/NerdleSolver

Interactive Nerdle Solver

Primary LanguageJavaScript

Nerdle Solver

The aim of this project is a fairly silly interactive html/javascript solver for https://nerdlegame.com/classic/

To try the automatic solver, you can run it from from your web browser here:

The idea is that the Nerdle Solver will generate a guess, you enter it into Nerdle itself, manually assign the colours back into the Nerdle Solver, and it will iterate guesses from there.

It goes without saying, this is 100% unaffiliated with Nerdle. Currently it will only solve Nerdle Classic and Speed out of the box. There's hypothetical support for Mini and Instant, but these have to be enabled by hand editing the nerdleglobals.js currently. I'll either let you choose the game on startup, or maybe just cheat and have extra HTML files to set this up.

This just leaves Nerdle Pro, which is only supported if the operators match Nerdle Classic, and the expression matches Classic or Mini. Support for the extra operators may be added in the future. However the quality of the first guess is critical, since the more unique values that are discovered, the more information we are guarenteed to get on the first guess, I'll probably re-visit the auto-guess generation, especially on the first guess.

Usage

Using the Webpage is simply a matter of performing these repeating steps...

  1. Enter your guess into the Solver. This can either be your own guess, or the suggestion.
  • If you want to use the Suggestion, just click on Use Suggestion.
  • Otherwise, type in your guess and Press Enter.
  1. Enter that same guess into Nerdle.
  2. Click on each of the Numbers/Operators in the solver until the colours match Nerdle. Press Enter when done.
  • Super duper double check it matches.
  1. Go back to 1

The Gold Suggestion at the bottom is there to let you know what the current suggestion is.

About

So... how does this work?

Firstly, this needs to implement Nerdle itself, so I've essentially reverse engineered the parser by just throwing various inputs at it. So here's some rules.

  • Unary operators are allowed, and they stack, albeit I highly doubt they'll ever be used in any of the classic puzzles... infact the FAQ states that negative values won't ever be used in solutions.

    • This means the following are 100% legal.

      --1+-1=0 (this reduces to: 1 - 1 = 0)
+-+-1=+1 (this reduces to: 1     = 1)

  • Leading zeroes are allowed, and there's no limit to them, albeit again, the FAQ states these won't form part of a real solution.

    • This means the following are 100% legal.

      01*001=1 (this reduces to: 1 * 1 = 1)

  • Otherwise standard Wordle rules apply.

    • Green on the guess means a number/operator is in the right position.
    • Green on the keyboard means at least one instance of the number/operator has been correctly deduced in the right position.
    • Purple on the guess means a number/operator is needed, but the position is wrong.
    • Purple on the keyboard means at least one instance of the number/operator is needed, but the exact position is unknown.
    • Black on the guess means this instance of the number/operator isn't needed.
      • If it's the only time this number/operator has been guessed in this round, then it isn't used anywhere.
      • If it's not the only time this number/operator has been guessed in this round, then it is needed, but this instance indicates we have too many.
    • Black on the keyboard means the number/operator is not used anywhere in the solution.

So as indicated, black on guesses is where things get interesting, as what it tells us is how many instances of a number/operator are needed.

Secondly, the part you're here for, the solver...

Initial Game State

Initially we know a few things, there's 8 entries, but we can rule reduce some already.

Col : What we know...
=== : ===================================
  1 : Can only be one of 0123456789+-
  2 : Can only be one of 0123456789+-*/=
  3 : Can only be one of 0123456789+-*/=
  4 : Can only be one of 0123456789+-*/=
  5 : Can only be one of 0123456789+-*/=
  6 : Can only be one of 0123456789+-*/=
  7 : Can only be one of 0123456789+-*/=
  8 : Can only be one of 0123456789

We also know some things about the legal inputs:

Input : What we know...
===== : ===================================
    0 : Can occur between 0 and 7 times
    1 : Can occur between 0 and 7 times
    2 : Can occur between 0 and 7 times
    3 : Can occur between 0 and 7 times
    4 : Can occur between 0 and 7 times
    5 : Can occur between 0 and 7 times
    6 : Can occur between 0 and 7 times
    7 : Can occur between 0 and 7 times
    8 : Can occur between 0 and 7 times
    9 : Can occur between 0 and 7 times
    + : Can occur between 0 and 5 times
    - : Can occur between 0 and 5 times
    * : Can occur between 0 and 3 times
    / : Can occur between 0 and 3 times
    = : Must occur only once

The interesting part that helps define this is the very last row. Whilst we can position = in up to 6 positions, we know we can only have one of them. We can also condense the operators further. We know * and / must be nested between numbers, meaning for multiply/divide, the most you'll see is based on:

0*0*0=00

Whereas because +/- can also be unary operators, it's 5 times, with the following being the longest legal expression:

-----0=0

So any solutions proposed will be limited by the above knowledge. We know what items can go in what positions, and we know a min/max for each. The purpose of the solver therefore is once a solution is entered, use the colours returned to refine what we know about they game state, which will further reduce the potential solutions.

So each time a black entry is returned, we can refine two things:

  • We can refine the maximum number of times an input can occur.
  • We can refine the legal positions for the inputs since we know a position it can't go.
  • And obviously if the maximum number of times is zero (i.e. input only tried once, and black was returned), we can remove it entirely from consideration.

Each time a purple entry is returned, we can refine two things:

  • We can refine the minimum number if tunes an input can occur.
  • We can refine the legal positions for the inputs since we know a position it can't go.

Each time a green entry is returned, we can again refine two things:

  • We can refine the minimum number if tunes an input can occur.
  • We can refine the legal inputs for a given position since we know what it must be.

And after processing these...

  • If the total of all the minimums for all possible entries exactly matches the number of spaces digits can go, then we know exactly how many of each type of number we can actually use. Meaning we now just have to set all the max's to the min values.

The idea is we want to end up with a scenario where:

  • All the min/max's are the same, meaning we know exactly how many times each input is used.
  • All the positions have exactly one entry, meaning we know exactly what input is used in what position.

The closer we get to achieving this, the less possible permutations there are that satisfy a balanced equation.

In terms of the balanced equation, we need to find an equation that meets the known criteria:

  • All known positions are filled with the input.
  • All unknown positions are filled with the possible inputs.
  • The total of each of the inputs that are used/not used all satisfy the known ranges for each input.
  • If possible, avoid solutions which abuse stacking unary +/- and leading zeroes, these will only be used if all other possibilities have been eliminated.
  • If possible, avoid solutions which abuse multiplying by zero, these will only be used if all other possibility have been eliminated.

Currently solutions are offered in three passes.

  • Pass 1: Satisfying answers.
    • This assumes that the answer won't include NOP arithmetic, such as 10*0-0=0 (basically it will disallow any 0's), and that answers won't abuse unary operators or leading zeroes.
  • Pass 2: OK answers
    • This allows NOP arithmatic, but doesn't allow answers to abuse unary operators/leading zeroes.
  • Pass 3: Naff Answers
    • If the parser accepts it, this will propose it.

One interesting thing I found was the Nerdle tests the associativity of a solution, i.e. 1+2=3 and 2+1=3, as long as you enable "Allow commutative answers" in the settings of Nerdle (default), are interchangeable. This isn't something I believe my solver needs to be aware of due to the following:

  1. If my solver proposes an answer that's commutative with the real answer, Nerdle will just accept it, up until then Nerdle will NOT re-arrange the expression.
  2. If my solver proposes an answer that's commutative with the real answer and Nerdle doesn't accept it (i.e. it's disabled), my solver will eventually propose the real answer anyway.

Compatability

This goes back at least as far as Internet Explorer 11 (what I'm forced to use at work). Going back further would force me to give up const, which doesn't feel great, but if there's demand, I'm sure I can start regressing the code to support older standards........... maybe. Though a quick look implies:

  • Internet Explorer 10 - Change all const's to var's (yuck).
  • Internet Explorer 9 - DOMTokenList no longer exists, meaning I'd probably need to re-write all the code for setting/removing class's for the CSS style.

It's all doable, but again, given I've ostensibly designed this for a mobile, I'm a bit loathed to start making changes which overly comprimise the code. IE 11 was at least just a few simple wrappers / default parameter handling.

TO DO

  • Pro support
    • This will require reworking of the basic initialisation routine, as currently it's hard coded to three states.
      • Nerdle Classic - Fixed to 8 columns.
      • Nerdle Mini - Fixed to 6 columns.
      • Nerdle Instant - Identical to Nerdle Classic, but the first entry bypasses the entry validation rules pertaining to ensuring the equation is valid.
    • The first guess is now auto-generated, with rules in place to try and maximize the spread of digits.
      • In all honesty, I was surprised DFS worked at all here. I may re-visit this to see if there's a better / more browser friendly way of finding a solution. Especially now it's way faster, I can try and improve the quality of the first solution to encourage more unique digits.
    • Interestingly, Nerdle Pro seems to have a width limit of 16, I'm not going to implement that limit, albeit depending o how well I get the expression generator working, there may be natural limits to deal with anyway.
    • I do still dislike the fact the auto-generated solution invariably starts with a negative number due to how I bias the results. The reason I dislike this is because Nerdle explicitly states in the FAQ that their answers won't use negative values, meaning the first minus can never be correct.
  • Another UX pass... maybe some sort of interactive tutorial, or some better context clues as to how to use the solver.
    • Plus the UI is kind of misleading for Instant Mode, since there's no reason to offer a first suggestion.
  • Clearly some sort of heuristics needs to be added as to what the fundamental intent of the suggestion is. Currently one of the Nerdle Games got the solver into a horrible situation, whereby the answer was: 27+20=47
    • The Solver quickly got to (on move 2):
      • 2X+20=4X
    • But: 0, 2, 6, 7, 8 and 9 were all viable solutions, and could only be eliminated one at a time playing the naive strategy of going straight for the answer. A better approach would be to construct an algorithm with the data we know the least about here, being 6, 7, 8 and 9, since we know each of these can only appear 2 or no times to satisfy the equation. If they all go black (no times), then the solution is 50/50 between 0 and 2, meaning we solve in a worst case of 5 moves. If one of them goes green/purple, then we know the answer and solve in 4.
      • Whereas for the naive approach, you have a 1/3 chance of failing, as you can only try 4 of the 6 possible values.
    • I should at some stage do some sort of in-depth testing on generated solutions, and how well the Solver copes in order to evaluate changes. For example, I could just push a "fix" to the solution generator that prevents negative values being allowed in "good" solutions. This ends up changing the nature of the guesses so much this issue presented here quietly disappears... or more likely, simply shifts onto a different unsolveable solution.
      • Whereas with proper exhaustive testing (proably about the same time I update this antique hardware), I can evaluate how well the Solver copes with "good" puzzles, whether it ever fails, min/max/average attempts before solved and so on.