diff --git a/core/state/database_history.go b/core/state/database_history.go
index fbf4ab5f9c..170f79da93 100644
--- a/core/state/database_history.go
+++ b/core/state/database_history.go
@@ -62,7 +62,7 @@ func (r *historicStateReader) Account(addr common.Address) (*types.StateAccount,
 		Root:     common.BytesToHash(account.Root),
 	}
 	if len(acct.CodeHash) == 0 {
-		acct.CodeHash = types.EmptyCodeHash.Bytes()
+		acct.CodeHash = types.EmptyCodeHash[:]
 	}
 	if acct.Root == (common.Hash{}) {
 		acct.Root = types.EmptyRootHash
diff --git a/core/state/reader.go b/core/state/reader.go
index be07cec0f9..2163da8a9f 100644
--- a/core/state/reader.go
+++ b/core/state/reader.go
@@ -112,7 +112,7 @@ func (r *flatReader) Account(addr common.Address) (*types.StateAccount, error) {
 		Root:     common.BytesToHash(account.Root),
 	}
 	if len(acct.CodeHash) == 0 {
-		acct.CodeHash = types.EmptyCodeHash.Bytes()
+		acct.CodeHash = types.EmptyCodeHash[:]
 	}
 	if acct.Root == (common.Hash{}) {
 		acct.Root = types.EmptyRootHash
diff --git a/core/state/reader_test.go b/core/state/reader_test.go
new file mode 100644
index 0000000000..7a38971f28
--- /dev/null
+++ b/core/state/reader_test.go
@@ -0,0 +1,93 @@
+// Copyright 2026 The go-ethereum Authors
+// This file is part of the go-ethereum library.
+//
+// The go-ethereum library is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Lesser General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// The go-ethereum library is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Lesser General Public License for more details.
+//
+// You should have received a copy of the GNU Lesser General Public License
+// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
+
+package state
+
+import (
+	"testing"
+
+	"github.com/ethereum/go-ethereum/common"
+	"github.com/ethereum/go-ethereum/core/types"
+	"github.com/holiman/uint256"
+)
+
+// benchStateReader is a minimal database.StateReader backed by an in-memory map.
+// It isolates the conversion overhead in flatReader.Account (address hashing,
+// the *types.StateAccount allocation and the slim->full fixups) from any trie,
+// snapshot or disk access.
+type benchStateReader struct {
+	account *types.SlimAccount
+}
+
+func (r *benchStateReader) Account(hash common.Hash) (*types.SlimAccount, error) {
+	// Return a fresh copy on every call, mirroring the contract that the
+	// returned account is safe to modify by the caller.
+	if r.account == nil {
+		return nil, nil
+	}
+	cpy := *r.account
+	return &cpy, nil
+}
+
+func (r *benchStateReader) Storage(accountHash, storageHash common.Hash) ([]byte, error) {
+	return nil, nil
+}
+
+// benchmarkFlatReaderAccount measures flatReader.Account for a single address.
+// The provided slim account dictates which branches of the conversion are hit:
+// a slim account with nil Root/CodeHash exercises the EmptyCodeHash.Bytes() and
+// EmptyRootHash fixups (the lines that dominate the production profile), whereas
+// a fully populated one skips them.
+func benchmarkFlatReaderAccount(b *testing.B, slim *types.SlimAccount) {
+	r := newFlatReader(&benchStateReader{account: slim})
+	addr := common.HexToAddress("0x1234567890123456789012345678901234567890")
+
+	b.ReportAllocs()
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		acct, err := r.Account(addr)
+		if err != nil {
+			b.Fatal(err)
+		}
+		if acct == nil {
+			b.Fatal("unexpected nil account")
+		}
+	}
+}
+
+// BenchmarkFlatReaderAccountEmpty benchmarks the common EOA case: an account
+// with no code and no storage. This is the hot path from the profile, hitting
+// both the EmptyCodeHash and EmptyRootHash fixups.
+func BenchmarkFlatReaderAccountEmpty(b *testing.B) {
+	benchmarkFlatReaderAccount(b, &types.SlimAccount{
+		Nonce:   1,
+		Balance: uint256.NewInt(100),
+		// Root and CodeHash left nil: slim encoding of an EOA.
+	})
+}
+
+// BenchmarkFlatReaderAccountContract benchmarks a contract account with a
+// non-empty storage root and code hash, skipping the empty-value fixups.
+func BenchmarkFlatReaderAccountContract(b *testing.B) {
+	root := common.HexToHash("0xaabbccddeeff00112233445566778899aabbccddeeff00112233445566778899")
+	codeHash := common.HexToHash("0x112233445566778899aabbccddeeff00112233445566778899aabbccddeeff00")
+	benchmarkFlatReaderAccount(b, &types.SlimAccount{
+		Nonce:    7,
+		Balance:  uint256.NewInt(1000),
+		Root:     root.Bytes(),
+		CodeHash: codeHash.Bytes(),
+	})
+}
diff --git a/triedb/pathdb/buffer.go b/triedb/pathdb/buffer.go
index 5d3099285f..7afc1826c7 100644
--- a/triedb/pathdb/buffer.go
+++ b/triedb/pathdb/buffer.go
@@ -24,6 +24,7 @@ import (
 	"github.com/VictoriaMetrics/fastcache"
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/core/rawdb"
+	"github.com/ethereum/go-ethereum/core/types"
 	"github.com/ethereum/go-ethereum/ethdb"
 	"github.com/ethereum/go-ethereum/log"
 	"github.com/ethereum/go-ethereum/trie/trienode"
@@ -64,8 +65,9 @@ func newBuffer(limit int, nodes *nodeSet, states *stateSet, layers uint64) *buff
 	}
 }
 
-// account retrieves the account blob with account address hash.
-func (b *buffer) account(hash common.Hash) ([]byte, bool) {
+// account retrieves the decoded slim account with account address hash. A nil
+// account with found=true denotes a known-deleted account.
+func (b *buffer) account(hash common.Hash) (*types.SlimAccount, bool) {
 	return b.states.account(hash)
 }
 
diff --git a/triedb/pathdb/database.go b/triedb/pathdb/database.go
index e52949c93e..ca31647997 100644
--- a/triedb/pathdb/database.go
+++ b/triedb/pathdb/database.go
@@ -56,6 +56,16 @@ type layer interface {
 	// - no error will be returned if the requested account is not found in database.
 	account(hash common.Hash, depth int) ([]byte, error)
 
+	// accountObject directly retrieves the decoded slim account associated with
+	// a particular hash. It mirrors account but avoids re-encoding/decoding for
+	// the diff-layer hot read path, where accounts are already held in decoded
+	// form. An error will be returned if the read operation exits abnormally.
+	//
+	// Note:
+	// - the returned account is not a copy, please don't modify it.
+	// - a nil account is returned if the requested account is not found or was deleted.
+	accountObject(hash common.Hash, depth int) (*types.SlimAccount, error)
+
 	// storage directly retrieves the storage data associated with a particular hash,
 	// within a particular account. An error will be returned if the read operation
 	// exits abnormally. Specifically, if the layer is already stale.
diff --git a/triedb/pathdb/difflayer.go b/triedb/pathdb/difflayer.go
index 8ca3a39cf3..f92b380118 100644
--- a/triedb/pathdb/difflayer.go
+++ b/triedb/pathdb/difflayer.go
@@ -21,6 +21,7 @@ import (
 	"sync"
 
 	"github.com/ethereum/go-ethereum/common"
+	"github.com/ethereum/go-ethereum/core/types"
 	"github.com/ethereum/go-ethereum/log"
 )
 
@@ -102,25 +103,41 @@ func (dl *diffLayer) node(owner common.Hash, path []byte, depth int) ([]byte, co
 //
 // Note the returned account is not a copy, please don't modify it.
 func (dl *diffLayer) account(hash common.Hash, depth int) ([]byte, error) {
+	account, err := dl.accountObject(hash, depth)
+	if err != nil {
+		return nil, err
+	}
+	// Re-encode into slim-RLP form for the byte-oriented callers (e.g. the
+	// binary iterator and the public AccountRLP). The hot read path uses
+	// accountObject directly and avoids this encode.
+	return encodeSlimAccount(account), nil
+}
+
+// accountObject directly retrieves the decoded slim account associated with a
+// particular hash. A nil account is returned if the account does not exist or
+// was deleted in this hierarchy.
+//
+// Note the returned account is not a copy, please don't modify it.
+func (dl *diffLayer) accountObject(hash common.Hash, depth int) (*types.SlimAccount, error) {
 	// Hold the lock, ensure the parent won't be changed during the
 	// state accessing.
 	dl.lock.RLock()
 	defer dl.lock.RUnlock()
 
-	if blob, found := dl.states.account(hash); found {
+	if account, found := dl.states.account(hash); found {
 		dirtyStateHitMeter.Mark(1)
 		dirtyStateHitDepthHist.Update(int64(depth))
-		dirtyStateReadMeter.Mark(int64(len(blob)))
+		dirtyStateReadMeter.Mark(int64(slimAccountSize(account)))
 
-		if len(blob) == 0 {
+		if account == nil {
 			stateAccountInexMeter.Mark(1)
 		} else {
 			stateAccountExistMeter.Mark(1)
 		}
-		return blob, nil
+		return account, nil
 	}
 	// Account is unknown to this layer, resolve from parent
-	return dl.parent.account(hash, depth+1)
+	return dl.parent.accountObject(hash, depth+1)
 }
 
 // storage directly retrieves the storage data associated with a particular hash,
diff --git a/triedb/pathdb/disklayer.go b/triedb/pathdb/disklayer.go
index 8c0a751932..27a73af53b 100644
--- a/triedb/pathdb/disklayer.go
+++ b/triedb/pathdb/disklayer.go
@@ -25,6 +25,7 @@ import (
 	"github.com/VictoriaMetrics/fastcache"
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/core/rawdb"
+	"github.com/ethereum/go-ethereum/core/types"
 	"github.com/ethereum/go-ethereum/crypto"
 	"github.com/ethereum/go-ethereum/ethdb"
 	"github.com/ethereum/go-ethereum/log"
@@ -169,29 +170,42 @@ func (dl *diskLayer) node(owner common.Hash, path []byte, depth int) ([]byte, co
 //
 // Note the returned account is not a copy, please don't modify it.
 func (dl *diskLayer) account(hash common.Hash, depth int) ([]byte, error) {
+	account, err := dl.accountObject(hash, depth)
+	if err != nil {
+		return nil, err
+	}
+	return encodeSlimAccount(account), nil
+}
+
+// accountObject directly retrieves the decoded slim account associated with a
+// particular hash. A nil account is returned if the account does not exist or
+// was deleted.
+func (dl *diskLayer) accountObject(hash common.Hash, depth int) (*types.SlimAccount, error) {
 	dl.lock.RLock()
 	defer dl.lock.RUnlock()
 
 	if dl.stale {
 		return nil, errSnapshotStale
 	}
-	// Try to retrieve the trie node from the not-yet-written node buffer first
-	// (both the live one and the frozen one). Note the buffer is lock free since
-	// it's impossible to mutate the buffer before tagging the layer as stale.
+	// Try to retrieve the account from the not-yet-written state buffer first
+	// (both the live one and the frozen one). The buffer holds accounts in
+	// decoded form, so no decoding is required here. Note the buffer is lock
+	// free since it's impossible to mutate the buffer before tagging the layer
+	// as stale.
 	for _, buffer := range []*buffer{dl.buffer, dl.frozen} {
 		if buffer != nil {
-			blob, found := buffer.account(hash)
+			account, found := buffer.account(hash)
 			if found {
 				dirtyStateHitMeter.Mark(1)
-				dirtyStateReadMeter.Mark(int64(len(blob)))
+				dirtyStateReadMeter.Mark(int64(slimAccountSize(account)))
 				dirtyStateHitDepthHist.Update(int64(depth))
 
-				if len(blob) == 0 {
+				if account == nil {
 					stateAccountInexMeter.Mark(1)
 				} else {
 					stateAccountExistMeter.Mark(1)
 				}
-				return blob, nil
+				return account, nil
 			}
 		}
 	}
@@ -214,7 +228,7 @@ func (dl *diskLayer) account(hash common.Hash, depth int) ([]byte, error) {
 			} else {
 				stateAccountExistMeter.Mark(1)
 			}
-			return blob, nil
+			return decodeAccountBlob(hash, blob), nil
 		}
 		cleanStateMissMeter.Mark(1)
 	}
@@ -237,7 +251,7 @@ func (dl *diskLayer) account(hash common.Hash, depth int) ([]byte, error) {
 		stateAccountExistMeter.Mark(1)
 		stateAccountExistDiskMeter.Mark(1)
 	}
-	return blob, nil
+	return decodeAccountBlob(hash, blob), nil
 }
 
 // storage directly retrieves the storage data associated with a particular hash,
@@ -598,8 +612,12 @@ func (dl *diskLayer) revert(h *stateHistory) (*diskLayer, error) {
 	batch := dl.db.diskdb.NewBatch()
 	writeNodes(batch, nodes, dl.nodes)
 
-	// Provide the original values of modified accounts and storages for revert
-	writeStates(batch, progress, accounts, storages, dl.states)
+	// Provide the original values of modified accounts and storages for revert.
+	// The history-derived accounts are in slim-RLP form; decode them so they
+	// share the single encode point in writeStates. This path is only hit on a
+	// deep reorg that rolls back the persistent layer, so the round-trip cost
+	// is irrelevant.
+	writeStates(batch, progress, decodeAccounts(accounts), storages, dl.states)
 	rawdb.WritePersistentStateID(batch, dl.id-1)
 	rawdb.WriteSnapshotRoot(batch, h.meta.parent)
 	if err := batch.Write(); err != nil {
diff --git a/triedb/pathdb/flush.go b/triedb/pathdb/flush.go
index 4f816cf6a6..8d7f791e3b 100644
--- a/triedb/pathdb/flush.go
+++ b/triedb/pathdb/flush.go
@@ -22,6 +22,7 @@ import (
 	"github.com/VictoriaMetrics/fastcache"
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/core/rawdb"
+	"github.com/ethereum/go-ethereum/core/types"
 	"github.com/ethereum/go-ethereum/ethdb"
 	"github.com/ethereum/go-ethereum/trie/trienode"
 )
@@ -74,12 +75,12 @@ func writeNodes(batch ethdb.Batch, nodes map[common.Hash]map[string]*trienode.No
 // TODO(rjl493456442) do we really need this generation marker? The state updates
 // after the marker can also be written and will be fixed by generator later if
 // it's outdated.
-func writeStates(batch ethdb.Batch, genMarker []byte, accountData map[common.Hash][]byte, storageData map[common.Hash]map[common.Hash][]byte, clean *fastcache.Cache) (int, int) {
+func writeStates(batch ethdb.Batch, genMarker []byte, accountData map[common.Hash]*types.SlimAccount, storageData map[common.Hash]map[common.Hash][]byte, clean *fastcache.Cache) (int, int) {
 	var (
 		accounts int
 		slots    int
 	)
-	for addrHash, blob := range accountData {
+	for addrHash, account := range accountData {
 		// Skip any account not yet covered by the snapshot. The account
 		// at the generation marker position (addrHash == genMarker[:common.HashLength])
 		// should still be updated, as it would be skipped in the next
@@ -88,12 +89,16 @@ func writeStates(batch ethdb.Batch, genMarker []byte, accountData map[common.Has
 			continue
 		}
 		accounts += 1
-		if len(blob) == 0 {
+		if account == nil {
 			rawdb.DeleteAccountSnapshot(batch, addrHash)
 			if clean != nil {
 				clean.Set(addrHash[:], nil)
 			}
 		} else {
+			// Encode the decoded account back into slim-RLP form for storage.
+			// This is the single point where the in-memory decoded accounts
+			// are serialized for the on-disk snapshot.
+			blob := encodeSlimAccount(account)
 			rawdb.WriteAccountSnapshot(batch, addrHash, blob)
 			if clean != nil {
 				clean.Set(addrHash[:], blob)
diff --git a/triedb/pathdb/generate_test.go b/triedb/pathdb/generate_test.go
index f38a1ed7c4..a3c13f7c42 100644
--- a/triedb/pathdb/generate_test.go
+++ b/triedb/pathdb/generate_test.go
@@ -69,7 +69,7 @@ func (t *genTester) addTrieAccount(acckey string, acc *types.StateAccount) {
 	)
 	t.acctTrie.MustUpdate(key.Bytes(), val)
 
-	t.states.accountData[key] = val
+	t.states.accountData[key] = decodeAccountBlob(key, types.SlimAccountRLP(*acc))
 	t.states.accountOrigin[addr] = nil
 }
 
diff --git a/triedb/pathdb/iterator_fast.go b/triedb/pathdb/iterator_fast.go
index 04bf89e874..0f8b931a1f 100644
--- a/triedb/pathdb/iterator_fast.go
+++ b/triedb/pathdb/iterator_fast.go
@@ -98,7 +98,11 @@ func newFastIterator(db *Database, root common.Hash, account common.Hash, seek c
 						if dl.stale {
 							return nil, errSnapshotStale
 						}
-						return dl.buffer.states.mustAccount(hash)
+						account, err := dl.buffer.states.mustAccount(hash)
+						if err != nil {
+							return nil, err
+						}
+						return encodeSlimAccount(account), nil
 					}),
 					priority: depth,
 				})
@@ -110,8 +114,15 @@ func newFastIterator(db *Database, root common.Hash, account common.Hash, seek c
 				// The state set in diff layer is immutable and will never be stale,
 				// so the read lock protection is unnecessary.
 				accountList := dl.states.accountList()
+				states := dl.states
 				fi.iterators = append(fi.iterators, &weightedIterator{
-					it:       newDiffAccountIterator(seek, accountList, dl.states.mustAccount),
+					it: newDiffAccountIterator(seek, accountList, func(hash common.Hash) ([]byte, error) {
+						account, err := states.mustAccount(hash)
+						if err != nil {
+							return nil, err
+						}
+						return encodeSlimAccount(account), nil
+					}),
 					priority: depth,
 				})
 			}
diff --git a/triedb/pathdb/iterator_test.go b/triedb/pathdb/iterator_test.go
index 191c2fadf5..3bce1989ea 100644
--- a/triedb/pathdb/iterator_test.go
+++ b/triedb/pathdb/iterator_test.go
@@ -122,8 +122,7 @@ func TestAccountIteratorBasics(t *testing.T) {
 	// Fill up a parent
 	for i := 0; i < 100; i++ {
 		hash := testrand.Hash()
-		data := testrand.Bytes(32)
-		accounts[hash] = data
+		accounts[hash] = randomAccount()
 
 		if rand.Intn(2) == 0 {
 			accStorage := make(map[common.Hash][]byte)
@@ -153,7 +152,7 @@ func TestStorageIteratorBasics(t *testing.T) {
 	// Fill some random data
 	for i := 0; i < 10; i++ {
 		hash := testrand.Hash()
-		accounts[hash] = testrand.Bytes(32)
+		accounts[hash] = randomAccount()
 
 		accStorage := make(map[common.Hash][]byte)
 		var nilstorage int
@@ -344,6 +343,18 @@ func TestAccountIteratorTraversalValues(t *testing.T) {
 	}
 	db := New(rawdb.NewMemoryDatabase(), config, false)
 
+	// layerAccount returns a valid slim-RLP account whose nonce uniquely encodes
+	// the (layer, key) pair, so that values are distinct per layer and the
+	// iterator's "newest layer wins" behavior is exercised.
+	layerAccount := func(layer, key byte) []byte {
+		return types.SlimAccountRLP(types.StateAccount{
+			Nonce:    uint64(layer)<<8 | uint64(key),
+			Balance:  uint256.NewInt(uint64(key)),
+			Root:     types.EmptyRootHash,
+			CodeHash: types.EmptyCodeHash[:],
+		})
+	}
+
 	// Create a batch of account sets to seed subsequent layers with
 	var (
 		a = make(map[common.Hash][]byte)
@@ -356,27 +367,27 @@ func TestAccountIteratorTraversalValues(t *testing.T) {
 		h = make(map[common.Hash][]byte)
 	)
 	for i := byte(2); i < 0xff; i++ {
-		a[common.Hash{i}] = fmt.Appendf(nil, "layer-%d, key %d", 0, i)
+		a[common.Hash{i}] = layerAccount(0, i)
 		if i > 20 && i%2 == 0 {
-			b[common.Hash{i}] = fmt.Appendf(nil, "layer-%d, key %d", 1, i)
+			b[common.Hash{i}] = layerAccount(1, i)
 		}
 		if i%4 == 0 {
-			c[common.Hash{i}] = fmt.Appendf(nil, "layer-%d, key %d", 2, i)
+			c[common.Hash{i}] = layerAccount(2, i)
 		}
 		if i%7 == 0 {
-			d[common.Hash{i}] = fmt.Appendf(nil, "layer-%d, key %d", 3, i)
+			d[common.Hash{i}] = layerAccount(3, i)
 		}
 		if i%8 == 0 {
-			e[common.Hash{i}] = fmt.Appendf(nil, "layer-%d, key %d", 4, i)
+			e[common.Hash{i}] = layerAccount(4, i)
 		}
 		if i > 50 && i < 85 {
-			f[common.Hash{i}] = fmt.Appendf(nil, "layer-%d, key %d", 5, i)
+			f[common.Hash{i}] = layerAccount(5, i)
 		}
 		if i%64 == 0 {
-			g[common.Hash{i}] = fmt.Appendf(nil, "layer-%d, key %d", 6, i)
+			g[common.Hash{i}] = layerAccount(6, i)
 		}
 		if i%128 == 0 {
-			h[common.Hash{i}] = fmt.Appendf(nil, "layer-%d, key %d", 7, i)
+			h[common.Hash{i}] = layerAccount(7, i)
 		}
 	}
 	// Assemble a stack of snapshots from the account layers
diff --git a/triedb/pathdb/reader.go b/triedb/pathdb/reader.go
index e087ef26ed..9a006a2dbd 100644
--- a/triedb/pathdb/reader.go
+++ b/triedb/pathdb/reader.go
@@ -127,18 +127,27 @@ func (r *reader) AccountRLP(hash common.Hash) ([]byte, error) {
 // - the returned account object is safe to modify
 // - no error will be returned if the requested account is not found in database
 func (r *reader) Account(hash common.Hash) (*types.SlimAccount, error) {
-	blob, err := r.AccountRLP(hash)
+	l, err := r.db.tree.lookupAccount(hash, r.state)
 	if err != nil {
 		return nil, err
 	}
-	if len(blob) == 0 {
+	// If the located layer is stale, fall back to the slow path to retrieve
+	// the account. See AccountRLP for the rationale of this fallback.
+	account, err := l.accountObject(hash, 0)
+	if errors.Is(err, errSnapshotStale) {
+		account, err = r.layer.accountObject(hash, 0)
+	}
+	if err != nil {
+		return nil, err
+	}
+	if account == nil {
 		return nil, nil
 	}
-	account := new(types.SlimAccount)
-	if err := rlp.DecodeBytes(blob, account); err != nil {
-		panic(err)
-	}
-	return account, nil
+	// The diff layers retain the account in decoded form and the returned
+	// object is documented as safe to modify, so hand back a shallow copy to
+	// protect the cached instance from caller mutation.
+	cpy := *account
+	return &cpy, nil
 }
 
 // Storage directly retrieves the storage data associated with a particular hash,
diff --git a/triedb/pathdb/states.go b/triedb/pathdb/states.go
index 27a6c1d422..6c79e62bfb 100644
--- a/triedb/pathdb/states.go
+++ b/triedb/pathdb/states.go
@@ -26,6 +26,7 @@ import (
 	"github.com/VictoriaMetrics/fastcache"
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/core/rawdb"
+	"github.com/ethereum/go-ethereum/core/types"
 	"github.com/ethereum/go-ethereum/ethdb"
 	"github.com/ethereum/go-ethereum/log"
 	"github.com/ethereum/go-ethereum/metrics"
@@ -60,7 +61,7 @@ func (c *counter) report(count, size *metrics.Meter) {
 // subsequent state access will be denied due to the stale flag. Therefore,
 // state access and mutation won't happen at the same time with guarantee.
 type stateSet struct {
-	accountData map[common.Hash][]byte                 // Keyed accounts for direct retrieval (nil means deleted)
+	accountData map[common.Hash]*types.SlimAccount     // Keyed accounts for direct retrieval (nil means deleted)
 	storageData map[common.Hash]map[common.Hash][]byte // Keyed storage slots for direct retrieval. one per account (nil means deleted)
 	size        uint64                                 // Memory size of the state data (accountData and storageData)
 
@@ -75,17 +76,81 @@ type stateSet struct {
 	listLock sync.RWMutex
 }
 
+// slimAccountSize returns the approximate encoded size of a slim account, used
+// for memory accounting. It estimates the slim-RLP byte length without actually
+// encoding the account. A nil account (deleted marker) contributes zero.
+func slimAccountSize(account *types.SlimAccount) int {
+	if account == nil {
+		return 0
+	}
+	// 9 bytes covers the RLP list header plus the nonce; the balance, root and
+	// code hash contribute their own byte lengths. This intentionally mirrors
+	// the previous behavior of accounting for the encoded blob length.
+	size := 9
+	if account.Balance != nil {
+		size += account.Balance.ByteLen()
+	}
+	size += len(account.Root)
+	size += len(account.CodeHash)
+	return size
+}
+
+// decodeAccountBlob decodes a single slim-RLP-encoded account blob. A nil/empty
+// blob denotes a non-existent or deleted account and decodes to a nil account.
+func decodeAccountBlob(hash common.Hash, blob []byte) *types.SlimAccount {
+	if len(blob) == 0 {
+		return nil
+	}
+	account := new(types.SlimAccount)
+	if err := rlp.DecodeBytes(blob, account); err != nil {
+		panic(fmt.Sprintf("failed to decode account %x: %v", hash, err))
+	}
+	return account
+}
+
+// decodeAccounts converts a map of slim-RLP-encoded accounts into decoded slim
+// account objects. A nil/empty blob denotes a deleted account and is preserved
+// as a nil entry in the resulting map.
+func decodeAccounts(accounts map[common.Hash][]byte) map[common.Hash]*types.SlimAccount {
+	decoded := make(map[common.Hash]*types.SlimAccount, len(accounts))
+	for hash, blob := range accounts {
+		if len(blob) == 0 {
+			decoded[hash] = nil // deleted account
+			continue
+		}
+		account := new(types.SlimAccount)
+		if err := rlp.DecodeBytes(blob, account); err != nil {
+			panic(fmt.Sprintf("failed to decode account %x: %v", hash, err))
+		}
+		decoded[hash] = account
+	}
+	return decoded
+}
+
+// encodeSlimAccount serializes a decoded slim account back into its slim-RLP
+// byte form. A nil account (deleted marker) encodes to a nil blob.
+func encodeSlimAccount(account *types.SlimAccount) []byte {
+	if account == nil {
+		return nil
+	}
+	blob, err := rlp.EncodeToBytes(account)
+	if err != nil {
+		panic(fmt.Sprintf("failed to encode account: %v", err))
+	}
+	return blob
+}
+
 // newStates constructs the state set with the provided account and storage data.
+// The accounts are supplied in slim-RLP form and are decoded once on ingest;
+// they are kept in decoded form for the lifetime of the set and only re-encoded
+// at the serialization boundaries (journal and disk flush).
 func newStates(accounts map[common.Hash][]byte, storages map[common.Hash]map[common.Hash][]byte, rawStorageKey bool) *stateSet {
 	// Don't panic for the lazy callers, initialize the nil maps instead.
-	if accounts == nil {
-		accounts = make(map[common.Hash][]byte)
-	}
 	if storages == nil {
 		storages = make(map[common.Hash]map[common.Hash][]byte)
 	}
 	s := &stateSet{
-		accountData:       accounts,
+		accountData:       decodeAccounts(accounts),
 		storageData:       storages,
 		rawStorageKey:     rawStorageKey,
 		storageListSorted: make(map[common.Hash][]common.Hash),
@@ -95,7 +160,9 @@ func newStates(accounts map[common.Hash][]byte, storages map[common.Hash]map[com
 }
 
 // account returns the account data associated with the specified address hash.
-func (s *stateSet) account(hash common.Hash) ([]byte, bool) {
+// A nil account with found=true denotes a known-deleted account, whereas
+// found=false indicates the account is unknown in this set.
+func (s *stateSet) account(hash common.Hash) (*types.SlimAccount, bool) {
 	// If the account is known locally, return it
 	if data, ok := s.accountData[hash]; ok {
 		return data, true
@@ -106,7 +173,7 @@ func (s *stateSet) account(hash common.Hash) ([]byte, bool) {
 // mustAccount returns the account data associated with the specified address
 // hash. The difference is this function will return an error if the account
 // is not found.
-func (s *stateSet) mustAccount(hash common.Hash) ([]byte, error) {
+func (s *stateSet) mustAccount(hash common.Hash) (*types.SlimAccount, error) {
 	// If the account is known locally, return it
 	if data, ok := s.accountData[hash]; ok {
 		return data, nil
@@ -143,8 +210,8 @@ func (s *stateSet) mustStorage(accountHash, storageHash common.Hash) ([]byte, er
 // Additionally, it computes the total memory size occupied by the maps.
 func (s *stateSet) check() uint64 {
 	var size int
-	for _, blob := range s.accountData {
-		size += common.HashLength + len(blob)
+	for _, account := range s.accountData {
+		size += common.HashLength + slimAccountSize(account)
 	}
 	for accountHash, slots := range s.storageData {
 		if slots == nil {
@@ -235,11 +302,13 @@ func (s *stateSet) merge(other *stateSet) {
 	)
 	// Apply the updated account data
 	for accountHash, data := range other.accountData {
+		dataSize := slimAccountSize(data)
 		if origin, ok := s.accountData[accountHash]; ok {
-			delta += len(data) - len(origin)
-			accountOverwrites.add(common.HashLength + len(origin))
+			originSize := slimAccountSize(origin)
+			delta += dataSize - originSize
+			accountOverwrites.add(common.HashLength + originSize)
 		} else {
-			delta += common.HashLength + len(data)
+			delta += common.HashLength + dataSize
 		}
 		s.accountData[accountHash] = data
 	}
@@ -293,11 +362,20 @@ func (s *stateSet) revertTo(accountOrigin map[common.Hash][]byte, storageOrigin
 		if !ok {
 			panic(fmt.Sprintf("non-existent account for reverting, %x", addrHash))
 		}
-		if len(data) == 0 && len(blob) == 0 {
+		// Decode the original slim-RLP value; an empty blob denotes a deleted
+		// account, represented as a nil entry.
+		var origin *types.SlimAccount
+		if len(blob) != 0 {
+			origin = new(types.SlimAccount)
+			if err := rlp.DecodeBytes(blob, origin); err != nil {
+				panic(fmt.Sprintf("failed to decode account %x: %v", addrHash, err))
+			}
+		}
+		if data == nil && origin == nil {
 			panic(fmt.Sprintf("invalid account mutation (null to null), %x", addrHash))
 		}
-		delta += len(blob) - len(data)
-		s.accountData[addrHash] = blob
+		delta += slimAccountSize(origin) - slimAccountSize(data)
+		s.accountData[addrHash] = origin
 	}
 	// Overwrite the storage data with original value blindly
 	for addrHash, storage := range storageOrigin {
@@ -346,9 +424,9 @@ func (s *stateSet) encode(w io.Writer) error {
 		AddrHashes: make([]common.Hash, 0, len(s.accountData)),
 		Accounts:   make([][]byte, 0, len(s.accountData)),
 	}
-	for addrHash, blob := range s.accountData {
+	for addrHash, account := range s.accountData {
 		enc.AddrHashes = append(enc.AddrHashes, addrHash)
-		enc.Accounts = append(enc.Accounts, blob)
+		enc.Accounts = append(enc.Accounts, encodeSlimAccount(account))
 	}
 	if err := rlp.Encode(w, enc); err != nil {
 		return err
@@ -395,7 +473,7 @@ func (s *stateSet) decode(r *rlp.Stream) error {
 	for i := range dec.AddrHashes {
 		accountSet[dec.AddrHashes[i]] = empty2nil(dec.Accounts[i])
 	}
-	s.accountData = accountSet
+	s.accountData = decodeAccounts(accountSet)
 
 	// Decode storages
 	type storage struct {
@@ -431,7 +509,7 @@ func (s *stateSet) write(batch ethdb.Batch, genMarker []byte, clean *fastcache.C
 // reset clears all cached state data, including any optional sorted lists that
 // may have been generated.
 func (s *stateSet) reset() {
-	s.accountData = make(map[common.Hash][]byte)
+	s.accountData = make(map[common.Hash]*types.SlimAccount)
 	s.storageData = make(map[common.Hash]map[common.Hash][]byte)
 	s.size = 0
 	s.accountListSorted = nil
diff --git a/triedb/pathdb/states_test.go b/triedb/pathdb/states_test.go
index 4d181cc914..ff1a1078e0 100644
--- a/triedb/pathdb/states_test.go
+++ b/triedb/pathdb/states_test.go
@@ -22,15 +22,39 @@ import (
 	"testing"
 
 	"github.com/ethereum/go-ethereum/common"
+	"github.com/ethereum/go-ethereum/core/types"
+	"github.com/ethereum/go-ethereum/internal/testrand"
 	"github.com/ethereum/go-ethereum/rlp"
+	"github.com/holiman/uint256"
 )
 
+// testAccountRLP returns the slim-RLP encoding of a deterministic account whose
+// nonce is derived from the provided tag. It lets the state-set tests use real,
+// decodable accounts while keeping the tag values readable.
+func testAccountRLP(tag byte) []byte {
+	return types.SlimAccountRLP(types.StateAccount{
+		Nonce:    uint64(tag),
+		Balance:  uint256.NewInt(uint64(tag)),
+		Root:     types.EmptyRootHash,
+		CodeHash: types.EmptyCodeHash[:],
+	})
+}
+
+// accountEqualsTag reports whether the decoded account matches the account
+// produced by testAccountRLP for the given tag.
+func accountEqualsTag(account *types.SlimAccount, tag byte) bool {
+	if account == nil {
+		return false
+	}
+	return bytes.Equal(encodeSlimAccount(account), testAccountRLP(tag))
+}
+
 func TestStatesMerge(t *testing.T) {
 	a := newStates(
 		map[common.Hash][]byte{
-			{0xa}: {0xa0},
-			{0xb}: {0xb0},
-			{0xc}: {0xc0},
+			{0xa}: testAccountRLP(0xa0),
+			{0xb}: testAccountRLP(0xb0),
+			{0xc}: testAccountRLP(0xc0),
 		},
 		map[common.Hash]map[common.Hash][]byte{
 			{0xa}: {
@@ -48,8 +72,8 @@ func TestStatesMerge(t *testing.T) {
 	)
 	b := newStates(
 		map[common.Hash][]byte{
-			{0xa}: {0xa1},
-			{0xb}: {0xb1},
+			{0xa}: testAccountRLP(0xa1),
+			{0xb}: testAccountRLP(0xb1),
 			{0xc}: nil, // delete account
 		},
 		map[common.Hash]map[common.Hash][]byte{
@@ -69,25 +93,25 @@ func TestStatesMerge(t *testing.T) {
 	)
 	a.merge(b)
 
-	blob, exist := a.account(common.Hash{0xa})
-	if !exist || !bytes.Equal(blob, []byte{0xa1}) {
+	account, exist := a.account(common.Hash{0xa})
+	if !exist || !accountEqualsTag(account, 0xa1) {
 		t.Error("Unexpected value for account a")
 	}
-	blob, exist = a.account(common.Hash{0xb})
-	if !exist || !bytes.Equal(blob, []byte{0xb1}) {
+	account, exist = a.account(common.Hash{0xb})
+	if !exist || !accountEqualsTag(account, 0xb1) {
 		t.Error("Unexpected value for account b")
 	}
-	blob, exist = a.account(common.Hash{0xc})
-	if !exist || len(blob) != 0 {
+	account, exist = a.account(common.Hash{0xc})
+	if !exist || account != nil {
 		t.Error("Unexpected value for account c")
 	}
 	// unknown account
-	blob, exist = a.account(common.Hash{0xd})
-	if exist || len(blob) != 0 {
+	account, exist = a.account(common.Hash{0xd})
+	if exist || account != nil {
 		t.Error("Unexpected value for account d")
 	}
 
-	blob, exist = a.storage(common.Hash{0xa}, common.Hash{0x1})
+	blob, exist := a.storage(common.Hash{0xa}, common.Hash{0x1})
 	if !exist || !bytes.Equal(blob, []byte{0x11}) {
 		t.Error("Unexpected value for a's storage")
 	}
@@ -118,9 +142,9 @@ func TestStatesMerge(t *testing.T) {
 func TestStatesRevert(t *testing.T) {
 	a := newStates(
 		map[common.Hash][]byte{
-			{0xa}: {0xa0},
-			{0xb}: {0xb0},
-			{0xc}: {0xc0},
+			{0xa}: testAccountRLP(0xa0),
+			{0xb}: testAccountRLP(0xb0),
+			{0xc}: testAccountRLP(0xc0),
 		},
 		map[common.Hash]map[common.Hash][]byte{
 			{0xa}: {
@@ -138,8 +162,8 @@ func TestStatesRevert(t *testing.T) {
 	)
 	b := newStates(
 		map[common.Hash][]byte{
-			{0xa}: {0xa1},
-			{0xb}: {0xb1},
+			{0xa}: testAccountRLP(0xa1),
+			{0xb}: testAccountRLP(0xb1),
 			{0xc}: nil,
 		},
 		map[common.Hash]map[common.Hash][]byte{
@@ -160,9 +184,9 @@ func TestStatesRevert(t *testing.T) {
 	a.merge(b)
 	a.revertTo(
 		map[common.Hash][]byte{
-			{0xa}: {0xa0},
-			{0xb}: {0xb0},
-			{0xc}: {0xc0},
+			{0xa}: testAccountRLP(0xa0),
+			{0xb}: testAccountRLP(0xb0),
+			{0xc}: testAccountRLP(0xc0),
 		},
 		map[common.Hash]map[common.Hash][]byte{
 			{0xa}: {
@@ -179,25 +203,25 @@ func TestStatesRevert(t *testing.T) {
 		},
 	)
 
-	blob, exist := a.account(common.Hash{0xa})
-	if !exist || !bytes.Equal(blob, []byte{0xa0}) {
+	account, exist := a.account(common.Hash{0xa})
+	if !exist || !accountEqualsTag(account, 0xa0) {
 		t.Error("Unexpected value for account a")
 	}
-	blob, exist = a.account(common.Hash{0xb})
-	if !exist || !bytes.Equal(blob, []byte{0xb0}) {
+	account, exist = a.account(common.Hash{0xb})
+	if !exist || !accountEqualsTag(account, 0xb0) {
 		t.Error("Unexpected value for account b")
 	}
-	blob, exist = a.account(common.Hash{0xc})
-	if !exist || !bytes.Equal(blob, []byte{0xc0}) {
+	account, exist = a.account(common.Hash{0xc})
+	if !exist || !accountEqualsTag(account, 0xc0) {
 		t.Error("Unexpected value for account c")
 	}
 	// unknown account
-	blob, exist = a.account(common.Hash{0xd})
-	if exist || len(blob) != 0 {
+	account, exist = a.account(common.Hash{0xd})
+	if exist || account != nil {
 		t.Error("Unexpected value for account d")
 	}
 
-	blob, exist = a.storage(common.Hash{0xa}, common.Hash{0x1})
+	blob, exist := a.storage(common.Hash{0xa}, common.Hash{0x1})
 	if !exist || !bytes.Equal(blob, []byte{0x10}) {
 		t.Error("Unexpected value for a's storage")
 	}
@@ -231,7 +255,7 @@ func TestStateRevertAccountNullMarker(t *testing.T) {
 	a := newStates(nil, nil, false) // empty initial state
 	b := newStates(
 		map[common.Hash][]byte{
-			{0xa}: {0xa},
+			{0xa}: testAccountRLP(0xa),
 		},
 		nil,
 		false,
@@ -244,12 +268,12 @@ func TestStateRevertAccountNullMarker(t *testing.T) {
 		nil,
 	) // revert the transition b
 
-	blob, exist := a.account(common.Hash{0xa})
+	account, exist := a.account(common.Hash{0xa})
 	if !exist {
 		t.Fatal("null marker is not found")
 	}
-	if len(blob) != 0 {
-		t.Fatalf("Unexpected value for account, %v", blob)
+	if account != nil {
+		t.Fatalf("Unexpected value for account, %v", account)
 	}
 }
 
@@ -258,7 +282,7 @@ func TestStateRevertAccountNullMarker(t *testing.T) {
 // entry in the set.
 func TestStateRevertStorageNullMarker(t *testing.T) {
 	a := newStates(map[common.Hash][]byte{
-		{0xa}: {0xa},
+		{0xa}: testAccountRLP(0xa),
 	}, nil, false) // initial state with account 0xa
 
 	b := newStates(
@@ -297,7 +321,7 @@ func TestStatesEncode(t *testing.T) {
 func testStatesEncode(t *testing.T, rawStorageKey bool) {
 	s := newStates(
 		map[common.Hash][]byte{
-			{0x1}: {0x1},
+			{0x1}: testAccountRLP(0x1),
 		},
 		map[common.Hash]map[common.Hash][]byte{
 			{0x1}: {
@@ -333,7 +357,7 @@ func TestStateWithOriginEncode(t *testing.T) {
 func testStateWithOriginEncode(t *testing.T, rawStorageKey bool) {
 	s := NewStateSetWithOrigin(
 		map[common.Hash][]byte{
-			{0x1}: {0x1},
+			{0x1}: testAccountRLP(0x1),
 		},
 		map[common.Hash]map[common.Hash][]byte{
 			{0x1}: {
@@ -375,17 +399,53 @@ func testStateWithOriginEncode(t *testing.T, rawStorageKey bool) {
 	}
 }
 
+// TestSlimAccountSize verifies the estimated memory size of a slim account for
+// the empty/deleted, EOA and contract cases. The accountData map keys are
+// accounted for separately (common.HashLength per entry), so this only covers
+// the value contribution.
+func TestSlimAccountSize(t *testing.T) {
+	// A nil account (deletion marker) contributes nothing.
+	if got := slimAccountSize(nil); got != 0 {
+		t.Fatalf("nil account size, want: 0, got: %d", got)
+	}
+	// An EOA with no code and empty storage root carries nil Root/CodeHash in
+	// slim form, so only the 9-byte list/nonce overhead plus the balance length
+	// is counted.
+	eoa := &types.SlimAccount{Nonce: 1, Balance: uint256.NewInt(0x100)}
+	if got, want := slimAccountSize(eoa), 9+eoa.Balance.ByteLen(); got != want {
+		t.Fatalf("EOA account size, want: %d, got: %d", want, got)
+	}
+	// A contract carries a 32-byte root and 32-byte code hash in addition.
+	contract := &types.SlimAccount{
+		Nonce:    2,
+		Balance:  uint256.NewInt(0x10000),
+		Root:     testrand.Bytes(32),
+		CodeHash: testrand.Bytes(32),
+	}
+	if got, want := slimAccountSize(contract), 9+contract.Balance.ByteLen()+64; got != want {
+		t.Fatalf("contract account size, want: %d, got: %d", want, got)
+	}
+}
+
+// TestStateSizeTracking verifies the storage-slot memory accounting through the
+// merge and revert transitions. Accounts are all identical real accounts, so
+// their size contribution is a constant (acctSize per entry) and the assertions
+// focus on the storage deltas.
 func TestStateSizeTracking(t *testing.T) {
-	expSizeA := 3*(common.HashLength+1) + /* account data */
+	// All accounts are identical, so each contributes the same estimated size.
+	acct := decodeAccountBlob(common.Hash{}, testAccountRLP(0x1))
+	acctSize := common.HashLength + slimAccountSize(acct)
+
+	expSizeA := 3*acctSize + /* account data: a, b, c */
 		2*(2*common.HashLength+1) + /* storage data of 0xa */
 		2*common.HashLength + 3 + /* storage data of 0xb */
 		2*common.HashLength + 1 /* storage data of 0xc */
 
 	a := newStates(
 		map[common.Hash][]byte{
-			{0xa}: {0xa0}, // common.HashLength+1
-			{0xb}: {0xb0}, // common.HashLength+1
-			{0xc}: {0xc0}, // common.HashLength+1
+			{0xa}: testAccountRLP(0x1),
+			{0xb}: testAccountRLP(0x1),
+			{0xc}: testAccountRLP(0x1),
 		},
 		map[common.Hash]map[common.Hash][]byte{
 			{0xa}: {
@@ -405,15 +465,15 @@ func TestStateSizeTracking(t *testing.T) {
 		t.Fatalf("Unexpected size, want: %d, got: %d", expSizeA, a.size)
 	}
 
-	expSizeB := common.HashLength + 2 + common.HashLength + 3 + common.HashLength + /* account data */
+	expSizeB := 2*acctSize + common.HashLength + /* account data: a, b updated; c deleted (key only) */
 		2*common.HashLength + 3 + 2*common.HashLength + 2 + /* storage data of 0xa */
 		2*common.HashLength + 2 + 2*common.HashLength + 2 + /* storage data of 0xb */
 		3*2*common.HashLength /* storage data of 0xc */
 	b := newStates(
 		map[common.Hash][]byte{
-			{0xa}: {0xa1, 0xa1},       // common.HashLength+2
-			{0xb}: {0xb1, 0xb1, 0xb1}, // common.HashLength+3
-			{0xc}: nil,                // common.HashLength, account deletion
+			{0xa}: testAccountRLP(0x1), // account update (same size)
+			{0xb}: testAccountRLP(0x1), // account update (same size)
+			{0xc}: nil,                 // common.HashLength, account deletion
 		},
 		map[common.Hash]map[common.Hash][]byte{
 			{0xa}: {
@@ -438,10 +498,13 @@ func TestStateSizeTracking(t *testing.T) {
 	}
 
 	a.merge(b)
-	mergeSize := expSizeA + 1 /* account a data change */ + 2 /* account b data change */ - 1 /* account c data change */
-	mergeSize += 2*common.HashLength + 2 + 2                                                  /* storage a change */
-	mergeSize += 2*common.HashLength + 2 - 1                                                  /* storage b change */
-	mergeSize += 2*2*common.HashLength - 1                                                    /* storage data removal of 0xc */
+	// Accounts a and b are overwritten with same-size accounts (no delta), but
+	// account c becomes a deletion marker, dropping its value contribution while
+	// retaining the key.
+	mergeSize := expSizeA - slimAccountSize(acct) /* account c value removed */
+	mergeSize += 2*common.HashLength + 2 + 2      /* storage a change */
+	mergeSize += 2*common.HashLength + 2 - 1      /* storage b change */
+	mergeSize += 2*2*common.HashLength - 1        /* storage data removal of 0xc */
 
 	if a.size != uint64(mergeSize) {
 		t.Fatalf("Unexpected size, want: %d, got: %d", mergeSize, a.size)
@@ -450,9 +513,9 @@ func TestStateSizeTracking(t *testing.T) {
 	// Revert the set to original status
 	a.revertTo(
 		map[common.Hash][]byte{
-			{0xa}: {0xa0},
-			{0xb}: {0xb0},
-			{0xc}: {0xc0},
+			{0xa}: testAccountRLP(0x1),
+			{0xb}: testAccountRLP(0x1),
+			{0xc}: testAccountRLP(0x1),
 		},
 		map[common.Hash]map[common.Hash][]byte{
 			{0xa}: {