A round is one of two kinds, alternating. Letters: nine tiles are read out and you call the longest word you can build from them. Numbers: six numbers and a three-digit target, reached with plus, minus, times and divide. The passenger types what you say, and at the end the app shows the best word that rack held and the shortest route to that target, because it worked both out before it dealt them.

Players: Two or more people in the car.

Driver: Calls out words and arithmetic. Nothing to read.

The bar: The best word in the rack and the best way to the target, both found first.

The app hides a code of digits. You call a guess out loud, the passenger types it, and the app reports two numbers back: how many digits you had in the right place, and how many were the right digit in the wrong place. You work back from that. Ask for the guesses so far at any point and the passenger re-reads them. The round is cleared by cracking the code in no more guesses than the app's own solver would need at its worst.

Players: Two or more people in the car.

Driver: Guesses out loud and works back from the feedback.

The bar: The most guesses its own solver would ever need on a code that shape.

Five quantities, one at a time -- how heavy, how many, how far. You reason out loud to a number and the passenger types it in any form you say it: 3m, 3 million, 3000000. You are scored on how many TIMES off you were rather than by how much, so being within double beats being within tenfold whether the answer is in the hundreds or the quadrillions. The round is cleared if your average over the five stays inside the level's factor. Nothing here is recall; every figure can be reasoned toward.

Players: Two or more people in the car.

Driver: Thinks out loud. Nothing to memorize.

The bar: How many times off you may be on average. Tightens every level.

The app holds a secret thing. You ask yes-or-no questions out loud, the passenger taps the one you asked from the list, and the app answers -- then says what share of the remaining candidates your question ruled out and what the best question available would have ruled out. Name the thing inside the question budget to clear the round.

Players: Two or more people in the car.

Driver: Asks the questions. The phone answers, so nobody argues.

The bar: The questions its own best-split strategy would need on the things in play.

The passenger reads out an event. You say where it belongs among the events already on the timeline -- after the moon landing, before Chernobyl -- and they tap that gap. Right, and it joins the timeline, which makes the next one harder because there are more places it could go. Wrong, and the round is over. Par is a streak long enough that luck cannot account for it, and the app shows the odds it used.

Players: Two or more people in the car.

Driver: Argues about dates.

The bar: A streak long enough that luck cannot account for it.

A board of piles is read out to you. You name a move out loud -- how many to take, and from which pile -- and the app answers with its own. It plays perfectly, and every board it deals you is one you can win, so the only way to lose is to hand a lost position back to it. Three games rotate and the rule for taking changes with them, which is what stops the strategy from being solved once and coasted on.

Players: Two or more people in the car.

Driver: Speaks every move. The phone is only the board.

The bar: Perfect play. Every board starts won, so clear them all without a losing move.

Four people and a couple of things to pin on each of them. The passenger reads the setup, then the clues one at a time; you hold the grid in your head and say the answer when you have it. The clue set is minimal -- drop any one of them and there are two possible answers -- so par is hearing each clue once. Asking for a re-read is the only thing that costs, and that is the whole game.

Players: Two or more people in the car.

Driver: Solves it in their head. Asking for a clue again is what costs.

The bar: One reading per clue. Every clue matters, so there are no spare ones.

The passenger reads out a short account of a day: stops, times, who was in the vehicle, what got spent. Exactly one sentence in it cannot be true, and you say which. The app wrote a true account, broke a single claim in it, and then checked that only that one sentence is to blame, so an argument about it is not possible. Par is a clock: one reading aloud, plus a re-read of the stretch holding the fault.

Players: Two or more people in the car.

Driver: Listens, then names the line that does not add up.

The bar: One reading aloud, plus a re-read of the stretch holding the fault.

The passenger reads a category out -- the fifty state capitals, the seven wonders. You say how many of them you can name BEFORE naming any. Then you deliver: fall one short of your own bid and the round is worth nothing at all, however many you actually got. Three wrong answers end it. What is being tested is knowing what you know, so a cautious bid you make beats an ambitious one you miss.

Players: Two or more people in the car.

Driver: Sets the bid, then has to deliver it.

The bar: A bid worth making, from the list's length and how hard it is.

The app names a place. You name one that starts with the letter the last one ended on, and the chain runs on from there. Anywhere on the map counts. Scoring runs inversely to how well known the place is -- Tokyo is worth almost nothing, somewhere nobody at the table can place is worth a great deal -- so par is a points target the easy answers cannot reach.

Players: Two or more people in the car.

Driver: Supplies the places from memory.

The bar: A score set by how obscure the places are, read off real populations.

The referee reads a prompt out -- somewhere in Peru, one of the seven dwarfs -- and counts the players down. On one, they both say a single answer at the same moment. They score only if they said the same thing, and then only in proportion to how unlikely it was: agreeing on Tokyo is worth nothing at all, agreeing on a town of nine thousand is worth a great deal. Five prompts to a round.

Players: Three or more people in the car.

Driver: Says one thing on the count of three, and tries to be read.

The bar: A total no run of obvious matches can reach, read off real populations.

The referee reads a category, and the players bid each other up on how many of it they can name -- before anybody names one. When everyone else passes, the last raiser has to deliver what they claimed. Falling short pays whoever they were bidding against, which is exactly why letting a rival have the last word is a move rather than a surrender.

Players: Three or more people in the car.

Driver: Bids out loud, then delivers or pays the other one.

The bar: The bid worth making, from the list's own length and difficulty.

One player places the event on the timeline. Before it locks, another may challenge and put it somewhere else instead. A correct challenge steals the point and keeps the streak alive; a failed one pays the player who was challenged, which is what stops challenging being free. The referee never sees the year until the placement resolves, so there is nothing to read off their face.

Players: Three or more people in the car.

Driver: Places events, and challenges the placements they think are wrong.

The bar: A streak long enough that luck cannot account for it.

The players question the same secret, turn about, and either may name it at any point. The first correct name takes the round; a wrong one costs that player their next turn. The question budget is SHARED, so rushing spends the other player's questions as well as your own -- and if it is found inside par, everybody scores.

Players: Three or more people in the car.

Driver: Asks, listens, and names it. A wrong name costs a turn.

The bar: The questions our own best-split strategy needs, spent between the two of you.

The referee reads a category and the players take turns naming things that belong to it. Stall, repeat something already said, or name something that is not on the list, and you are out. With more than two playing it runs until one is left. There is no allowance and nothing to bid: this is Bid and Deliver with the safety net taken away.

Players: Three or more people in the car.

Driver: Keeps naming until somebody runs dry.

The bar: How deep the table should get on this list, from its length and difficulty.

Both players estimate the same quantity out loud, and the referee types both numbers in before either of them hears how close anybody was. The closer estimate takes the point, on the same how-many-times-off scoring Order of Magnitude uses. The table still has an accuracy to hit across the five quantities though, and that number does not care who won: a round you were both wildly out on is lost by both of you.

Players: Three or more people in the car.

Driver: Reasons out loud to a number, then hears how close it was.

The bar: The log-error threshold, tightened for having more than one go at it.