diff --git a/src/manifest.js b/src/manifest.js
index ba5a83e..629db76 100644
--- a/src/manifest.js
+++ b/src/manifest.js
@@ -63,6 +63,7 @@ async function fetchManifests(manifests, resolver) {
     // Inherit sequence number from manifest if not present in entry
     for (const entry of entries) {
       entry.partition_spec_id = manifest.partition_spec_id ?? 0
+      if (entry.snapshot_id == null) entry.snapshot_id = manifest.added_snapshot_id
 
       if (entry.sequence_number === undefined) {
         // When reading v1 manifests with no sequence number column,
diff --git a/src/write/manifest.js b/src/write/manifest.js
index f27307c..047b264 100644
--- a/src/write/manifest.js
+++ b/src/write/manifest.js
@@ -39,6 +39,12 @@ function manifestEntrySchema(schema, partitionSpec, formatVersion, manifestConte
     mapField('nan_value_counts', 137, 'k138_v139', 138, 139, 'long'),
     mapField('lower_bounds', 125, 'k126_v127', 126, 127, 'bytes'),
     mapField('upper_bounds', 128, 'k129_v130', 129, 130, 'bytes'),
+    {
+      name: 'split_offsets',
+      type: ['null', { type: 'array', items: 'long', 'element-id': 133 }],
+      default: null,
+      'field-id': 132,
+    },
     { name: 'sort_order_id', type: ['null', 'int'], default: null, 'field-id': 140 },
   ]
   if (manifestContent === 1) {
@@ -216,6 +222,63 @@ export function writeDeleteManifest({ writer, schema, partitionSpec, snapshotId,
   })
 }
 
+/**
+ * Write a data manifest containing already-existing data entries. Used when
+ * manifests are rewritten without touching the data files they reference —
+ * merging many small manifests into one, for instance. Carried-over files
+ * must keep their original data and file sequence numbers rather than
+ * inheriting the rewriting snapshot's, which would misattribute their data
+ * sequence numbers and change which delete files apply to them. All entries
+ * must belong to the supplied partition spec.
+ *
+ * @param {object} options
+ * @param {Writer} options.writer
+ * @param {Schema} options.schema
+ * @param {PartitionSpec} options.partitionSpec
+ * @param {ManifestEntry[]} options.entries
+ * @param {2|3} [options.formatVersion]
+ * @returns {void | Promise<void>} resolves when the writer's `finish()` lands
+ */
+export function writeExistingDataManifest({ writer, schema, partitionSpec, entries, formatVersion = 2 }) {
+  const records = entries.map(entry => {
+    const dataFile = entry.data_file
+    if (entry.status === 2) {
+      throw new Error('writeExistingDataManifest cannot rewrite deleted entries as existing')
+    }
+    if (dataFile.content !== 0) {
+      throw new Error(`writeExistingDataManifest expects data files (content=0), got content=${dataFile.content}`)
+    }
+    if (entry.partition_spec_id !== partitionSpec['spec-id']) {
+      throw new Error(`existing data entry partition spec ${entry.partition_spec_id} does not match ${partitionSpec['spec-id']}`)
+    }
+    const record = manifestEntryRecord(dataFile, schema, partitionSpec, 0n, formatVersion, 0)
+    record.status = 0
+    record.snapshot_id = entry.snapshot_id ?? null
+    record.sequence_number = entry.sequence_number ?? null
+    record.file_sequence_number = entry.file_sequence_number ?? null
+    if (record.snapshot_id == null) {
+      throw new Error('existing data manifest entry missing snapshot id')
+    }
+    if (record.sequence_number == null || record.file_sequence_number == null) {
+      throw new Error('existing data manifest entry missing sequence numbers')
+    }
+    return record
+  })
+
+  return avroWrite({
+    writer,
+    schema: manifestEntrySchema(schema, partitionSpec, formatVersion, 0),
+    records,
+    metadata: {
+      'format-version': String(formatVersion),
+      content: 'data',
+      schema: icebergSchemaJson(schema),
+      'partition-spec': partitionSpecJson(partitionSpec),
+      'partition-spec-id': String(partitionSpec['spec-id']),
+    },
+  })
+}
+
 /**
  * Write a delete manifest containing already-existing delete entries. Used
  * when a v3 deletion vector replaces an older vector in a mixed manifest: the
@@ -292,6 +355,7 @@ function manifestEntryRecord(dataFile, schema, partitionSpec, snapshotId, format
     nan_value_counts: encodeMap(dataFile.nan_value_counts),
     lower_bounds: encodeMap(dataFile.lower_bounds),
     upper_bounds: encodeMap(dataFile.upper_bounds),
+    split_offsets: dataFile.split_offsets?.length ? dataFile.split_offsets : null,
     sort_order_id: dataFile.content === 1 ? null : dataFile.sort_order_id ?? 0,
   }
   if (manifestContent === 1) {
diff --git a/test/manifest.test.js b/test/manifest.test.js
index a381021..744c64a 100644
--- a/test/manifest.test.js
+++ b/test/manifest.test.js
@@ -2,7 +2,8 @@ import { describe, expect, it } from 'vitest'
 import fs from 'fs'
 import { icebergManifests } from '../src/manifest.js'
 import { icebergMetadata } from '../src/metadata.js'
-import { localResolver } from './helpers.js'
+import { writeDataManifest } from '../src/write/manifest.js'
+import { localResolver, memResolver } from './helpers.js'
 
 describe('Iceberg Manifests', () => {
   const tableUrl = 's3://hyperparam-iceberg/spark/bunnies'
@@ -91,4 +92,43 @@ describe('Iceberg Manifests', () => {
     expect(manifests).toHaveLength(1)
     expect(calls).toEqual([{ url: manifestPath, byteLength: manifestLength }])
   })
+
+  it('inherits a null entry snapshot id from the manifest list', async () => {
+    const { resolver: memory } = memResolver()
+    const manifestPath = 'http://test/inherited-snapshot-id.avro'
+    const writer = memory.writer?.(manifestPath)
+    if (!writer) throw new Error('expected resolver.writer')
+    await writeDataManifest({
+      writer,
+      schema: { type: 'struct', 'schema-id': 0, fields: [] },
+      partitionSpec: { 'spec-id': 0, fields: [] },
+      snapshotId: /** @type {any} */ (null),
+      dataFiles: [{
+        content: 0,
+        file_path: 'http://test/data.parquet',
+        file_format: 'parquet',
+        partition: {},
+        record_count: 1n,
+        file_size_in_bytes: 1n,
+      }],
+    })
+
+    const metadata = /** @type {any} */ ({
+      'current-snapshot-id': 77,
+      snapshots: [{
+        'snapshot-id': 77,
+        manifests: [{
+          manifest_path: manifestPath,
+          manifest_length: BigInt(writer.offset),
+          partition_spec_id: 0,
+          content: 0,
+          sequence_number: 5n,
+          added_snapshot_id: 77n,
+        }],
+      }],
+    })
+
+    const manifests = await icebergManifests({ metadata, resolver: memory })
+    expect(manifests[0].entries[0].snapshot_id).toBe(77n)
+  })
 })
diff --git a/test/write/manifest.test.js b/test/write/manifest.test.js
index fca6ee7..70a75b3 100644
--- a/test/write/manifest.test.js
+++ b/test/write/manifest.test.js
@@ -1,11 +1,11 @@
 import { describe, expect, it } from 'vitest'
 import { ByteWriter } from 'hyparquet-writer'
-import { writeDataManifest } from '../../src/write/manifest.js'
+import { writeDataManifest, writeExistingDataManifest } from '../../src/write/manifest.js'
 import { avroMetadata } from '../../src/avro/avro.metadata.js'
 import { avroRead } from '../../src/avro/avro.read.js'
 
 /**
- * @import {DataFile, PartitionSpec, Schema} from '../../src/types.js'
+ * @import {DataFile, ManifestEntry, PartitionSpec, Schema} from '../../src/types.js'
  */
 
 describe('writeDataManifest', () => {
@@ -119,3 +119,179 @@ describe('writeDataManifest', () => {
     expect(records[0].data_file.first_row_id).toBe(1000n)
   })
 })
+
+describe('writeExistingDataManifest', () => {
+  /** @type {Schema} */
+  const schema = {
+    type: 'struct',
+    'schema-id': 0,
+    fields: [
+      { id: 1, name: 'id', required: true, type: 'long' },
+      { id: 2, name: 'name', required: false, type: 'string' },
+    ],
+  }
+
+  /** @type {PartitionSpec} */
+  const unpartitioned = { 'spec-id': 0, fields: [] }
+
+  /** @type {DataFile} */
+  const dataFile = {
+    content: 0,
+    file_path: 's3://bucket/table/data/abc.parquet',
+    file_format: 'parquet',
+    partition: {},
+    record_count: 3n,
+    file_size_in_bytes: 421n,
+    sort_order_id: 0,
+  }
+
+  /** @type {ManifestEntry} */
+  const entry = {
+    status: 1,
+    snapshot_id: 111n,
+    sequence_number: 7n,
+    file_sequence_number: 7n,
+    partition_spec_id: 0,
+    data_file: dataFile,
+  }
+
+  it('writes EXISTING entries that keep their original snapshot and sequence numbers', async () => {
+    const writer = new ByteWriter()
+    writeExistingDataManifest({ writer, schema, partitionSpec: unpartitioned, entries: [entry] })
+    const buffer = writer.getBuffer()
+
+    const reader = { view: new DataView(buffer), offset: 0 }
+    const { metadata, syncMarker } = await avroMetadata(reader)
+    expect(metadata.content).toBe('data')
+
+    const records = await avroRead({ reader, metadata, syncMarker })
+    expect(records).toHaveLength(1)
+    // status 0 (EXISTING) with explicit numbers: an ADDED entry would leave
+    // these null and inherit the rewriting snapshot's sequence number, which
+    // would move the file forward in time and change which deletes apply.
+    expect(records[0]).toMatchObject({
+      status: 0,
+      snapshot_id: 111n,
+      sequence_number: 7n,
+      file_sequence_number: 7n,
+      data_file: { file_path: 's3://bucket/table/data/abc.parquet' },
+    })
+  })
+
+  it('carries v3 first_row_id through', async () => {
+    const writer = new ByteWriter()
+    writeExistingDataManifest({
+      writer,
+      schema,
+      partitionSpec: unpartitioned,
+      entries: [{ ...entry, data_file: { ...dataFile, first_row_id: 1000n } }],
+      formatVersion: 3,
+    })
+    const buffer = writer.getBuffer()
+
+    const reader = { view: new DataView(buffer), offset: 0 }
+    const { metadata, syncMarker } = await avroMetadata(reader)
+    const records = await avroRead({ reader, metadata, syncMarker })
+    expect(records[0].data_file.first_row_id).toBe(1000n)
+  })
+
+  it('round-trips stat maps from a read-decoded manifest', async () => {
+    // The Avro reader hands Iceberg maps back as {key, value} record arrays
+    // rather than plain objects, and a manifest rewrite feeds exactly those
+    // decoded entries back in. Sequence numbers are supplied here the way
+    // `icebergManifests` materializes them from the manifest list.
+    const first = new ByteWriter()
+    writeDataManifest({
+      writer: first,
+      schema,
+      partitionSpec: unpartitioned,
+      snapshotId: 111n,
+      dataFiles: [{
+        ...dataFile,
+        value_counts: { 1: 3n, 2: 3n },
+        lower_bounds: { 1: new Uint8Array([1, 0, 0, 0, 0, 0, 0, 0]) },
+        split_offsets: [4n, 100n],
+      }],
+    })
+    const firstBuffer = first.getBuffer()
+
+    const firstReader = { view: new DataView(firstBuffer), offset: 0 }
+    const firstMeta = await avroMetadata(firstReader)
+    const decoded = (await avroRead({
+      reader: firstReader,
+      metadata: firstMeta.metadata,
+      syncMarker: firstMeta.syncMarker,
+    }))[0]
+
+    const second = new ByteWriter()
+    writeExistingDataManifest({
+      writer: second,
+      schema,
+      partitionSpec: unpartitioned,
+      entries: [/** @type {ManifestEntry} */ ({
+        ...decoded,
+        sequence_number: 7n,
+        file_sequence_number: 7n,
+        partition_spec_id: 0,
+      })],
+    })
+    const secondBuffer = second.getBuffer()
+
+    const secondReader = { view: new DataView(secondBuffer), offset: 0 }
+    const secondMeta = await avroMetadata(secondReader)
+    const records = await avroRead({
+      reader: secondReader,
+      metadata: secondMeta.metadata,
+      syncMarker: secondMeta.syncMarker,
+    })
+    expect(records[0].data_file.value_counts).toEqual(decoded.data_file.value_counts)
+    expect(records[0].data_file.lower_bounds).toEqual(decoded.data_file.lower_bounds)
+    expect(records[0].data_file.split_offsets).toEqual(decoded.data_file.split_offsets)
+    expect(records[0].data_file.record_count).toBe(3n)
+  })
+
+  it('rejects deleted entries', () => {
+    expect(() => writeExistingDataManifest({
+      writer: new ByteWriter(),
+      schema,
+      partitionSpec: unpartitioned,
+      entries: [{ ...entry, status: 2 }],
+    })).toThrow('writeExistingDataManifest cannot rewrite deleted entries as existing')
+  })
+
+  it('rejects delete files', () => {
+    expect(() => writeExistingDataManifest({
+      writer: new ByteWriter(),
+      schema,
+      partitionSpec: unpartitioned,
+      entries: [{ ...entry, data_file: { ...dataFile, content: 1 } }],
+    })).toThrow('writeExistingDataManifest expects data files')
+  })
+
+  it('rejects entries without sequence numbers', () => {
+    expect(() => writeExistingDataManifest({
+      writer: new ByteWriter(),
+      schema,
+      partitionSpec: unpartitioned,
+      entries: [{ ...entry, file_sequence_number: undefined }],
+    })).toThrow('existing data manifest entry missing sequence numbers')
+  })
+
+  it('rejects entries without a materialized snapshot id', () => {
+    expect(() => writeExistingDataManifest({
+      writer: new ByteWriter(),
+      schema,
+      partitionSpec: unpartitioned,
+      entries: [{ ...entry, snapshot_id: undefined }],
+    })).toThrow('existing data manifest entry missing snapshot id')
+  })
+
+  it('rejects entries from another partition spec', () => {
+    expect(() => writeExistingDataManifest({
+      writer: new ByteWriter(),
+      schema,
+      partitionSpec: unpartitioned,
+      entries: [{ ...entry, partition_spec_id: 1 }],
+    })).toThrow('existing data entry partition spec 1 does not match 0')
+  })
+})