when I read the result; Looking back is what I think now. They don't always agree.

Where I started

He had already decided on his phone.

What it does

Six pieces, each with what people said.

Starts from cold. Before the system can learn from behaviour it needs a first impression. A handful of posters, “I'd watch this” or “skip,” and it starts building your stream.

First, something worth judging. My first catalogue only held films released before 2008, each with a title and a synopsis. I tried it myself and couldn't tell whether I'd like anything. So I added trailers. They were old, and the picture quality was terrible.

I swapped in a current catalogue. Before you model taste, make sure the thing on screen is worth judging. A signal can only be as good as the material behind it.

Trailer first. The first recommendation plays immediately. No wall of posters; title, tags and a one-line synopsis sit over the picture.

It shows what it's learning. A side panel lists what this choice moved, your running taste profile, and what is queued next.

Change like channels. Up and down switch trailers, the way you'd flip through channels, and the panel updates as you go.

Familiar, then. But what does a scroll mean? My prototype logged every action and gave it a weight. “Try next” counted as a weak negative: probably not a dislike, but not a vote in favour either. Then I asked people.

Three readings of one gesture. Here is a row from my own test log:

Dwell time is observable. Intent is not.

Details when you want them. Press right and a side panel opens with seasons, episodes, cast and similar titles, without leaving the trailer.

Then comes the question I haven't solved: how does a viewer open the panel? Up and down already browse. Three answers, from three people, none of them wrong:

I haven't chosen between the first two. My own answer, buttons permanently in the lower-left corner, turned out to solve a different problem. More on that below.

It explains itself. Pause, and it tells you why: which parts of your taste this title matched.

What the tour doesn't show

Three things I didn't see coming.

A button that already existed. Subject A asked for what every streaming service already has: Play, and a way to save a title for later.

I had invented a vocabulary for something that already had one. It was worth it for what it showed: a skip is only ambiguous because the button for what people actually meant wasn't there.

That is where my lower-left idea from earlier came back. Here is the row added to the prototype: move focus to Add and press OK, then back to Play and press OK.

A front door. Subject A didn't even notice the “Play First” label. When I asked where the entry should go, the answers split:

Someone who didn't need it.

That is where my notes end for now. The promo promised a feed that learns what you like. What I have instead is a smaller claim, and a guess: it may suit people who are tired of choosing, and it needs buttons that let them say what they mean. The next thing to test is whether the guess holds for someone who isn't already tired.

Appendix

Specimens, for anyone who wants the mechanics.

How the feed learns within a session, how it starts from cold, and how it explains why it holds something back.

Movie data from TMDB.