Games that make pairs by accident

The ones that qualify

• An escape room, physical or online, with pairs that are assigned rather than chosen. It is contrived, it is artificial, and it does one real thing: two people who do not know each other have to communicate precisely, under time pressure, about a shared object. Afterwards each of them has evidence about how the other handles being wrong, being rushed, and not knowing. That is the beginning of the knowledge that a working pair runs on.
• Any competitive format with assigned small teams. The assignment is the whole trick. Self-selected teams reproduce the existing grid exactly, because people pick the colleagues they already work with, and the event then strengthens the pairs that needed it least.
• Cooking, building, or making anything together where the output has to exist by a time. The constraint does the work.
• A game with genuinely asymmetric information, where one person can see something the other cannot and has to describe it. This is the purest version: it is a simulation of the exact thing a distributed pair does all day.

The ones that do not

Anything played individually with a scoreboard. Trivia with individual answers, step challenges, puzzle streaks. Everybody participates in parallel and nobody depends on anybody.

Anything where the group is large enough that two or three people carry it. An eight-person team playing one game together produces two active participants and six spectators, and spectating is not an exchange.

Anything self-selecting, for the reason above.

Anything where the interdependence is decorative — a shared score with no shared decisions, a team name over individual effort.

Why it still does not transfer by itself

An escape room produces a warm pair and no shared work, so the pair has nothing to live on. Three weeks later the two people are exactly where they were except that each knows the other is good under pressure.

That knowledge is not nothing. It lowers the cost of the first real request between them from high to moderate, which is the difference between a request that happens and one that does not. But it decays, and it decays faster than people expect: a few months without an occasion and the pair is back to cold with a pleasant memory attached.

So the rule for any of these is the same as for an offsite. The event starts something; a piece of work within the next few weeks is what keeps it. Without the follow-up the money bought an afternoon, which may have been worth it on its own terms and did not change the count.

The assignment is the whole trick

It is worth dwelling on, because it is the one detail that decides whether any of these formats does anything and it is the one most often left to the room.

Let people choose their own small teams and they will choose the colleagues they already work with. This is not laziness; it is sensible, because a familiar partner makes the next hour easier and more likely to go well. The result is that the event reproduces the existing grid exactly, strengthens the pairs that needed it least, and leaves everybody with the impression that the team works well together.

Assign the groups from the grid instead — cold pairs first, near-misses second — and the same hour, the same budget and the same game produce a different afternoon.

The cheapest version of all of this

Nothing in the catalogue matches a shared duty. An on-call week covered by two people, a release run by a rotating pair, a review roster where the roster is deliberate.

The panel above has three people in it and two pairs, and the second one — the only working pair on this entire team that does not involve the manager — came out of a shared on-call week between two engineers. Nobody arranged it as team-building. It cost nothing, it recurred, and it produced the one connection that the quiz, the all-hands and a year of meetings did not.

That is the case for the last part of this site in a sentence: the things that make pairs are mostly already on your team's calendar, doing something else.