diff --git a/.gas-snapshot b/.gas-snapshot
index e3a596da..af471e8e 100644
--- a/.gas-snapshot
+++ b/.gas-snapshot
@@ -3,36 +3,37 @@ LibDecimalFloatCompareTest:testCompareOnePointSix() (gas: 5040)
 LibDecimalFloatCompareTest:testCompareTen() (gas: 5333)
 LibDecimalFloatDecimalAddTest:testAdd123456789987654321() (gas: 6549)
 LibDecimalFloatDecimalAddTest:testAdd123456789e9987654321() (gas: 6830)
-LibDecimalFloatDecimalAddTest:testAddNeverRevertIsNormalized(int256,int256,int256,int256) (runs: 2063, μ: 14270, ~: 14152)
+LibDecimalFloatDecimalAddTest:testAddNeverRevertIsNormalized(int256,int256,int256,int256) (runs: 2050, μ: 14296, ~: 14175)
 LibDecimalFloatDecimalAddTest:testAddOneOneNotNormalized() (gas: 5958)
 LibDecimalFloatDecimalAddTest:testAddOneOnePreNormalized() (gas: 4390)
 LibDecimalFloatDecimalAddTest:testAddOneZero() (gas: 5275)
-LibDecimalFloatDecimalAddTest:testAddSameExponentSameCoefficient(int256,int256) (runs: 2063, μ: 6848, ~: 6862)
+LibDecimalFloatDecimalAddTest:testAddSameExponentSameCoefficient(int256,int256) (runs: 2050, μ: 6849, ~: 6874)
 LibDecimalFloatDecimalAddTest:testAddZero() (gas: 4388)
-LibDecimalFloatDecimalAddTest:testAddZeroAnyExponent(int128) (runs: 2063, μ: 10055, ~: 9992)
+LibDecimalFloatDecimalAddTest:testAddZeroAnyExponent(int128) (runs: 2050, μ: 10061, ~: 9992)
 LibDecimalFloatDecimalAddTest:testAddZeroOne() (gas: 5256)
-LibDecimalFloatDecimalAddTest:testAddingSmallToLargeReturnsLargeExamples() (gas: 22033)
-LibDecimalFloatDecimalAddTest:testAddingSmallToLargeReturnsLargeFuzz(int256,int256,int256,int256) (runs: 2054, μ: 18704, ~: 18883)
+LibDecimalFloatDecimalAddTest:testAddZeroToAny(int256,int256,int256) (runs: 2050, μ: 17340, ~: 17323)
+LibDecimalFloatDecimalAddTest:testAddingSmallToLargeReturnsLargeExamples() (gas: 22282)
+LibDecimalFloatDecimalAddTest:testAddingSmallToLargeReturnsLargeFuzz(int256,int256,int256,int256) (runs: 2049, μ: 18698, ~: 18818)
 LibDecimalFloatDecimalAddTest:testGasAddOne() (gas: 1084)
 LibDecimalFloatDecimalAddTest:testGasAddZero() (gas: 1106)
-LibDecimalFloatDecimalTest:testFixedDecimalRoundTripLossless(uint256,uint8) (runs: 2063, μ: 10144, ~: 10196)
+LibDecimalFloatDecimalTest:testFixedDecimalRoundTripLossless(uint256,uint8) (runs: 2050, μ: 10145, ~: 10196)
 LibDecimalFloatDecimalTest:testFromFixedDecimalLossyComplicated() (gas: 1022048)
 LibDecimalFloatDecimalTest:testFromFixedDecimalLossyNormalizedMax() (gas: 760206)
 LibDecimalFloatDecimalTest:testFromFixedDecimalLossyNormalizedMaxPlusOne() (gas: 866032)
 LibDecimalFloatDecimalTest:testFromFixedDecimalLossyOne() (gas: 975616)
 LibDecimalFloatDecimalTest:testFromFixedDecimalLossyOneMillion() (gas: 952412)
 LibDecimalFloatDecimalTest:testFromFixedDecimalLossyOverflow() (gas: 838681)
-LibDecimalFloatDecimalTest:testFromFixedDecimalLossyTruncateOne(uint256,uint8) (runs: 2063, μ: 9574, ~: 7195)
-LibDecimalFloatDecimalTest:testFromFixedDecimalLossyTruncateZero(uint256,uint8) (runs: 2063, μ: 8556, ~: 7207)
-LibDecimalFloatDecimalTest:testToFixedDecimalLosslessScaleUp(int256,int256,uint8) (runs: 2056, μ: 16000, ~: 16228)
-LibDecimalFloatDecimalTest:testToFixedDecimalLossyExponentOverflow(int256,int256,uint8) (runs: 2063, μ: 14938, ~: 14718)
-LibDecimalFloatDecimalTest:testToFixedDecimalLossyIdentity(int256,uint8) (runs: 2063, μ: 10110, ~: 9778)
-LibDecimalFloatDecimalTest:testToFixedDecimalLossyNegative(int256,int256,uint8) (runs: 2063, μ: 9916, ~: 10161)
-LibDecimalFloatDecimalTest:testToFixedDecimalLossyScaleUpOverflow(int256,int256,uint8) (runs: 2055, μ: 13915, ~: 14172)
-LibDecimalFloatDecimalTest:testToFixedDecimalLossyTruncate(int256,int256,uint8) (runs: 2063, μ: 14457, ~: 14188)
+LibDecimalFloatDecimalTest:testFromFixedDecimalLossyTruncateOne(uint256,uint8) (runs: 2050, μ: 9549, ~: 7195)
+LibDecimalFloatDecimalTest:testFromFixedDecimalLossyTruncateZero(uint256,uint8) (runs: 2050, μ: 8541, ~: 7207)
+LibDecimalFloatDecimalTest:testToFixedDecimalLosslessScaleUp(int256,int256,uint8) (runs: 2049, μ: 16001, ~: 16228)
+LibDecimalFloatDecimalTest:testToFixedDecimalLossyExponentOverflow(int256,int256,uint8) (runs: 2050, μ: 14938, ~: 14718)
+LibDecimalFloatDecimalTest:testToFixedDecimalLossyIdentity(int256,uint8) (runs: 2050, μ: 10110, ~: 9778)
+LibDecimalFloatDecimalTest:testToFixedDecimalLossyNegative(int256,int256,uint8) (runs: 2050, μ: 9916, ~: 10161)
+LibDecimalFloatDecimalTest:testToFixedDecimalLossyScaleUpOverflow(int256,int256,uint8) (runs: 2049, μ: 13915, ~: 14172)
+LibDecimalFloatDecimalTest:testToFixedDecimalLossyTruncate(int256,int256,uint8) (runs: 2050, μ: 14458, ~: 14188)
 LibDecimalFloatDecimalTest:testToFixedDecimalLossyTruncateLossless() (gas: 14500)
-LibDecimalFloatDecimalTest:testToFixedDecimalLossyUnderflow(int256,int256,uint8) (runs: 2063, μ: 13700, ~: 13580)
-LibDecimalFloatDecimalTest:testToFixedDecimalLossyZero(int256,uint8) (runs: 2063, μ: 4575, ~: 4575)
+LibDecimalFloatDecimalTest:testToFixedDecimalLossyUnderflow(int256,int256,uint8) (runs: 2050, μ: 13699, ~: 13580)
+LibDecimalFloatDecimalTest:testToFixedDecimalLossyZero(int256,uint8) (runs: 2050, μ: 4575, ~: 4575)
 LibDecimalFloatDivideTest:testDivide1Over3() (gas: 6388)
 LibDecimalFloatDivideTest:testDivide1Over3Gas0() (gas: 827)
 LibDecimalFloatDivideTest:testDivide1Over3Gas10() (gas: 7934)
@@ -46,18 +47,21 @@ LibDecimalFloatLog10Test:testExactLogs() (gas: 19667)
 LibDecimalFloatLog10Test:testInterpolatedLookups() (gas: 18818)
 LibDecimalFloatMixedTest:testDivide1Over3() (gas: 7926)
 LibDecimalFloatMultiplyTest:testMultiply123456789987654321() (gas: 4551)
-LibDecimalFloatMultiplyTest:testMultiply123456789987654321WithExponents(int128,int128) (runs: 2063, μ: 14168, ~: 14254)
+LibDecimalFloatMultiplyTest:testMultiply123456789987654321WithExponents(int128,int128) (runs: 2050, μ: 14170, ~: 14254)
 LibDecimalFloatMultiplyTest:testMultiply1e181e19() (gas: 4732)
 LibDecimalFloatMultiplyTest:testMultiplyGasOne() (gas: 1284)
 LibDecimalFloatMultiplyTest:testMultiplyGasZero() (gas: 326)
 LibDecimalFloatMultiplyTest:testMultiplyOneOne() (gas: 4753)
 LibDecimalFloatMultiplyTest:testMultiplyOneZero() (gas: 3628)
 LibDecimalFloatMultiplyTest:testMultiplyZero0Exponent() (gas: 3642)
-LibDecimalFloatMultiplyTest:testMultiplyZeroAnyExponent(int64,int64) (runs: 2063, μ: 3930, ~: 3930)
+LibDecimalFloatMultiplyTest:testMultiplyZeroAnyExponent(int64,int64) (runs: 2050, μ: 3930, ~: 3930)
 LibDecimalFloatMultiplyTest:testMultiplyZeroOne() (gas: 3621)
-LibDecimalFloatPackTest:testNormalizedRoundTrip(int256,int256) (runs: 2063, μ: 11025, ~: 10932)
-LibDecimalFloatPackTest:testPartsRoundTrip(int128,int128) (runs: 2063, μ: 4588, ~: 4588)
+LibDecimalFloatPackTest:testNormalizedRoundTrip(int256,int256) (runs: 2050, μ: 11025, ~: 10932)
+LibDecimalFloatPackTest:testPartsRoundTrip(int128,int128) (runs: 2050, μ: 4588, ~: 4588)
 LibDecimalFloatPower10Test:testInterpolatedLookupsPower() (gas: 43971)
+LibDecimalFloatSubTest:testMinusIsSubZero(int256,int256,int256) (runs: 2050, μ: 15675, ~: 15579)
+LibDecimalFloatSubTest:testSubIsAdd(int256,int256,int256,int256) (runs: 2050, μ: 17940, ~: 18083)
+LibDecimalFloatSubTest:testSubMinSignedValue(int256,int256,int256) (runs: 2050, μ: 17687, ~: 17559)
 LibLogTableBytesTest:testToBytesAntiLogTableDec() (gas: 158027)
 LibLogTableBytesTest:testToBytesAntiLogTableDecSmall() (gas: 160620)
 LibLogTableBytesTest:testToBytesLogTableDec() (gas: 141631)
diff --git a/src/lib/LibDecimalFloat.sol b/src/lib/LibDecimalFloat.sol
index cc41bdb1..eb5ba656 100644
--- a/src/lib/LibDecimalFloat.sol
+++ b/src/lib/LibDecimalFloat.sol
@@ -343,7 +343,11 @@ library LibDecimalFloat {
             // The early return here allows us to do unchecked pow on the
             // multiplier and means we never revert due to overflow here.
             if (alignmentExponentDiff > ADD_MAX_EXPONENT_DIFF) {
-                return (signedCoefficientA, exponentA);
+                if (signedCoefficientA == 0) {
+                    return (signedCoefficientB, exponentB);
+                } else {
+                    return (signedCoefficientA, exponentA);
+                }
             }
             multiplier = 10 ** alignmentExponentDiff;
         }
@@ -359,6 +363,18 @@ library LibDecimalFloat {
         return LibDecimalFloatImplementation.normalize(signedCoefficientA, exponentB);
     }
 
+    /// Subtract two floats together as a normalized result.
+    ///
+    /// This is effectively shorthand for adding the two floats with the second
+    /// float negated. Therefore, the same caveats apply as for `add`.
+    /// @param signedCoefficientA The signed coefficient of the first floating
+    /// point number.
+    /// @param exponentA The exponent of the first floating point number.
+    /// @param signedCoefficientB The signed coefficient of the second floating
+    /// point number.
+    /// @param exponentB The exponent of the second floating point number.
+    /// @return signedCoefficient The signed coefficient of the result.
+    /// @return exponent The exponent of the result.
     function sub(int256 signedCoefficientA, int256 exponentA, int256 signedCoefficientB, int256 exponentB)
         internal
         pure
@@ -368,6 +384,9 @@ library LibDecimalFloat {
         return add(signedCoefficientA, exponentA, signedCoefficientB, exponentB);
     }
 
+    /// Negates and normalizes a float.
+    /// Equivalent to `0 - x`.
+    ///
     /// https://speleotrove.com/decimal/daops.html#refplusmin
     /// > minus and plus both take one operand, and correspond to the prefix
     /// > minus and plus operators in programming languages.
@@ -377,6 +396,12 @@ library LibDecimalFloat {
     /// > (where a and b refer to any numbers) are calculated as the operations
     /// > add(’0’, a) and subtract(’0’, b) respectively, where the ’0’ has the
     /// > same exponent as the operand.
+    ///
+    /// @param signedCoefficient The signed coefficient of the floating point
+    /// number.
+    /// @param exponent The exponent of the floating point number.
+    /// @return signedCoefficient The signed coefficient of the result.
+    /// @return exponent The exponent of the result.
     function minus(int256 signedCoefficient, int256 exponent) internal pure returns (int256, int256) {
         (signedCoefficient, exponent) = LibDecimalFloatImplementation.normalize(signedCoefficient, exponent);
         return (-signedCoefficient, exponent);
diff --git a/test/src/LibDecimalFloat.add.t.sol b/test/src/LibDecimalFloat.add.t.sol
index 8d43d928..e95e09ce 100644
--- a/test/src/LibDecimalFloat.add.t.sol
+++ b/test/src/LibDecimalFloat.add.t.sol
@@ -198,4 +198,22 @@ contract LibDecimalFloatDecimalAddTest is Test {
         assertEq(signedCoefficient, expectedSignedCoefficient);
         assertEq(exponent, expectedExponent);
     }
+
+    /// Adding any zero to any value returns the non-zero value.
+    function testAddZeroToAny(int256 exponentZero, int256 signedCoefficient, int256 exponent) external pure {
+        exponentZero = bound(exponentZero, EXPONENT_MIN / 10, EXPONENT_MAX / 10);
+        exponent = bound(exponent, EXPONENT_MIN / 10, EXPONENT_MAX / 10);
+
+        (int256 expectedSignedCoefficient, int256 expectedExponent) =
+            LibDecimalFloatImplementation.normalize(signedCoefficient, exponent);
+        (int256 signedCoefficientAddZero, int256 exponentAddZero) =
+            LibDecimalFloat.add(0, exponentZero, signedCoefficient, exponent);
+        assertEq(signedCoefficientAddZero, expectedSignedCoefficient);
+        assertEq(exponentAddZero, expectedExponent);
+
+        // Reverse order.
+        (signedCoefficientAddZero, exponentAddZero) = LibDecimalFloat.add(signedCoefficient, exponent, 0, exponentZero);
+        assertEq(signedCoefficientAddZero, expectedSignedCoefficient);
+        assertEq(exponentAddZero, expectedExponent);
+    }
 }
diff --git a/test/src/LibDecimalFloat.minus.t.sol b/test/src/LibDecimalFloat.minus.t.sol
new file mode 100644
index 00000000..88ed4db3
--- /dev/null
+++ b/test/src/LibDecimalFloat.minus.t.sol
@@ -0,0 +1,22 @@
+// SPDX-License-Identifier: CAL
+pragma solidity =0.8.25;
+
+import {LibDecimalFloat, EXPONENT_MIN, EXPONENT_MAX} from "src/lib/LibDecimalFloat.sol";
+import {LibDecimalFloatImplementation} from "src/lib/implementation/LibDecimalFloatImplementation.sol";
+
+import {Test} from "forge-std/Test.sol";
+
+contract LibDecimalFloatSubTest is Test {
+    /// Minus is the same as `0 - x`.
+    function testMinusIsSubZero(int256 exponentZero, int256 signedCoefficient, int256 exponent) external pure {
+        exponentZero = bound(exponentZero, EXPONENT_MIN / 10, EXPONENT_MAX / 10);
+        exponent = bound(exponent, EXPONENT_MIN / 10, EXPONENT_MAX / 10);
+
+        (int256 signedCoefficientMinus, int256 exponentMinus) = LibDecimalFloat.minus(signedCoefficient, exponent);
+        (int256 expectedSignedCoefficient, int256 expectedExponent) =
+            LibDecimalFloat.sub(0, exponentZero, signedCoefficient, exponent);
+
+        assertEq(signedCoefficientMinus, expectedSignedCoefficient);
+        assertEq(exponentMinus, expectedExponent);
+    }
+}
diff --git a/test/src/LibDecimalFloat.sub.t.sol b/test/src/LibDecimalFloat.sub.t.sol
new file mode 100644
index 00000000..739453a3
--- /dev/null
+++ b/test/src/LibDecimalFloat.sub.t.sol
@@ -0,0 +1,48 @@
+// SPDX-License-Identifier: CAL
+pragma solidity =0.8.25;
+
+import {LibDecimalFloat, EXPONENT_MIN, EXPONENT_MAX} from "src/lib/LibDecimalFloat.sol";
+import {LibDecimalFloatImplementation} from "src/lib/implementation/LibDecimalFloatImplementation.sol";
+
+import {Test} from "forge-std/Test.sol";
+
+contract LibDecimalFloatSubTest is Test {
+    /// Sub is the same as add, but with the second coefficient negated.
+    function testSubIsAdd(int256 signedCoefficientA, int256 exponentA, int256 signedCoefficientB, int256 exponentB)
+        external
+        pure
+    {
+        exponentA = bound(exponentA, EXPONENT_MIN / 10, EXPONENT_MAX / 10);
+        exponentB = bound(exponentB, EXPONENT_MIN / 10, EXPONENT_MAX / 10);
+
+        // The min signed value cannot be negated directly so we can't test it
+        // in this function.
+        vm.assume(signedCoefficientB != type(int256).min);
+
+        (int256 signedCoefficient, int256 exponent) =
+            LibDecimalFloat.sub(signedCoefficientA, exponentA, signedCoefficientB, exponentB);
+        (int256 expectedSignedCoefficient, int256 expectedExponent) =
+            LibDecimalFloat.add(signedCoefficientA, exponentA, -signedCoefficientB, exponentB);
+        assertEq(signedCoefficient, expectedSignedCoefficient);
+        assertEq(exponent, expectedExponent);
+    }
+
+    /// We can sub the min signed value as it will be normalized.
+    function testSubMinSignedValue(int256 signedCoefficientA, int256 exponentA, int256 exponentB) external pure {
+        exponentA = bound(exponentA, EXPONENT_MIN / 10, EXPONENT_MAX / 10);
+        exponentB = bound(exponentB, EXPONENT_MIN / 10, EXPONENT_MAX / 10);
+
+        // Able to sub the non-normalized min signed value.
+        int256 signedCoefficientB = type(int256).min;
+        (int256 signedCoefficient, int256 exponent) =
+            LibDecimalFloat.sub(signedCoefficientA, exponentA, signedCoefficientB, exponentB);
+
+        // Which behaves the same as normalizing then adding its negation.
+        (signedCoefficientB, exponentB) = LibDecimalFloatImplementation.normalize(signedCoefficientB, exponentB);
+        (int256 expectedSignedCoefficient, int256 expectedExponent) =
+            LibDecimalFloat.add(signedCoefficientA, exponentA, -signedCoefficientB, exponentB);
+
+        assertEq(signedCoefficient, expectedSignedCoefficient);
+        assertEq(exponent, expectedExponent);
+    }
+}