Dicey Dungeons Reunion Design: The Witch

Time for another Dicey Dungeons Reunion design thread! Up next, it’s everyone’s favourite contestant, the Witch:

Very early in development, I had a plan for a Witch episode that I assumed was going to be a slam dunk – so, I put that idea to one side and worked on some other episodes first.

That idea… did not work out. There was something there, maybe, but after a few rounds of iteration, I just couldn’t really get it to click. So, I decided to just scrap it and start over! So the Witch episode ended up being the last one I made for the DLC.

The thing about the Witch is; her episodes are a massive difficulty spike.

Lots of people start out liking the character, but get frustrated with how difficult she ends up being. Lady Luck makes it clear that she’s her favourite, though! (it’s me, I’m Lady Luck)

This new Witch episode idea was an attempt to make an “easy” Witch episode. It had a simplified take on her spellbook mechanic, and I was hoping that it would help people see what’s good about her. Kind of like a primer to the original episodes?

But that was the wrong way to think about it! This episode only really came together when I decided, screw it, I’m just going to make something completely self-indulgent.

So, I made another puzzle episode.

One of my favourite episodes in the whole game (pre-reunion) was the Witch Halloween Special episode – a series of handcrafted puzzles using the spellbook mechanic.

Initially, I *didn’t* want to make another puzzle episode, because I assumed it would just be more of the same!

Halloween puzzles were *real* hard to come up with. It’s the spellbook that makes it so complicated – so many possible combinations of things to try.

I had a go at coming up with more puzzles in this style for Reunion, but that didn’t seem promising. I wanted *easier* puzzles!

Sadly, I don’t seem to have kept a screenshot of this in game, but here’s the readme from this episode from when I tested it on the dicecord:

“how do I take the dice I’ve gotten, and turn them into the dice that I need?”

This episode ditches the spellbook, and focuses on pure dice manipulation puzzles. Here’s an example starter puzzle:

Each puzzle gives you a board of equipment, a couple of dice, and challenges you to turn them all into the same number. My first attempt just had a normal floor layout, like the other episodes. Eventually, though, I realised that it was much better to have it be one big fight!

Shout out to the Christmas Special Mod (matty707.itch.io/christmas-spec…), which is what inspired that choice – it did more or less the same thing for its Robot episode!

Coming up with the puzzles was a fun process! There are about 40 puzzles that can randomly come up. How did I come up with that many? I wrote a tool to help!

My tool would take a board and dice arrangement that I gave it, and analyse it for interesting solutions. When I spotted one that looked promising, I’d try to solve it, and see if it was interesting.

Most of the time, it would turn out that solutions “looped” – that is, there were ways to solve a puzzle unintentionally by just cycling dice around in circles. These sort of solutions were rubbish, and it would have been impossible to spot them without making this kind of tool.

A loop is something like “put this 3 in a Bump, and then put the 4 in a Nudge” – now you’ve got a 3 and you’ve used both pieces of equipment.

In that example above, [3, 3] is a possible solution to the puzzle, and there are 1,462 combinations that lead to it. *But at least one* of those combinations contained a loop, which meant it couldn’t be considered.

Interestingly, the first version of this tool wasn’t searching for solutions with matching dice – it was looking for any combination at all which might be interesting. It just turned out in practice that all the “matching dice” puzzles were the best ones.

I’ve never *really* worked this way before, where I’d written a tool to help me with design, and I’ve gotta say, the process was really fun. Like searching for signal in static.

The tool had its limits, though. I did the “constraint” puzzles manually. This one is probably my favourite:

Something I really love about this episodes is that it doesn’t demand absolute perfection – you can make a few mistakes, and it’ll be ok! Here’s a puzzle that has an interesting solution for both 3 1s and 4 1s:

The best moment I had developing this episode was when my tool insisted there was a solution to a certain puzzle, but the graph exporter wasn’t working and I couldn’t figure out how to solve it!

I thought the tool was just *wrong*, but I exported the raw graph data and combed through it myself – and discovered a solution that I’d missed. You better believe this puzzle made it in!

I’m getting into the weeds here a bit, I know, but there’s one more thing that was really interesting to figure out: the final puzzle.

Something that happened a bunch in testing is that people would get to the final puzzle, and realise they could solve it without getting a “perfect” solution.

If a puzzle required 4 1s to solve, for example, you could sometimes get that before you’d used all equipment.

In those cases, obviously, people would just use the 4 1s as soon as they had them, and win the episode prematurely. Which felt… underwhelming.

But what can you do about that? With this kind of dice manipulation, you’re almost always going to end up in that situation, and enforcing a perfect solution on the final puzzle just felt bad.

The answer: I got my tool to search each of my existing puzzles, looking for early solutions. Virtually every puzzle had a bunch of them – but exactly two had no early solutions at all.

I categorised these as “perfect” puzzles, and made sure that one of them would always appear as the final one! It’s maybe a dumb detail, but I felt extremely proud of myself, lol

…unfortunately, I discovered after release that I’d gotten this wrong, and that a perfect puzzle wasn’t actually guaranteed at the end. OH WELL

That about wraps it up for the Witch! Two more episodes left, Robot and Inventor, and thankfully both are a little easier to talk about.