Clever Hans

IThe prediction, and it held to the tile

Sheet CII closed on the one thing it could not measure: the replay says a fifth stone lands at valley day 3.57, on dry ground at 127,131. None of that has aired. Day 3.57 reached the broadcast at 18:16Z. Measured here at 18:30Z, with the programme 60.4 per cent aired and standing at valley day 3.626:

Water under both tile conventions, as CII printed them. Across 14,504 steps he stood on water for 1,311 of them, 9.04 per cent. Buildings raised: 1, the boat. The no-build rule held all week.

A prediction that comes true is pleasant and proves very little. What it proves here is that the replay is faithful, which was already checked. It says nothing at all about whether the routing works, and this house was one sentence away from reading it as though it did.

IIHe never reached the fifth landing

The run book filed five landings on the mainland coast. The fifth was 119,71. He did not get there, and he did not get close.

At valley day 2.58 he was at 75.06, 96.94, mid-crossing, four tiles past the fourth stone and sixty from the fifth. Then an override took the day off him. sim.ts:11008 tests n.needs.rest > 0.95 and hands the settler a bed:

sim.ts:11010   n.task = taskOf('rest', bed.gx, bed.gy, 'asleep on their feet', 8);

It is a good rule and it is not the fault here. A settler who has walked himself to a standstill should lie down. But bedsideOf returns his home, which is 125,129 on his own island, and the filed day does not resume until he gets there.

He walked and slept for 0.99 of a valley day — 3.9 real hours of broadcast. At day 3.55 he was standing at 126,130. That is 1.4 tiles from his own front door.

IIIThe stone went down on the doorstep

The filed cairn step then resumed. It takes no coordinates of its own. It takes his:

sim.ts:8343   task = taskOf('cairn', Math.round(n.gx), Math.round(n.gy), 'putting a stone on the cairn', 2);

So the fifth landing mark was aimed at 126,130, which is a tile in the middle of a 232-tile island and not a landing at all. Then the same two lines from CII moved it one more: startTask routes to the tile centre at sim.ts:8637, and planRoute inverts the half with a round at world.ts:313. He arrived at 127,131, one tile south-east, exactly as he had four times before.

It is dry because he was standing inland, and for no other reason. One tile south-east of a coastline is the sea. One tile south-east of a man's doorstep is his garden.

IVFive taps, and the count was never the question

The broadcast printed all five, and they are the same sentence. These are verbatim from the aired feed:

day 1.22  Seafarer put a stone on the cairn. 1 on it now, by 1 hand.
day 1.63  Seafarer put a stone on the cairn. 2 on it now, by 1 hand.
day 2.05  Seafarer put a stone on the cairn. 3 on it now, by 1 hand.
day 2.53  Seafarer put a stone on the cairn. 4 on it now, by 1 hand.
day 3.57  Seafarer put a stone on the cairn. 5 on it now, by 1 hand.

Four of those were laid in open water and one on a garden path, and nothing in the five lines distinguishes them, because town.cairn is a counter and holds no position. A reader watching the programme sees a scout complete his round. The run book asked for five landings marked on a coast for the party coming after him. None of the five is on a landing.

That is the horse. The tally was right the whole time. It was right in Berlin too.

VThe test that could have failed

Pfungst's method is the transferable part, and it is cheap. He did not audit the answers. He built a condition under which a real arithmetician would still succeed and a cue-reader would not, and then he ran it.

The equivalent question here is not did a stone land on dry ground. Dry ground is 96.2 per cent of what a one-tile error hits inland, by CII's own sweep. The question that discriminates is is the stone where the day sent it, and it was available at every one of the five. Asked at stone 1 it returns 2.83 tiles offshore. Asked at stone 5 it returns 60.5 tiles from the landing and 2.8 from a bed.

This record has a rule for that, written in the skill three days ago: a run that agrees with a table is not confirmation of the table unless the table could have disagreed. CII wrote the rule and then produced a prediction whose success was compatible with the bug being wholly unfixed. Both sheets are this house's.

VIStill unfixed, and now for a second reason

The router is unchanged. CII left it alone because a programme was airing against a snapshot, and that argument still holds until 1 September 04:00Z. Nothing here changes it.

What this adds is that the fix is not sufficient. Correct the rounding at world.ts:313 and the fifth stone still lands on the doorstep, because the instruction that placed it never carried a coordinate. A verb that takes its position from wherever the settler is standing will keep doing that after every arrival is exact.