Rule Hunt Strategy: Finding the Hidden Rule in Fewer Tests

Updated October 2026

Rule Hunt gives you three minutes to uncover as many hidden rules as possible, and the only feedback you ever get is a card landing in the accept column or the reject column. Because the set of possible rules is small and completely known, this is a game you can play almost perfectly with the right method. This guide lists every rule the machine can choose, explains how the starting cards are picked, and shows how few tests a well-planned hunt really needs.

The cards and every possible rule

Each card has a colour (red, green or blue), a shape (circle, square or triangle) and a count of one, two or three symbols. That makes 27 different cards. A hidden rule is always one of four forms, and knowing how many cards each form accepts is the most useful fact in the game:

• Single value, such as "blue" or "two". There are 9 of these, and each accepts 9 of the 27 cards.
• Negation, such as "not a circle". Again 9 rules, but each accepts 18 cards, two thirds of the deck.
• And, such as "green and triangle". There are 27, and each accepts only 3 cards.
• Or, such as "red or three". There are 27, each accepting 15 cards.

An and or an or rule always joins two different properties: a colour with a shape, a colour with a count, or a shape with a count. Something like "red or blue" never appears; you would see it as "not green" instead. Altogether there are 72 distinct rules.

Which rules can appear when

The machine widens its choice as you succeed. Your first two rules are always single values. The third and fourth can also be negations. The fifth and sixth add and rules, and from the seventh rule onward all 72 are possible. The step only moves up when you solve a rule; a wrong answer gives you a fresh rule from the same set. So a strong start matters: the first few rules are the cheapest points you will get all run.

How the starting cards are chosen

Every puzzle opens with two accepted and two rejected cards, picked on purpose to leave several rules still possible. In our simulation of thousands of puzzles, the four starting cards never identified the rule on their own. At the first two rules they always left exactly two candidates; at rules three and four, usually four; at rules five and six, usually five; and from rule seven, five or six about two thirds of the time, with the rest ranging up to eleven.

This means you should expect to test at least once on every puzzle. It also gives you a quick sanity check: on the first two rules, your job is just to tell two single values apart, and one well-chosen card always does it.

Reading the starting evidence

Before you build anything, look for what the two accepted cards share and what the two rejected cards share.

• Accepted cards with nothing in common rule out single values and and rules immediately, because both of those require a shared property. You are looking at a negation or an or rule.
• Accepted cards that share two properties make an and rule worth checking: those two shared values are the only and rule both could satisfy.
• A rejected card that shares a value with both accepted cards eliminates that value as a single-value rule on the spot.

Write out the short list of candidates in your head. Up to the sixth rule it never had more than five entries in our simulation, and even later it is usually under ten. The game's own design guarantees that the list is all you need to consider.

Choosing the test card

The ideal test is a card that half of your remaining candidates accept and half reject. Whatever the answer, you throw away about half the list. Build it on purpose with the three rows of buttons rather than by habit.

For example, suppose the candidates are "blue", "square" and "blue or square". A blue circle splits them neatly: if it is rejected, only "square" survives; if it is accepted, you are down to "blue" or "blue or square", and a red square settles that. Two tests, every time.

In our simulation, choosing the best splitting card each time needed exactly one test on the first two rules, about two on rules three and four, about 2.4 on rules five and six, and about 2.7 from rule seven, never more than four. If you are routinely using six or seven tests, you are choosing cards that do not split your list.

Note that testing a card already shown in either column does nothing at all. It is not counted and gives no new answer, so check the columns before pressing Test.

The final check and when to commit

Pressing I know it switches you to the check stage, and you cannot go back to testing. Four new cards appear, and you mark each one as following the rule or not. All four correct scores the rule; any mistake reveals the rule and starts a new puzzle. A correct answer brings the next puzzle after about a second, a wrong one after about two.

The four check cards are picked to make the remaining candidates answer differently. The game only deals starting evidence for which such a set of four cards exists, and since every test you add can only shorten the list, a separating set exists at every point of the puzzle. In our simulation of 5,000 puzzles per level band, with zero to six extra tests before confirming, the check cards separated every candidate in every puzzle. So once your list is down to one rule, you cannot fail the check; with two or more, you are guessing.

Guessing is tempting on the first two rules, where the odds are one in two, but a single test always settles it for a few seconds of effort. From rule seven onward a blind guess between the remaining candidates succeeded only about one time in six in our simulation. Test until one candidate is left.

Pacing three minutes

The clock starts when the page loads and runs on real time, so it keeps going if you switch apps. Know which buttons you want before you arrive. A rough budget is ten to fifteen seconds for each of the early rules and twenty to thirty for the later ones. Spend the saved time on the and and or rules, which are where careless commits happen. The rule-hunting habit that pays most is simple: list the candidates, split them in half, and commit only when one remains.

Your best score is saved on your device, and any run with at least one rule found can go to the Hall of Fame.