Il nome della zebra

[Ben-Wormald]

My second time participating! The aim is to generate a solvable zebra puzzle with however many clues it takes to hit 50k words. I read The Name of the Rose this year and a monastery murder mystery seems like a good setting for the puzzle.

[ikarth]

Reminds me a bit of Calvino's "Prose and Anticombinatorics", or Georges Perec's approach to writing Life: A User's Manual.

I think there's a lot of potential in using a structure (such as a puzzle) for a generated novel.

[Ben-Wormald]

https://github.com/Ben-Wormald/nanogenmo-2023 Started with a proof-of-concept for generating clues, looks like it's working well with a small fixed number of items

[verachell]

A cool project. Can't wait to see how this turns out!

[mmoskowitz]

Very interested in this as another person who has done a puzzle-based NaNoGenMo.

[Ben-Wormald]

I've made the clue generator use a dynamic number of items/attributes - unfortunately the brute force approach breaks down at 5x5. I'm wondering if taking a small random sample of the solution space is viable, otherwise I might need to change this approach of accepting clues which narrow down the solutions

[Ben-Wormald]

No luck with the sampling approach so instead started again with a graph representation - build the full graph for a random solution then pick a random MST for the final set of clues. It's much better for handling large numbers!

There's redundant clues with this method - one optimisation to make is leave out one item from each group in the MST as that can be inferred, otherwise potentially attempting to remove clues and test if it's solvable.

But now it's probably time to start turning the set of clues into some actual prose!

[Ben-Wormald]

Here's my first complete iteration, 51,045 words which should be fully solvable: https://raw.githubusercontent.com/Ben-Wormald/nanogenmo-2023/9132d60a3b65233bbde160cedd631709f293edc9/output.txt

Some bits I still want to do before the end:

• Add some more attributes - currently I have a cell number, name, town of origin, and age for each monk. The England's Immigrants dataset was really helpful in getting a big list of timely, varied names and towns. Having just 4 attributes needed 500 different entities (there's some very old monks in this abbey!), so it would be nice to come up with some more attribute lists for a more interesting set of clues.
• Add some intro and outro text to introduce the setting and context. Also set up a clue to the culprit in such a way that you'd need to process all the clues to determine their exact identity.
• Refine the clue generation to reduce redundant clues and potentially introduce more types such as neighbours and negation, though this might be tricky with the graph/MST approach.

[mmoskowitz]

Oh this is very cool. For another attribute, can I suggest the care of a specific saint's relics, or even multiple relic types per saint? Edited: And I see you are way ahead of me. Never mind, and again, wow!