In calculated questions using certain formulae (apparently only ones that produce a negative answer) and geometric tolerances, the minimum and maximum tolerance limits are reversed, and the correct answer is marked as incorrect.

To reproduce, use the following example.

Formula for answers 1 and 3:

-1*(({C1a}+{C2a})*({R1}+{R2})/1000000)*log(4.5)*1000

Formula for answers 2 and 4:

(({C1a}+{C2a})*({R1}+{R2})/1000000)*log(4.5)*1000

Tolerance for answers 1 and 2: 0.01 (geometric)
Tolerance for answers 3 and 4: 0.1 (geometric)
All answers displayed to 3 significant figures.

Wild card C1a: 2.4
Wild card C2a: 1.1
Wild card R1: 2.5
Wild card R2: 7.8

The 'Answers tolerance parameters' section of the form reports the following.

For answer 1 (-ve):
-1*((2.4+1.1)*(2.5+7.8)/1000000)*log(4.5)*1000 = -0.0542
ERROR Correct answer : -0.0542 outside limits of true value
Min: -0.053685138766123 — Max: -0.054764210055323

For answer 2 (+ve):
((2.4+1.1)*(2.5+7.8)/1000000)*log(4.5)*1000 = 0.0542
Correct answer : 0.0542 inside limits of true value
Min: 0.053685138766123 — Max: 0.054764210055323

For answer 3 (-ve):
-1*((2.4+1.1)*(2.5+7.8)/1000000)*log(4.5)*1000 = -0.0542
ERROR Correct answer : -0.0542 outside limits of true value
Min: -0.049292718321622 — Max: -0.059644189169164

For answer 4 (+ve):
((2.4+1.1)*(2.5+7.8)/1000000)*log(4.5)*1000 = 0.0542
Correct answer : 0.0542 inside limits of true value
Min: 0.049292718321622 — Max: 0.059644189169164

Clearly, for the two negative answers in this example the minimum and maximum limits are the wrong way around.

In the question preview, the 'Fill in correct responses' button enters the correct answer of -0.0542 (answer 1 is the one for which 100% is awarded), but then clicking the 'Check' button marks the answer wrong and gives the feedback "The correct answer is: -0.0542".