Divisions A and B 2020-2021

[pinkagape]

A few people with a puzzle in hand leaves it a bit unsettled, but a good catch up from Michael here.

[mwhite]

I should have completed through to today (which I think was the last one), sorry for getting a little behind. I've enjoyed the way you've set this round up Adam - I do obviously enjoy the 'sit and puzzle for 90 minutes' format for GP/LMI but its nice to have a round that's a bit different

[emma]

My 358 seems to be the worst score. I don't like the interface for sudoku, and in fact after I went wrong, I gave up and printed it and solved on paper and then copied in the answers - which took rather less time than the time I'd spent failing to solve it online!

[pinkagape]

There's worse in Division A, Emma. It's easy to have a bit of a nightmare here, I think. Friday's maxiloop really threw me through a loop!

Division A is all over bar the shouting, as it doesn't matter what David scores on today's MaxiLoop, the positions are fixed. The missing puzzles in Division B will have more of an effect, but we've seen a number of very long times so far, so it's not guaranteed to change!



Here is Division A after Round 3 is completed. The missing Division B puzzles need to be solved by Tuesday evening at the latest please.

[pinkagape]

I should have completed through to today (which I think was the last one), sorry for getting a little behind. I've enjoyed the way you've set this round up Adam - I do obviously enjoy the 'sit and puzzle for 90 minutes' format for GP/LMI but its nice to have a round that's a bit different

Thanks Matthew. That’s exactly what I was going for. Glad you enjoyed it.

[pinkagape]

I’ve just realised that if David beats Freddie’s time by enough he would increase everyone else’s point sufficiently that he could catch up, so it’s still worth a go!

[Puzzle_Maestro]

Yeah, David definitely has a chance to claim 4th/5th place. 3rd is still theoretically possible but he’d have to solve at an absolutely blazing pace!

[david mcneill]

Probably won't (try to) solve the Saturday puzzle until tomorrow. I had a few nightmares already this week. The Ripple Loop turned out to be very hard and I didn't give myself a long enough window to solve it. Had to return to it over an hour later. The Magic Snail was another disaster as I didn't know it was a 1-3 snail. I submitted a valid 1-2 snail! However, I quite liked trying out Puzzle Duel even if it's never going to showcase my undoubted talent.

[Nilz]

I should have completed through to today (which I think was the last one), sorry for getting a little behind. I've enjoyed the way you've set this round up Adam - I do obviously enjoy the 'sit and puzzle for 90 minutes' format for GP/LMI but its nice to have a round that's a bit different

Thanks Matthew. That’s exactly what I was going for. Glad you enjoyed it.

Yeah I agree with this too. I really dislike solving with a mouse (I have a mouse in hand so much already for work and/or pleasure that I really don't want more reasons to potentially give myself RSI), but it was a refreshing change to only have to set aside a small number of minutes at a time, rather than 90-120 mins.
I think the scoring system would work better if all puzzles were of the same difficulty. Is taking 90 seconds to solve a puzzle that some people can solve in 30 seconds really that much worse than taking 5 mins to solve a puzzle that some people can solve in 2 mins? Not sure. And there's obviously no real solution to this, afaik we have no control over which puzzles came up this week- it was just pot luck, which at least had the advantage of being quite fair for everyone.
Actually, just thinking aloud, I wonder if setting the 4th or 5th time to be the 100-mark would work better- it would lessen the impact of uber-quick times. Anyway, if we're going to use the same system again in future, we've got time to play around with the numbers, this worked well enough for a first go for sure.

I'd also be curious to see a combined A & B table if you've got the data for that Adam? It's not going to simply be the exact same scores just meshed together because the 100-point mark is different in each division.

[pinkagape]

Thanks everyone for trying this new format. I think I agree with Neil, especially when you see the combined table, that using second place as the benchmark probably isn't right for future years but it's given a decent result this year.



And here's the combined one as requested by Neil.



This leaves us with a current Division B standings like this:

[pinkagape]

Round 4 will be the next Puzzle Ramayan in two weekends' time.

[Nilz]

Crikey, that combined table is interesting. If Freddie and Sam had been in the same division, their combined impact has a huge impact on Friday's scores (enough to move Adam above Tom and Matthew, for example- incredibly, by just a single point!). I think that definitely argues for a slight alteration of the system ahead of next time (assuming there is a next time). But to reiterate, the system was perfectly fine for this round, I'm trying to help constructively going forward, not grumble about the past.

[detuned]

One of the interesting things about the puzzle duel site is that it gives a fairly detailed personal rating history. I mention that as last week was objectively the worst I had for 3 months, and so also don’t want to be seen as grumbling about the past

I started blogging about this earlier in the year. To get a good idea of measuring performance, you need at least some idea of a benchmark to measure it against. Most commonly the benchmark is set vs. The population at that point in time, typically the median time - but I guess taken 2nd place in the sub populations here is a bit volatile.

Another alternative is to have a consistent panel of testers to help set the benchmark, similar to how you have testers for a contest to assign points.

I’m not really sure why a linear scale is any better (or worse) than any other, but for example the puzzle duel rates performances (for the purposes of the ratings) vs. the median solve time via the formula:

Points = 2000 - 500 * log_2 ( your time/ median time + error counts)

The old croco formula was slightly more complex and introduced the second parameter of the best time:

Points = 3000 / 2 ^ [(your time - best time)/(median time - best time) + error counts]

I have previously observed that taking the natural log of raw solve times roughly produces a normal distribution, in which case you could also take the standard z score for log(your time) vs the days distribution.

If the puzzles were all of a similar difficultly, you could just add up the times untransformed! You could still do that anyway if you are less bothered by puzzles having to be equally weighted individual events.

[pinkagape]

Logic Masters India have released their Instruction Booklet for this weekend’s Puzzle Ramayan, which doubles up as Round 4 of Divisions A and B of the UK Puzzle League.

The puzzles are Fillomino, Double Choco, Sashigane, Spiral Galaxies, Compass, Combi-Block, Max Spiral Galaxies and Max Sashigane

[pinkagape]

Round 4 is complete. Some excellent work by the UK in general resulted in a familiar looking top of both leaderboards, but Division B also had their fourth different second place finished.

Division A:

Freddie Hand - 132.07 points
Neil Zussman - 117.63 points
Adam Bissett - 55.4 points
Matthew White - 55 points

Division B:

Sam Cappleman-Lynes - 125.37 points
Michael Collins - 67 points
Emma McCaughan - 66.7 points
AJ Moore - 20 points
Roderick Grafton - 4 points

This leaves the league tables looking like this, ahead of May's Grand Prix as Round 5. Division B is pretty close on the whole, for second through to fifth place.