Reading mutation results and strengthening tests

The function

mean_absolute_deviation <- function(x, center) {
  mean(abs(x - center))
}

mean_absolute_deviation computes the average distance from a center point: mean(abs(x - center)). The subtraction x - center is the step that arithmetic operator mutators target.

Weak test

test_that("mean absolute deviation is non-negative", {
  expect_gte(mean_absolute_deviation(c(1, 3, 5), 3), 0)
})

The test asserts that the result is non-negative (expect_gte(..., 0)). This checks a property of the output rather than its value.

Mutation testing output:

ℹ Mutation Testing
  |   K |   S |   E |   T |   % | Mutator  | File 
x |   0 |   1 |   0 |   1 |   0 | - → +    | mad.R 


── Results ─────────────────────────────────────────────────────────────────────
[ KILLED 0 | SURVIVED 1 | ERRORS 0 | TOTAL 1 | SCORE 0.0% ]

- → + survives. Changing x - center to x + center gives a different number, but abs(x + center) is also non-negative for all inputs. For x = c(1,3,5) and center = 3:

• Original: abs(c(-2, 0, 2)) → mean = 4/3 ≈ 1.33
• Mutant: abs(c(4, 6, 8)) → mean = 6

Both are ≥ 0. The assertion cannot distinguish them.

This pattern is common in LLM-generated tests that check what is obviously true about the result (it is positive, it is numeric) rather than what is specifically correct.

The fix

test_that("mean absolute deviation equals average distance from center", {
  expect_equal(mean_absolute_deviation(c(1, 3, 5), 3), 4 / 3)
})

Replace the directional assertion with expect_equal and a value computed by hand. Now the mutant returns 6, which is not 4/3, and the test fails.

After the fix:

ℹ Mutation Testing
  |   K |   S |   E |   T |   % | Mutator  | File 
✔ |   1 |   0 |   0 |   1 | 100 | - → +    | mad.R 


── Results ─────────────────────────────────────────────────────────────────────
[ KILLED 1 | SURVIVED 0 | ERRORS 0 | TOTAL 1 | SCORE 100.0% ]

When an arithmetic mutant survives, replace directional assertions (expect_gt, expect_gte) with exact-value assertions (expect_equal). Compute the expected value by hand and hard-code it.

The function

shipping_cost <- function(weight_kg) {
  if (weight_kg > 5) 15.00 else 5.00
}

shipping_cost returns a flat rate based on whether the package is over or under 5 kg. The strict > operator means a 5 kg package pays the lower rate; >= would charge it the higher rate.

Weak test

test_that("heavy packages cost more than light ones", {
  expect_gt(shipping_cost(10), shipping_cost(2))
})

The test confirms that a heavy package costs more than a light one. It passes inputs of 10 and 2 kg — both clearly on opposite sides of the boundary — so it never exercises the value 5 itself.

Mutation testing output:

ℹ Mutation Testing
  |   K |   S |   E |   T |   % | Mutator  | File 
✔ |   1 |   0 |   0 |   1 | 100 | > → <    | shipping.R 
x |   1 |   1 |   0 |   2 |  50 | > → >=   | shipping.R 


── Results ─────────────────────────────────────────────────────────────────────
[ KILLED 1 | SURVIVED 1 | ERRORS 0 | TOTAL 2 | SCORE 50.0% ]

> → >= survives. Changing weight_kg > 5 to weight_kg >= 5 only affects input weight_kg = 5 exactly. For inputs 10 and 2, the function returns identical results under both operators, so the test cannot tell the operators apart.

In production, >= 5 instead of > 5 would route 5 kg shipments to the expensive tier and no test would catch the regression.

The fix

test_that("heavy packages cost more than light ones", {
  expect_gt(shipping_cost(10), shipping_cost(2))
})

test_that("5kg falls into the lower-cost tier", {
  expect_equal(shipping_cost(5), 5.00)
})

Add a test that passes the exact boundary value 5. With > 5, the condition is FALSE and the function returns 5.00. With >= 5, it is TRUE and returns 15.00. This difference kills the mutant.

After the fix:

ℹ Mutation Testing
  |   K |   S |   E |   T |   % | Mutator  | File 
✔ |   1 |   0 |   0 |   1 | 100 | > → <    | shipping.R 
✔ |   2 |   0 |   0 |   2 | 100 | > → >=   | shipping.R 


── Results ─────────────────────────────────────────────────────────────────────
[ KILLED 2 | SURVIVED 0 | ERRORS 0 | TOTAL 2 | SCORE 100.0% ]

When a comparison mutant survives, find the boundary value implied by the operator and add a test that passes exactly that value.