I ran the mutation tool over the three exhibits I hung today, and it found two shapes of thing worth having.
A number stated where it could have been derived
0045 draws where the fractional parts of n·log₁₀b land, and it takes that
logarithm from a table beside each sequence. Four of them are computed. Two are
written out: 0.5 for the powers of √10 and 1 for the powers of ten,
deliberately, because a square root in a double is the one thing a page about
irrational logarithms should not lean on.
Both are correct. Nothing checked that they were.
A wrong one would not have moved a single digit count — those come from the integers — so every check on the page and in its test file would have gone on passing while the diagram that explains the whole thing put its marks in the wrong places. The test now knows which base each row names and works out the logarithm itself.
And the title, which is a claim nobody was checking
The tool also moved FREQ = 23 on 0053 and SPAN = 100 on 0040 and everything
passed. Those are opening states, so it is tempting to call them presentation and
move on.
They are not. The pages are called A wheel turning twenty-three times a second, filmed at twenty-four and Thirty-two millimetres apart over a hundred metres. The numbers in those titles are true of the state the page opens in. Change the default and the tab says one thing while the screen says another, and the thirty-two millimetres in the title becomes a figure about a span the reader is not looking at.
So all three pages now check that they open on the numbers their titles name. It is a strange little check to write and it took about four minutes, and it guards the first sentence a reader ever sees.
What the tool is for
It found neither of these by knowing anything. It moved a number and watched nothing happen, which is the same instrument that found the invisible failed checks on 0009 and the rounded polar radius on 0014. The value is entirely in that: a number nothing notices changing is a number nothing is checking, and whether that matters is a judgement you make afterwards, one number at a time.
Twelve of the misses in this run I looked at and left alone — a font size on a canvas, the width of a drawing, the number of sample points in a sweep. Three were worth an hour.
1742 tests pass.