I am going for number 11, mainly because other adventurers with predictions similar to 11 did unusually well.
Thank you for posting this. Getting a very good or pefect answer felt a lot easier than most, however getting from a very good answer to a perfect answer seemed more difficult than most. I identified a very good answer very quickly just by looking for combinations that were present in the dataset. It was then rather frustrating to make a lot of progress in untangling the rules and still being unable to find a better solution than the first one I found. Overalll I would rate it as difficulty = 2/5 playability 2/5 where 3 is an average D and D puzzle.
!> I was able to generalise the formula to a wide range of scenrios involving heades, but the eyes are more complicated. Aside from forceing the tax to 0 when it would otherwise be relatively small I didn't find a consistent pattern. As I can't find anything obviously better I will stick with my original entry.
The tax is always the same for the same set of monster parts so no randomness is involved.
I then looked for entries where only one type of part was present. With the exception of the heads this gave some obvious formulas:
When only eyes are present no tax is paid
When only heads are present tax is 2.8 for 1, 8.4 for 2 21 for 3 and 29.4 for 4.
When only skulls are present tax is the number of skulls
When only hands are present the tax is 0.2 times the number of hands.
When only horns are present and their number is < 5 the tax is 1.4*number of horns, and 1.75*number of horns when >= 5 are present.
Next I looked for records where only two types of parts were present, but with the following the exceptions it didn't give anything obvious:
When only skulls and hands are present the tax rate is #SKULL + 0.2*#HAND
When only horns and hands are present the tax rate is:
1.4*#HORN + 0.4*#HAND provided the total tax bill is less than 6
else when horns < 5 and the total tax is < 18: 2.1*#HORN + 0.6*#HAND
else when horns < 5: 2.8*#HORN + 0.8*#HAND
When horns >= 5 : 1.75*#HORN + 0.5*#HAND
After much looking at the data a lot I was then able to find the following formulas when skulls, horns, and hands were all present:
1.4*#HORN + 0.4*#HAND + 2*#SKULL provided the result is < 6
Else 1.75*HORN + 0.5*HAND + 2.5*SKULL provided there are at least 5 horns
Else 2.1*#HORN + 0.6*#HAND + 3*SKULL provided the result is less than 18
Else 2.8*#HORN + 0.8*#HAND + 4*SKULL provided ther result is less than 40
Else 3.5*#HORN + 1*#HAND + 5*#SKULL
Eyes and particularly heads seem to introduce a lot of extra complexity.
The best record I could find with 4 eyes and 4 heads had 4 eyes, 4 heads and 1 hand, so I tried to give these to 1 adventurer, and then allocate the rest amonst the remaining 3 according to these formulas. However the result was worse than the best I could find by looking up the tax for various combinations in the datafile. I will therefor use this as my entry if I can't work out what is going on with the eyes/heads.
Adventurer 1: EYE(1)HEAD(1)SKULL(5)HORN(6)HAND(2)TAX: 23
Adventurer2: EYE(1)HEAD(1)SKULL(0)HORN(1)HAND(0)TAX: 0
Adventurer3: EYE(1)HEAD(1)SKULL(0)HORN(0)HAND(3)TAX: 0
Adventurer4: EYE(1)HEAD(1)SKULL(0)HORN(0)HAND(3)TAX: 0
Total estimated tax is 23
Thank you for posting this. I rate this as about 3/5 for both complexity and quality. It was relatively easy to make some progress, but getting a perfect answer that you know is perfect looks hard. I think adding more rooms to the dungeon would have made things more difficult as it would have added more combinations to consider.
I only had time to construct a simple model based on the average value of the score for different encounters in different dungeons. Based on this my submission is:
COG/GOB/WHD
And when Goblins aren't present:
CON/WOB/NHD
I was expecting earlier choices of foraging location to have a much stronger impact, and mistook some of the randomness for affects of earlier choices. In retrospect it would have been better to spend longer exploreing various possibilites rather than settling on an exploit strategy so soon. Adding an explicit target was a big improvement as it gave some idea of "how good a strategy" we should be searching for.
I starte off by going to each one in turn. And then visiting those sites that yielded the best results. It rapidly became clear that some sites were better at different times of the day, and visiting one site too much could result in diminishing returns. Then I spent an entire day at each site in turn and then visited the sites that produce the most food at a given time. The relative order of visiting sites did have an affect so this didn't always work particularly well. After a bit of playing around and experimentation I ended up settling on the following cycle towards the end:
08:00 Pig Swamp
09:00 Bunny Ride
10:00 Snake Shores
11:00 Dog Valley
12:00 Dog Valley
13:00 Dog Valley
14:00 Dog Valley
15:00 Rooster Peaks
16:00 Monkey Meadow
17:00 Bunny Ridge
18:00 Pig Swamp
19:00 Pig Swamp
20:00 Tiger Forest
21:00 Tiger Forest
22:00 Tiger Forest
23:00 Tiger Forest
This got me 2761 food total.
I enjoyed this. Though I think the 10 minuite estimate is far too low. This leaves less than 1s per decision!
Thank you for posting this. Overall I would rate this as a middle of the road (ie good) scenario. Complexity 3/5, quality 3/5.
I thought the bonus objective was in principal a good addition, though it could have done with an extra couple of known words. As it is unless you spot the anomaly with Cadagals boots it seems next to impossible to figure out what it might mean.
Overall I think a gap of about 2 months is better than short gaps of one month followed by longer gaps of several months. Though possibly not this time as that would put it right in the middle of the Christmas period!
Thank you for posting this. I found it more difficult than most to untangle the rules behind what was going on, but easier than most to find a candidate that I could be reasonably confident would do a lot better than random selection (Which then turned out to be the best). Possibly having more candidates to choose from, with a smaller difference between the top candidates, would have made ML more effective relative to pure analysis.
I would prefer it if the next one came a week later as I am expecing to be very busy that week.