Puzzle Duel

[pinkagape]

Thought this might be a good place to discuss Puzzle Duel and to congratulate AJ on winning August’s Yellow division.

[pinkagape]

I thought I'd share some feelings after having the best time on 12 December.

In August, I was doing Puzzle Duel and had a snake puzzle I couldn't see a start on. So I guessed, the solution fell out and I got the second best time. I felt pleased but also knew that this was all down to a guess.



Then last week I was doing No Four In A Row and took a guess early on which paid off as I hit a solution. I was a bit nervous that there might be a four that I hadn't spotted but it looked fine so I submitted and ended up with a great time that stayed #1 all the way through until closing. I feel much happier with this one, possibly due to the risk/reward of having an error, but also because I feel like knowing where to guess is often the key to a No Four In A Row.



All in all, I'm very pleased with myself and wanted to share my joy, but also see if anyone else felt like a fast solve caused by a guess was worthy of credit or not.

[detuned]

Hey Adam, definitely worth a post!

For what it’s worth I think there are now a new generation of solvers who are far more intuitive than before - and I suppose you’d have to say it’s not exactly random guesswork if the same people keep doing so well using it (which isn’t to say these people are always intuiting either). Very interesting to hear about the examples where you can have some hope of getting an edge over the likes of Freddie and co!

Back in the day there’s be one or two types where I could top the charts, but I don’t think I’d be able to pull it off these days. A lot of it is the interface… I think some are better on phone whilst others are better with keyboard/mouse - however in most cases I think a tablet/smart pen is simply a lot quicker these days and I don’t have that setup.

I still would rather not advertise my puzzle duel username, but because of the above and the league system I’ve been attempting to prioritise accuracy rather than speed these days (I’m not sure staying up to do them at midnight is all that sensible either…). Over the summer months it was working out quite nicely but I’ve stuffed up a couple in December and will have to start again in January

[pinkagape]

Just a quick message to congratulate David on his 3rd place this month!

[pinkagape]

And again, to celebrate Maurice’s third place this month. UK in a golden patch recently.

[detuned]

Well done both!

There are a fair few good solvers piled up in orange league right now - just goes to show how difficult it is to maintain a 100% record, and that you’ll be on a good thing if you can keep it up. My own solving times have come right down - I’m typically spending at least 30-60 secs on each puzzle simply checking.

[pinkagape]

I'm wondering about doing a league next week made up of solvers who are using the site every day, after this messy Puzzle League version from this week with people not playing.

I'm thinking:

Me, Maurice, Matthew, AJ, Tom, David, Todd and Desirée

It wouldn't require any additional thought as I believe we're all solving every day as it is but might be a nice contest.

1) Does anyone not want to be included?
2) Does anyone else want to be included? It doesn't extend well beyond 8 and stay in a week but if we had more we could maybe look at two smaller leagues?

[detuned]

Seeing as I do the puzzles anyway (although sometimes bunched up after a few days, rather than on regular schedule), I don’t mind being a part of this suggestion

[pinkagape]

Here are the fixtures that I'll be working to this week then!

[pinkagape]

Lightning time from Maurice on Day 1, beating Freddie's time. Here are the results so far.

[pinkagape]

As the month ends and everyone clears their backlog, we had a tricky Clouds puzzle which has introduced some large time gaps. Desirée will be pleased to count a win as I had a nightmare, and there's only two players left with a perfect record. It could be tight.

[detuned]

After we had 3rd in the red league with David (in Feb) and Maurice in March, I’m quite pleased to claim 2nd in April - though a loooong way behind Hideaki Jo in 1st which is hardly a surprise.

Here are a few observations:

1) I haven't been going out of my way to solve quickly - and in particular I’ve kept up my approach through 2024 which has seen no errors and a progression through yellow and orange, to a 4th place in March and now a 2nd place in April. Maybe there’s something in the interface but it seems remarkable how frequently people submit with an error. Conversely the puzzle duel structure really rewards not making a mistake.

2) there’s more thinking to be done on puzzle ranking systems here, with the contrast of where you finish in the daily rankings (which is simply an extension of a choice of a given 1v1 match up within the ranking) vs what your aggregate time implies. It does seem on any one given puzzle I’m not doing anything spectacular (as evidenced by the recent contests as organised by Adam)

3) I’m not particularly keen to share my puzzle duel username, but there’s now more than enough evidence on these forums to work out that particular puzzle.

4) getting into the habit of sacrificing time on daily puzzles does seem to get me into good habits - I do seem to be making less silly mistakes in the offline competitions than I used to

[pinkagape]

David continues his strong run and gives Matthew his first defeat of the week. His tough start continues tomorrow with a match against AJ who is also going well.

[pinkagape]

Matthew continues his strong week with a great time to build his time advantage should David drop a match.

[pinkagape]

A tricky puzzle today could give the table a bit of a twist with three solvers yet to try it. Maurice took a big time hit, but until David or Matthew get their fifth win on the board, it's still an open game.

[pinkagape]

Argh! Spent too long checking today!

[pinkagape]

I'm really enjoying this league. I hope you are too. Todd's had a great couple of days and Matthew's had the luck here that he didn't get in the Division A week. Here are the remaining fixtures and current standings.

[pinkagape]

Matthew is all done now and just waiting on David as the only one able to stop him, but with a sudoku as Sunday's puzzle, it's hard to see past David.

I'll keep an eye out over the next couple of days and post when I spot that we're all done.

[pinkagape]

Not quite done, but here's a little update while we await the final result.

[detuned]

Turns out this is a really interesting source of data to play around with!

My first effort was to compare these results by looking at the sum of ranks by day, with lower ranks being better. This in general doesn't feel like a great way of doing things as there's much dependency on the total participation in each event - and also you have to wait for everyone to participate to be able to get final results. In this case the daily participation didn't vary, and I've taken the convention that Todd is currently ranked 8th on the puzzle he hasn't completed, and the results go something like:

1 Tom (24)
2 Matthew (26)
3= David (28)
3= Maurice (28)
5 Adam (32)
6 AJ (34)
7 Todd (35)
8 Desiree (44)

However, one thing that struck me is that any puzzle event that produces a ranking naturally provides a set of 1v1 results, where 1st place beats everyone else, 2nd place beats 3rd down and below, and so on. This means you can track pairwise head-to-head matchups across a series of different events. In this case, a series of 7 individual puzzles.

I thought it would be fun to run the data on these 7 puzzles, and look at different metrics - and see what kind of different orderings I could produce. For example, in the Adam vs AJ head to head over the week, the result was 4-2 in Adam's favour (with a tie on Friday!). In this case we award 2 points to Adam for the pairwise victory - and where we have a 3-3 that would be 1 point each. Compiling a results table that way gives something like:

1 Matthew (12)
2 David (11)
3 Tom (10)
4 Maurice (9)
5 Todd (7)
6 Adam (4)
7 AJ (3)
8 Desiree (0)

Interesting that that produces a bit of a shake-up compared to Adam's work - I guess Adam's is very dependent on the choice of puzzle used for the head-to-head, which adds some inherent volatility to proceedings. Another way of interpreting "volatility" is a bit of excitement, where a good performance on a single puzzle can be well-rewarded. Using head-to-heads means the importance of any one single puzzle becomes less important.

Still, if this is the kind of thing to generalise a bit further (you could do something like this to produce 24hpc rankings for example), it strikes me that having a wider definition of draws might be helpful. So I played around and decided that if the head-to-head had a difference of 1 (so e.g. any of 4-3, 3-3 or 3-4) then I would call that a draw. In a more general setting, where not every player plays in every event, this feels like a good thing to do because it means you'd need at least 2 head-to-heads before you determine who takes maximum points.

For Adam's event, doing that shakes up the rankings even further!

1 Tom (12)
2 Matthew (9)
3= David (8)
3= Maurice (8)
5= Adam (7)
5= Todd (7)
7 AJ (5)
8 Desiree (0)

Interestingly here is the observation the ranking is very similar - but not identical - to the ranking produced by just summing everyone's daily ranks. On t'other hand, where you are forcing there to be more draws, it probably makes sense to learn the lesson from football and give greater rewards to wins (in this case where the difference in the record is at least 2 - so 4-2 or 5-2) - so we award 3 points instead of 2:

1 Tom (17)
2 Matthew (12)
3 Maurice (10)
4= David (9)
4= Adam (9)
4= Todd (9)
7 AJ (6)
8 Desiree (0)

There's a bit more bunching up there admittedly.

Final thoughts:

(1) I quite like the idea of doing rankings via head-to-heads, as it appeals to the old football cliche that the league table never lies. Quite what head-to-heads mean in a puzzling context is open to a bit of interpretation. It could be a single event, could be an all-time record, or it could be a selection of recent events.

(2) I also like the idea of introducing draws as a way of maintaining a semblance of stability to the rankings. OK, from a competition point of view it probably makes things a bit boring, but I reckon for a rating system having rankings that don't change radically on one result is probably a desirable property. I suppose you could try and introduce draws via a different method - rather than using rankings as the head-to-head comparison, you might look at solving times and/or points totals as a way of deciding whether two performances on a single event could be said to be "close". I'm less keen on that approach, as there's a certain dependency on evaluating performances by using wider contexts as benchmarks. E.g. a time difference of 5 seconds is probably close on a 10 minute puzzle, but is it really all that close on a 30 second puzzle? Maybe you could work things with % differences instead, but my spreadsheet isn't really set up that way.

(3) I'm tempted to try this out to produce some UK rankings where I consider the rankings of lots of different events. I'll probably experiment further by adding weightings to different events (stuff like UK Open and WSPCs would get a higher weight), and also considering how to balance all-time records vs. more recent records. I'll probably share via a google sheet if there's interest!

[pinkagape]

Very interesting Tom!

Here is the final table of this competition, using the basic head to head rules. Well done Matthew!

[pinkagape]

So - with the Puzzle Duel Snake on 9th September, I tried all 49 combination of the left and right brackets and couldn’t get any to work. I must have spent about 4 hours on various efforts. The comments and rating suggest that it was just more of the same, but does anyone remember having more of a start than that? I did (eventually) spot the clue that there were only 6 cells in the unclued rows.

[Mblount]

I also got through it by a lot of trial and error, though I don't have much intuition/understanding of how snake puzzles work. I don't know if there's much mileage in looking at the total cell count to get a feel for how wiggly a snake has to be, which might have worked here.

After I spotted the unclued rows towards the end, I used them to more quickly (I think...) rule out some options, as they mean the snake can't go back and forth too much across those rows - e.g if the bottom right corner is shaded, then the second-to-bottom row 4 clue forces the snake to go down and up through the middle unclued row, forcing at least 3 shaded cells there. Together with the 2 shaded cells in the bottom row, this leaves only one cell to cross the top unclued row which then forces the 5 in the top row which then breaks the puzzle because of the 6 clue in the column.

[pinkagape]

Well done David! That was an exciting month!

[david mcneill]

Thanks Adam. Slightly higher level of difficulty across the month. Got a bit lucky with that last puzzle.