diff --git a/bench/merge.bench.js b/bench/merge.bench.js
index 2c650ae6..2bb3a7b8 100644
--- a/bench/merge.bench.js
+++ b/bench/merge.bench.js
@@ -2,7 +2,18 @@
 
 const fs = require('node:fs')
 const path = require('node:path')
-const { merge } = require('../src/lib/cidr')
+const assert = require('node:assert/strict')
+const { parseArgs } = require('node:util')
+
+const { values } = parseArgs({ options: {
+  baseline: { type: 'string' },
+  candidate: { type: 'string' },
+  extended: { type: 'boolean', default: false },
+  json: { type: 'string' },
+  filter: { type: 'string' }
+} })
+const baseline = values.baseline ? require(path.resolve(values.baseline)).merge : null
+const { merge } = require(values.candidate ? path.resolve(values.candidate) : '../src/lib/cidr')
 
 const DATA = path.join(__dirname, '..', 'data')
 const readLines = (file) => fs.readFileSync(file, 'utf8').trim().split('\n').filter(Boolean)
@@ -17,22 +28,43 @@ function loadCases () {
 
   const cncity = fs.readdirSync(path.join(DATA, 'cncity')).map((f) => readLines(path.join(DATA, 'cncity', f)))
 
-  return [
+  const cases = [
     { name: 'CN.txt', inputs: [readLines(path.join(DATA, 'country', 'CN.txt'))] },
     { name: 'non-china (all countries except CN)', inputs: [nonChina] },
     { name: `cncity/*.txt (${cncity.length} files)`, inputs: cncity },
     { name: `every data/**/*.txt separately (${files.length} files)`, inputs: files.map(readLines) },
     { name: 'all data/**/*.txt concatenated', inputs: [everything] }
   ]
+  if (values.extended) {
+    const ascending = merge(everything, true)
+    const shuffled = ascending.slice()
+    let seed = 42
+    for (let i = shuffled.length - 1; i > 0; i--) {
+      seed = (Math.imul(seed, 1664525) + 1013904223) >>> 0
+      const j = seed % (i + 1)
+      ;[shuffled[i], shuffled[j]] = [shuffled[j], shuffled[i]]
+    }
+    const v6 = Array.from({ length: 10000 }, (_, i) => `2001:db8:${i.toString(16)}::1/128`)
+    const v6Networks = Array.from({ length: 10000 }, (_, i) => `2001:db8:${i.toString(16)}::/64`)
+    cases.push(
+      { name: 'IPv4 ascending', inputs: [ascending] },
+      { name: 'IPv4 descending', inputs: [ascending.slice().reverse()] },
+      { name: 'IPv4 shuffled (seed 42)', inputs: [shuffled] },
+      { name: 'IPv6 synthetic 10000 /128', inputs: [v6] },
+      { name: 'IPv6 synthetic 10000 /128 sort=false', inputs: [v6], sort: false },
+      { name: 'IPv6 synthetic 10000 /64', inputs: [v6Networks] },
+      { name: 'mixed synthetic sort=false', inputs: [ascending.slice(0, 10000).concat(v6)], sort: false }
+    )
+  }
+  return cases.filter((c) => !values.filter || c.name.includes(values.filter))
 }
 
-function bench (fn, inputs, minMs = 1500) {
-  for (let i = 0; i < 3; i++) for (const input of inputs) fn(input, true) // warm-up
+function bench (fn, inputs, sort, minMs) {
   let iters = 0
   const t0 = process.hrtime.bigint()
   let elapsed = 0
   while (elapsed < minMs || iters < 5) {
-    for (const input of inputs) fn(input, true)
+    for (const input of inputs) fn(input, sort)
     iters++
     elapsed = Number(process.hrtime.bigint() - t0) / 1e6
   }
@@ -40,17 +72,42 @@ function bench (fn, inputs, minMs = 1500) {
 }
 
 const fmt = (ms) => (ms < 1 ? `${(ms * 1000).toFixed(1)} µs` : `${ms.toFixed(2)} ms`)
+const median = (samples) => samples.slice().sort((a, b) => a - b)[Math.floor(samples.length / 2)]
 
 console.log(`node ${process.version}\n`)
 const rows = []
+const results = []
 for (const c of loadCases()) {
   const size = c.inputs.reduce((n, a) => n + a.length, 0)
-  const b = bench(merge, c.inputs)
+  const sort = c.sort !== false
+  for (const input of c.inputs) {
+    if (baseline) assert.deepEqual(merge(input, sort), baseline(input, sort), c.name)
+  }
+  const before = []
+  const after = []
+  for (let i = 0; i < 3; i++) {
+    for (const input of c.inputs) {
+      if (baseline) baseline(input, sort)
+      merge(input, sort)
+    }
+  }
+  // Alternate execution order to reduce drift; report the median of five batches.
+  for (let round = 0; round < (baseline ? 5 : 1); round++) {
+    const runBefore = () => before.push(bench(baseline, c.inputs, sort, 300))
+    const runAfter = () => after.push(bench(merge, c.inputs, sort, baseline ? 300 : 1500))
+    if (baseline && round % 2 === 0) runBefore()
+    runAfter()
+    if (baseline && round % 2 !== 0) runBefore()
+  }
+  const b = median(after)
+  results.push({ name: c.name, size, sort, before, after, beforeMs: baseline ? median(before) : null, afterMs: b })
   rows.push({
     case: c.name,
     'input CIDRs': size,
+    ...(baseline ? { baseline: fmt(median(before)), speedup: `${(median(before) / b).toFixed(2)}x` } : {}),
     'src/lib/cidr': fmt(b),
     'CIDRs / second': Math.round(size / (b / 1000)).toLocaleString('en-US')
   })
 }
 console.table(rows)
+if (values.json) fs.writeFileSync(values.json, JSON.stringify({ node: process.version, results }, null, 2) + '\n')
diff --git a/src/lib/cidr.js b/src/lib/cidr.js
index 3da1bd7d..270516d6 100644
--- a/src/lib/cidr.js
+++ b/src/lib/cidr.js
@@ -46,7 +46,7 @@ function parseV4Packed (s) {
       if (++digits > 3) return bad(s)
     } else if (c === 46) { // '.'
       if (digits === 0 || octet > 255 || ++dots > 3) return bad(s)
-      ip = ip * 256 + octet
+      ip = (ip << 8) | octet
       octet = 0
       digits = 0
     } else if (c === 47) { // '/'
@@ -58,7 +58,7 @@ function parseV4Packed (s) {
     }
   }
   if (digits === 0 || octet > 255 || dots !== 3) return bad(s)
-  ip = ip * 256 + octet
+  ip = (ip << 8) | octet
 
   let prefix = 32
   if (i < len) {
@@ -74,7 +74,9 @@ function parseV4Packed (s) {
     if (prefix > 32) return bad(s)
   }
   const hb = 32 - prefix
-  const start = ip - (ip % POW2[hb])
+  const ip32 = ip >>> 0
+  // JS masks shift counts to five bits, so /0 needs an explicit case.
+  const start = hb === 32 ? 0 : ((ip32 >>> hb) << hb) >>> 0
   return start * 64 + hb
 }
 
@@ -95,20 +97,21 @@ function emitV4 (s, e, out, starts) {
   }
 }
 
-function mergeV4 (keys, count, out, starts) {
+function mergeV4 (keys, count, out, starts, ordered) {
   if (count === 0) return
   const sorted = count === keys.length ? keys : keys.subarray(0, count)
-  sorted.sort()
+  // Project data files are already ordered by normalized packed key.
+  if (!ordered) sorted.sort()
 
   let k = sorted[0]
-  let hb = k % 64
-  let curS = (k - hb) / 64
+  let curS = Math.floor(k / 64)
+  let hb = k - curS * 64
   let curE = curS + POW2[hb] - 1
 
   for (let i = 1; i < count; i++) {
     k = sorted[i]
-    hb = k % 64
-    const s = (k - hb) / 64
+    const s = Math.floor(k / 64)
+    hb = k - s * 64
     if (s > curE + 1) {
       emitV4(curS, curE, out, starts)
       curS = s
@@ -221,7 +224,7 @@ function emitV6 (s, e, out, starts) {
     const hbAlign = s === 0n ? 128 : bitLength(s & -s) - 1
     const hb = hbLen < hbAlign ? hbLen : hbAlign
     out.push(v6ToString(s, 128 - hb))
-    starts.push(s)
+    if (starts !== null) starts.push(s)
     s += 1n << BigInt(hb)
   }
 }
@@ -255,6 +258,8 @@ function merge (cidrs, sort = false) {
   const len = cidrs.length
   const keys = new Float64Array(len)
   let count = 0
+  let ordered = true
+  let previous = -1
   let v6 = null
 
   for (let i = 0; i < len; i++) {
@@ -264,20 +269,24 @@ function merge (cidrs, sort = false) {
       if (v6 === null) v6 = []
       v6.push(parseV6(c))
     } else {
+      if (k < previous) ordered = false
+      previous = k
       keys[count++] = k
     }
   }
 
   const out4 = []
   if (v6 === null) {
-    mergeV4(keys, count, out4, null)
+    mergeV4(keys, count, out4, null, ordered)
     return out4
   }
 
-  const starts4 = []
+  // Start arrays are only needed to interleave the two address families.
+  const trackStarts = sort && count !== 0
+  const starts4 = trackStarts ? [] : null
   const out6 = []
-  const starts6 = []
-  mergeV4(keys, count, out4, starts4)
+  const starts6 = trackStarts ? [] : null
+  mergeV4(keys, count, out4, starts4, ordered)
   mergeV6(v6, out6, starts6)
 
   if (!sort || out4.length === 0 || out6.length === 0) return out4.concat(out6)
diff --git a/test/cidr.test.js b/test/cidr.test.js
index 1987932f..054682c4 100644
--- a/test/cidr.test.js
+++ b/test/cidr.test.js
@@ -79,6 +79,28 @@ test('edge cases IPv4', () => {
   check([])
 })
 
+test('IPv4 bit masks cover every prefix and unsigned boundary', () => {
+  for (let prefix = 0; prefix <= 32; prefix++) {
+    for (const ip of ['0.0.0.0', '0.0.0.1', '127.255.255.255', '128.0.0.0', '255.255.255.255']) {
+      check([`${ip}/${prefix}`])
+    }
+  }
+})
+
+test('IPv4 ordering detection uses normalized packed keys across IPv6 entries', () => {
+  const inputs = [
+    ['10.0.0.255/24', '10.0.0.1/32', '10.0.0.1/32', '10.0.2.0/24'],
+    ['10.0.0.1/32', '10.0.0.255/24', '10.0.2.0/24'],
+    ['255.255.255.255', '128.0.0.0/1', '0.0.0.0/1'],
+    ['0.0.0.1', '::1', '0.0.0.2', '::2'],
+    ['0.0.0.2', '::1', '0.0.0.1', '::2']
+  ]
+  for (const input of inputs) {
+    check(input, true)
+    check(input, false)
+  }
+})
+
 test('edge cases IPv6 and mixed', () => {
   check(['::/0'])
   check(['::1', '::'])