A cleared control was announced by a line that called it a failure

The banner refuses to say "PASSED" when a negative control fails, for the
reason its own comment gives: it would tell a reader the exact opposite of
what was measured. The per-test loop eight lines above it was never given
the same treatment, so a refuted control printed

  [UITest] Issue31059: FAILED (expected)   <- green
  CONTROL DID NOT CLEAR                    <- red

one after the other, and only the second was true. In the control arm a
failing test is the refutation, not the expectation.

Both directions are fixed, because a change to the failing branch alone
leaves a cleared control reported as "PASSED (should fail!)" - the same
inversion the other way round. The reproduction arm's wording is
byte-identical: Validate-ReplicationCandidate.Tests.ps1 builds a console
fixture from that exact string, and the markdown report is deliberately
untouched so existing validators keep reading what they read today.

The full-verification renderers were checked and left alone; their arms
already label themselves correctly.

Mutation tested: reverting the change fails 3 of 5, making only the
failing branch arm-aware fails 2, drifting the reproduction wording
fails 1. Suites 1731 passed, 2 failed, both pre-existing.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 735ac9a2-7bec-4baa-ad19-c298e5bc795a

Author: Copilot CI

.github/skills/verify-tests-fail-without-fix/scripts/Verify-TestsFail.Tests.ps1

@@ -1801,3 +1801,54 @@ Describe 'The authoritative test result is kept as evidence' {
             ConvertFrom-Json).resultFile | Should -BeExactly ''
     }
 }
+
+Describe 'The per-test line knows which control arm it is reporting' {
+    # The banner already refuses to say "PASSED" for a failing negative control,
+    # for the reason its own comment gives: it would tell a reader the exact
+    # opposite of what was measured. The per-test loop a few lines above it was
+    # not given the same treatment, so build 15085903 printed
+    #   [UITest] Issue31059: FAILED (expected)      <- green
+    #   CONTROL DID NOT CLEAR                       <- red
+    # one after the other, and only the second was true.
+    BeforeAll {
+        $script:PerTestBlock = $script:VerifierSource.Substring(
+            $script:VerifierSource.IndexOf('# Show per-test results'),
+            2200)
+    }
+
+    It 'reports a failing negative control as the refutation, not as the expectation' {
+        $script:PerTestBlock | Should -Match 'control did not clear'
+    }
+
+    It 'reports a passing negative control as the expected outcome' {
+        $script:PerTestBlock | Should -Match 'expected with the trigger removed'
+    }
+
+    It 'branches on Purpose for both the failing and the passing case' {
+        # Guarding both halves: a fix applied to only the failing branch leaves a
+        # passing control reported as "PASSED (should fail!)", which is the same
+        # inversion in the other direction.
+        ([regex]::Matches($script:PerTestBlock, "Purpose -eq 'NegativeControl'")).Count |
+            Should -BeGreaterOrEqual 2
+    }
+
+    It 'leaves the reproduction arm wording byte-identical' {
+        # Validate-ReplicationCandidate.Tests.ps1 builds a console fixture
+        # containing this exact string, and the report renderer is deliberately
+        # untouched so existing validators keep reading what they read today.
+        $script:PerTestBlock | Should -Match 'FAILED \u2705 \(expected\)'
+        $script:PerTestBlock | Should -Match 'PASSED \u274c \(should fail!\)'
+    }
+
+    It 'still classifies env and build errors before either verdict' {
+        # An infrastructure failure is neither arm's result, and it has to keep
+        # outranking both so a build break cannot be read as a cleared control.
+        $envAt   = $script:PerTestBlock.IndexOf('ENV ERROR')
+        $buildAt = $script:PerTestBlock.IndexOf('BUILD ERROR')
+        $verdict = $script:PerTestBlock.IndexOf("Purpose -eq 'NegativeControl'")
+        $envAt   | Should -BeGreaterThan 0
+        $buildAt | Should -BeGreaterThan 0
+        $envAt   | Should -BeLessThan $verdict
+        $buildAt | Should -BeLessThan $verdict
+    }
+}

.github/skills/verify-tests-fail-without-fix/scripts/verify-tests-fail.ps1

@@ -2373,9 +2373,23 @@ if ($DetectedFixFiles.Count -eq 0) {
         } elseif ($r.Error) {
             Write-Host "  $icon [$($r.TestType)] $($r.TestName): ⚠️ ERROR — $($r.Error)" -ForegroundColor Yellow
         } elseif (-not $r.Passed) {
-            Write-Host "  $icon [$($r.TestType)] $($r.TestName): FAILED ✅ (expected)" -ForegroundColor Green
+            if ($Purpose -eq 'NegativeControl') {
+                # A failing test is the expected result for the reproduction arm
+                # and the refutation for this one. Printing the reproduction's
+                # green "(expected)" here told a reader the exact opposite of
+                # what was measured, directly above the CONTROL DID NOT CLEAR
+                # banner that says so - the same mistake the banner below was
+                # already written to avoid.
+                Write-Host "  $icon [$($r.TestType)] $($r.TestName): FAILED ❌ (control did not clear)" -ForegroundColor Red
+            } else {
+                Write-Host "  $icon [$($r.TestType)] $($r.TestName): FAILED ✅ (expected)" -ForegroundColor Green
+            }
         } else {
-            Write-Host "  $icon [$($r.TestType)] $($r.TestName): PASSED ❌ (should fail!)" -ForegroundColor Red
+            if ($Purpose -eq 'NegativeControl') {
+                Write-Host "  $icon [$($r.TestType)] $($r.TestName): PASSED ✅ (expected with the trigger removed)" -ForegroundColor Green
+            } else {
+                Write-Host "  $icon [$($r.TestType)] $($r.TestName): PASSED ❌ (should fail!)" -ForegroundColor Red
+            }
         }
     }
     Write-Host ""