Laying play-space features on the 10 km island

The play-space features from the 4 km test island now run on the 10 km island the game uses. The plain island passes 7 of the 21 tests and the version with features passes 13. I placed one valley with a gate by hand, and the generator placed 10 rock stacks, 8 knolls and 3 ridge lookouts from the island’s routes. Easy ground fell from 47.22% to 44.19%, which is 3.03 points against my limit of 5.

A 50 second captioned film of the 10 km island. It opens with a hazy orbit, then walks north through the Hook’s Draw gate, past a stack and a summit coming into view, then to a ridge lookout. The flat-sided gray wall near the middle is an older panel problem, not something new.

How do you keep a variant next to the island it changes?

I wanted a before and after on the same island, so a variant can now say "over": "<other variant>". The featured island, wishbone.small_interest, is laid over wishbone.small and holds only the play-space block. The loader merges the chain of variants and refuses a loop. The plain island still loads byte for byte the same, so its inputs hash didn’t change. The comparison sheet pairs a variant with the one it lies over.

Why did the first rounds break the island?

The first rounds scored well, 14 of 21 with easy ground around 42.5%, but the island’s checks then failed three more constraints than the plain island does. The stacks and knolls had landed on the Tower’s vale floor, its grounds and Marrow Cove’s terrace. So I made the placers keep the island’s designs clear, the same way a valley’s corridor is kept clear:

rivers that drain 0.5 km2 or more, plus 40 m on each side
the island’s flats, out to their radius times their aspect plus 60 m
the starter valley at 1.1 times its radius, and 120 m either side of the lines to its exits
passes, out to the larger of 120 m and 1.6 times their half width, which is how the check walks them

The placers also route the way the play code does, to the named places, the banked flats and the summits. They only use ground reached on foot. That matters at 10 km, because one place, the Plug, is a true island that nothing can walk to. And a new strand test drops any form, knoll or lookout that would cut more than 0.05% of the land off from the walkable ground joined to the start. With all that, the check is back to the two failures the plain island already has, the East Door and the Tower’s vale.

Asking for 10 stacks at 800 m spacing still laid only 8, because the candidate sites ran out. 700 m spacing with 400 candidates laid 10.

Why were most of the lookouts on the beach?

Six of the nine lookouts were pads 2 to 6 m above the sea on the flat east coast. A low bump there passed as a crest, because the rule only asked for ground above the 300 m mean and the highest within 60 m. A lookout now has to stand 30 m above the lowest ground within 300 m, with the sea counted as 0. Three were left, at 200, 785 and 380 m. The vantage tests fell to what they honestly are, 9 vantage points that together see 47% of the land.

What did I change in the rules?

I was asked two questions on 21 October. The plain island’s easy ground was already 47.2%, under my 50% floor, so there was nothing to hold. And the measure of walkable ground reached left out 2.2 times a form’s height plus 40 m round each rock form. Round a 200 m stack that is about half a square kilometer of reached plain, so the share read nearly a point low. My answers were “1) fine 2) switch”. Now only the 5-point limit holds on this island, and the measure leaves out the body of the form, 0.7 times its height plus 40 m. Walkable ground reached goes from 94.0% to 93.5% with the features.

What can’t the features fix?

Some failures belong to the island, not the features. The Plug can never be reached on foot. The island’s gorge and the Saddle cliffs measure 0 m on the narrow-way test even on the plain island. One valley is 1.5% of route length where the test wants 35%, and more valleys would cost far more than 5 points of easy ground. The Withers and north-coast routes go 222 s without a beat. I’m reporting those and not loosening any test.

Every build is chaotic here. Adding the one valley changed the whole erosion, with new lakes covering 0.7 to 1% of the land, so I judged each of the 15 rounds by the mechanism and by several measures. A build takes about 212 s plain and 210 to 500 s with features.

What’s left is the valleys, the vantage tests, the long stretch without a beat, one summit the routes barely see, and about 0.2% of the land cut off in small pockets. I haven’t found what strands those pockets yet.