diff --git a/pcsx2/EeFpuFormat.h b/pcsx2/EeFpuFormat.h
index be0c1f8160..6e41107fc4 100644
--- a/pcsx2/EeFpuFormat.h
+++ b/pcsx2/EeFpuFormat.h
@@ -58,7 +58,7 @@ __fi static u32 eeFprNarrow(double stored)
 }
 
 /*	Whether the FPR file is in this domain right now. It follows the EE FPU's
-	clamp mode: only the mode-3 recompiler computes here, and modes 0-2 leave the
+	clamp mode: only iFPUd computes here, and modes 0 to 2 leave the
 	architectural word in the slot's low half. Read it through
 	FPRreg::Word()/SetWord().
 */
diff --git a/pcsx2/R5900.h b/pcsx2/R5900.h
index ecff6146eb..3e45753f95 100644
--- a/pcsx2/R5900.h
+++ b/pcsx2/R5900.h
@@ -151,11 +151,10 @@ union GPR_reg64 {
 	s8  SC[8];
 };
 
-/*	One EE FPR. The architectural register is 32 bits and the slot is 64, so the
-	word is not the slot and there is no u32 view of it: it is read through
-	Word() and written through SetWord(), and how a slot encodes it is those two
-	functions' business. In eeClampMode 3 it is the relocation of EeFpuFormat.h,
-	and otherwise the low half.
+/*	One EE FPR. The architectural register is 32 bits and the slot is 64: read
+	the word through Word(), write it through SetWord(). From eeClampMode 3 up a
+	slot holds the relocation of EeFpuFormat.h, below that the word in its low
+	half.
 */
 union FPRreg {
 	double d;
diff --git a/pcsx2/arm64/AsmHelpers.h b/pcsx2/arm64/AsmHelpers.h
index c5fd878aed..cba28964ea 100644
--- a/pcsx2/arm64/AsmHelpers.h
+++ b/pcsx2/arm64/AsmHelpers.h
@@ -301,7 +301,7 @@ __fi static void armEmitEeFprStoreSlotWord(const vixl::aarch64::MemOperand& slot
 }
 
 // Bridge for emitters holding a result as the architectural single in an S
-// register while the slot they store it to holds the stored double: the mode-3
+// register while the slot they store it to holds the stored double: iFPUd's
 // bodies, the fast path's SQRT, and LWC1's fastmem load. Clobbers `tmp`.
 __fi static void armEmitEeFprFromS(const vixl::aarch64::VRegister& slot,
 	const vixl::aarch64::VRegister& src, const vixl::aarch64::Register& tmp)
diff --git a/pcsx2/arm64/iFPU-arm64.cpp b/pcsx2/arm64/iFPU-arm64.cpp
index 76159ab16c..80792433df 100644
--- a/pcsx2/arm64/iFPU-arm64.cpp
+++ b/pcsx2/arm64/iFPU-arm64.cpp
@@ -633,7 +633,8 @@ static void emitFpuMul(const a64::VRegister& dst, const a64::VRegister& s, const
 //------------------------------------------------------------------
 
 // "Full" / DOUBLE-precision emitters (iFPUd-arm64.cpp), selected per-op when
-// CHECK_FPU_FULL (GameDB eeClampMode:3). Default config uses the fast paths.
+// CHECK_FPU_FULL (GameDB eeClampMode 3 and up). Default config uses the fast
+// paths.
 // MOV.S is a raw bit-copy in BOTH modes (x86 DOUBLE::recMOV_S_xmm == the fast
 // body), so it has no DOUBLE selection.
 namespace DOUBLE {
@@ -1459,7 +1460,8 @@ void recMSUBA_S()
 // CVT.S: fd = (float)int_bits_of(fpr[fs])
 // Single NEON-scalar SCVTF Sd,Sn — the int32 bits are already in the V file;
 // the old Fmov-to-GPR bounce cost an extra insn + cross-file hazard (GE-02).
-// A relocated slot puts them back out of reach, so mode 3 pays the bounce.
+// A relocated slot puts them back out of reach, so the double tier pays the
+// bounce.
 static void recCVT_S_xmm(int info)
 {
 	if (!CHECK_FPU_FULL)
@@ -1485,8 +1487,8 @@ void recCVT_S()
 // NaN by sign — positive NaN → 0x7fffffff, negative NaN → 0x80000000. Fix up
 // the NaN case only (cold branch over the source-sign select).
 //
-// A relocated slot holds no NaN, so mode 3 unscales into the value and converts,
-// and Fcvtzs's own saturation covers it.
+// A relocated slot holds no NaN, so the double tier unscales into the value and
+// converts, and Fcvtzs's own saturation covers it.
 static void recCVT_W_xmm(int info)
 {
 	if (CHECK_FPU_FULL)
diff --git a/pcsx2/arm64/iFPUd-arm64.cpp b/pcsx2/arm64/iFPUd-arm64.cpp
index 5232e9934e..f75190a0f8 100644
--- a/pcsx2/arm64/iFPUd-arm64.cpp
+++ b/pcsx2/arm64/iFPUd-arm64.cpp
@@ -7,7 +7,7 @@
 // widens each single to IEEE double, performs the op in double, then narrows
 // back to a PS2 single with the hardware's overflow/underflow/clamp semantics.
 // It is selected when CHECK_FPU_FULL (EmuConfig.Cpu.Recompiler.fpuFullMode, the
-// GameDB `eeClampMode:3` path — FFX, Max Payne, Dark Cloud 2, Klonoa 2 …).
+// GameDB eeClampMode 3 and up — FFX, Max Payne, Dark Cloud 2, Klonoa 2 …).
 // Default config runs the single-precision fast path in iFPU-arm64.cpp.
 //
 // It serves eeClampMode 3 and 4, which differ at emitDefectiveFmul and at
diff --git a/pcsx2/arm64/iR5900-arm64.cpp b/pcsx2/arm64/iR5900-arm64.cpp
index c1f5b69fbc..06ca5f2665 100644
--- a/pcsx2/arm64/iR5900-arm64.cpp
+++ b/pcsx2/arm64/iR5900-arm64.cpp
@@ -323,13 +323,8 @@ static const void* _DynGen_EnterRecompiledCode()
 	armAsm->Ldr(a64::s9, -FLT_MAX);
 
 	// Same convention, three registers along: d11 = 2^kEeFprScaleExp, which
-	// turns a slot into the value it denotes (NEON_RESERVED_EEFPU_UNSCALE; the
-	// contract is on the constant, the consumers are in iFPUd-arm64.cpp).
-	// Parking it here is what makes a mode-3 widening one instruction instead
-	// of eleven — every consumer reads it, none materializes it, and because
-	// the low 64 bits of d8-d15 are callee-saved there is no C-call seam,
-	// branch fork, superblock side exit or backpatched fastmem thunk that can
-	// invalidate it.
+	// turns a slot into the value it denotes. The contract is on
+	// NEON_RESERVED_EEFPU_UNSCALE; iFPUd-arm64.cpp is the consumer.
 	//
 	// Emitted unconditionally rather than under CHECK_FPU_FULL: two
 	// instructions once per JIT entry are not worth a dispatcher that goes
@@ -533,8 +528,8 @@ void iFlushCall(int flushtype)
 	// GE-15: FPR-class slots (NEONTYPE_FPREG/FPACC) in the callee-saved
 	// q10-q15 range survive plain C-helper seams — AAPCS64 preserves the
 	// LOWER 64 bits of v8-v15, and this class only ever reads/writes lane 0,
-	// at S width or (eeClampMode 3, where the slot is a relocated double) D
-	// width. Both fit inside the preserved half, and _writebackNEONreg stores
+	// at S width, or at D width where eeClampMode 3 and up put a relocated
+	// double in the slot. Both fit the preserved half, and _writebackNEONreg stores
 	// the same width it filled, so post-call garbage above it is never
 	// observed. Writeback if
 	// dirty but KEEP mapped — 4248's writeback-dirty-but-keep passes, whose
diff --git a/pcsx2/arm64/recVTLB-arm64.cpp b/pcsx2/arm64/recVTLB-arm64.cpp
index 164d35f6ed..932af6e2a1 100644
--- a/pcsx2/arm64/recVTLB-arm64.cpp
+++ b/pcsx2/arm64/recVTLB-arm64.cpp
@@ -1049,10 +1049,10 @@ void recLWC1()
 		vtlbFastmemReadFPR32(9, ftreg);
 		if (CHECK_FPU_FULL)
 		{
-			// The load lands the architectural word; in mode 3 the slot holds
-			// it relocated. Emitted after the backpatch site, so it runs on the
-			// thunk's return path too (the slow path ends in the same
-			// Fmov S<n>, w0).
+			// The load lands the architectural word; on the double tier the
+			// slot holds it relocated. Emitted after the backpatch site so it
+			// runs on the thunk's return path too (the slow path ends in the
+			// same Fmov S<n>, w0).
 			armEmitEeFprFromS(armDRegister(ftreg), armSRegister(ftreg), RXSCRATCH);
 		}
 	}