What "84% compatibility" actually means: how Compromise scores a decision

When a Compromise room finishes, it says something like Arrival — 84% — suits the group, and then explains itself in a few lines. The explanation is the part that matters. But the number is worth understanding too, because it is not a vote count, not an average rating, and not a probability. It is a specific claim, and once you know what it is you can read the whole results page at a glance.

Every answer is points for every option

Start with one person answering one question about one option. Whatever they say, it becomes a number between −10 (strongest objection) and +10 (strongest endorsement):

• A rating from one to five maps straight across: 1 is −10, 3 is 0, 5 is +10.
• A yes/no is +10 if it matches the favourable answer and −10 if it doesn't. Some of these can also rule the option out entirely — more on that below.
• A choice ("what are you in the mood for?") is +10 if the option's tags overlap what you picked, and a small penalty if they don't. An empty answer means "doesn't matter to me" and contributes nothing.
• A slider ("how long can it be?") gives full marks inside your limit and falls off gradually past it. Ten minutes over your ceiling costs a little; an hour over costs a lot.
• A text note is never scored. It is shown next to the result, for people to read.

Each question also has a weight — how much it counts. "Any dietary needs?" counts more than "what kind of meal are you up for?", because it should.

The number is a fraction of the best possible feeling

Now add up everyone's points for one option, weighted. That gives a total. On its own the total means nothing — a room with ten people and eight questions has bigger totals than a room with two people and three.

So it is measured against the strongest feeling the group could have had: the total you'd get if every person had answered every question with maximum enthusiasm for this option. Compatibility is where the real total sits on that scale:

• 100 — unanimous enthusiasm. Every answer was the best possible one for this option.
• 50 — indifference. The endorsements and objections cancel out, or everyone shrugged.
• 0 — unanimous objection. Every answer was the worst possible one.

So 84% means: across everything everyone said, this option got 84% of the way from "everyone hates it" to "everyone loves it". A 70 is a solid compromise; a 55 is "nobody minds"; a 40 means the group is choosing among options it doesn't much like.

A veto is not a very low score

One kind of answer does not go into the points at all. Some questions — is this a no-go for you?, any dietary needs?, can you play it? — can rule an option out. When they do, the option leaves the ranking. It doesn't get −10; it gets removed, and it is listed separately with the person's reason.

This is deliberate. A −10 from one person can be outweighed by +10s from three others, and for a preference that is correct. For an allergy it is not. One "no" should beat ten "yes" goes into why.

A skipped question lowers confidence, not score

Here is the subtlety most scoring gets wrong. Suppose someone types four film titles into a movie room and nothing else — no runtime, no genre, no rating. Everyone answers "how long can it be?" — but the room has no runtime to check the answer against.

The naive thing is to score that question as zero for every film, which drags every film toward 50 and makes the whole result mushy. The right thing is to leave that question out — out of the total and out of the maximum it is measured against. The film's compatibility is then computed only over the questions that could actually apply to it.

The effect is that a film with less detail gets a less certain number, not a worse one: it is computed over fewer questions, so it rests on less evidence, but nothing has dragged it toward the middle. That is what lets someone type four titles and nothing else and still get a real ranking.

This is also why the same rule applies to a person who skipped a question. Their answer drops out of that question's numerator and denominator alike. They do not become a neutral vote; they become someone who didn't weigh in on that one thing.

The bar shows the shape, not just the size

Two rooms can both land on 70% and be completely different situations. In one, six people are each mildly content. In the other, three are delighted and three are unhappy. Those call for different decisions — the first is a safe pick; the second might be worth a conversation.

So the bar under the number is split into segments, one per person, coloured by how that person feels about the winner — delighted, content, indifferent, unhappy. You can see at a glance whether 70% is "everyone's fine" or "half and half", without anyone's individual answers being exposed.

The explanation is built from the same arithmetic

Every line on the results page — Everyone likes Sci-Fi, Fits everyone's time limit, Mike preferred Comedy, Anna has already seen it — comes from the evaluations that produced the number. Nothing is written by hand and nothing is a summary of something you can't see. If the number says 84 and the lines say four ticks and one "preferred something else", that is the whole story.

And once the host closes the room, all of it freezes. The number, the bar and the lines are stored as they were at that moment, so a later change to how the scoring works cannot rewrite a decision a group already made.

Make a room and read the results page with this in hand. It should make sense the first time.