diff --git a/BREAKING-CHANGES.md b/BREAKING-CHANGES.md
index 10b714cab..24203c315 100644
--- a/BREAKING-CHANGES.md
+++ b/BREAKING-CHANGES.md
@@ -34,6 +34,28 @@ read first.
 | **silent** | `"1/(x^4 + 4)".Integrate("x")` | unevaluated | the antiderivative over its two quadratic factors |
 | **silent** | `"sin(-x) + sin(x)".Simplify()` | `sin(-x) + sin(x)` | `0` |
 | **silent** | `"cos(-x)".Simplify()` and `"abs(-x)".Simplify()` | unchanged | `cos(x)`, `abs(x)` |
+| **silent** | `"cos(0 ^ y)".InnerSimplified` | `-(-1) provided ...` | `1 provided ...` |
+
+### `InnerSimplified` is idempotent again
+
+An exact trigonometric value reached through a half turn — `cos(x)` read off the table as
+`-cos(pi - x)` — was built as a negation over the value it turned to and handed back unfolded. Where
+the answer is then *wrapped* rather than rebuilt, nothing normalises it again:
+
+```
+"cos(0 ^ y)".InnerSimplified
+
+was  -(-1) provided y / 2 * (1 + 1 / sgn(y) ^ 2) > 0
+is       1 provided y / 2 * (1 + 1 / sgn(y) ^ 2) > 0
+```
+
+Both are the same value, so nothing was wrong — but applying `InnerSimplified` twice gave a different
+tree from applying it once, and a great deal of the library treats what it hands back as settled.
+`Simplify` was unaffected, since it normalises again anyway.
+
+Found by `canoncheck`, the canonical-form harness: it was the only idempotence failure in 834
+generated expressions, and the count is now zero.
+[#930](https://github.com/asc-community/AngouriMath/issues/930).
 
 ### The parity identities are applied
 
diff --git a/Sources/AngouriMath/Functions/Simplification/TrigonometricTableValues.cs b/Sources/AngouriMath/Functions/Simplification/TrigonometricTableValues.cs
index 38fe31cb6..8b78be55b 100644
--- a/Sources/AngouriMath/Functions/Simplification/TrigonometricTableValues.cs
+++ b/Sources/AngouriMath/Functions/Simplification/TrigonometricTableValues.cs
@@ -92,7 +92,14 @@ private static bool OrHalfTurn(System.Func<Complex, (bool Found, Entity? Value)>
             }
             if (lookup(HalfTurn(arg)) is (true, { } turned))
             {
-                res = oddOverHalfTurn ? -turned : turned;
+                // Folded here rather than left to whoever receives it. The negation is built
+                // above a value that is already in its final form, so nothing downstream is
+                // obliged to look at it again -- and where the answer is wrapped rather than
+                // rebuilt, nothing does: cos(0 ^ y) came back as `-(-1) provided ...` and
+                // needed a second InnerSimplified to become `1 provided ...`, which is an
+                // idempotence failure rather than a cosmetic one.
+                // https://github.com/asc-community/AngouriMath/issues/930
+                res = oddOverHalfTurn ? (-turned).InnerSimplified : turned;
                 return true;
             }
             res = null;
diff --git a/Sources/Tests/UnitTests/Common/InnerSimplifyTest.cs b/Sources/Tests/UnitTests/Common/InnerSimplifyTest.cs
index 34a503063..6a2cc846d 100644
--- a/Sources/Tests/UnitTests/Common/InnerSimplifyTest.cs
+++ b/Sources/Tests/UnitTests/Common/InnerSimplifyTest.cs
@@ -427,6 +427,34 @@ public void InequalityShouldBeNaN(string expr) {
             expr.ToEntity().Evaled.ShouldBe(MathS.NaN);
             expr.ToEntity().InnerSimplified.ShouldBe(MathS.NaN);
         }
+
+        /// <summary>
+        /// Normalising a normalised expression leaves it alone. `cos(0 ^ y)` did not: the
+        /// exact value of a cosine reached through a half turn was built as a negation over
+        /// the value it turned to and handed back unfolded, so the answer arrived as
+        /// `-(-1) provided ...` and became `1 provided ...` only if something normalised it
+        /// again. https://github.com/asc-community/AngouriMath/issues/930
+        /// </summary>
+        /// <remarks>
+        /// Compared as entities rather than as strings, deliberately. The general failure here
+        /// includes two trees that print alike and differ — `(x + y) + a` and `x + (y + a)` do
+        /// — so a comparison of printed forms would wave those through. See
+        /// `Docs/Contributing/CanonicalForm.md`.
+        /// </remarks>
+        [Theory]
+        [InlineData("cos(0 ^ y)")]
+        [InlineData("sin(0 ^ y)")]
+        [InlineData("cos(0)")]
+        [InlineData("cos(pi)")]
+        [InlineData("sin(pi / 6)")]
+        [InlineData("cos(pi / 3)")]
+        [InlineData("tan(pi / 4)")]
+        [InlineData("x + sin(0 ^ y)")]
+        public void InnerSimplifiedIsIdempotent(string expr)
+        {
+            var once = expr.ToEntity().InnerSimplified;
+            Assert.Equal(once, once.InnerSimplified);
+        }
     }
 }