diff --git a/.buildkite/pipelines/format_and_validation.yml.sh b/.buildkite/pipelines/format_and_validation.yml.sh
index ffff9e54d5..b21d1c0fdb 100755
--- a/.buildkite/pipelines/format_and_validation.yml.sh
+++ b/.buildkite/pipelines/format_and_validation.yml.sh
@@ -18,4 +18,13 @@ steps:
     notify:
       - github_commit_status:
           context: "Validate formatting with clang-format"
+  - label: "Validate changelog entries"
+    key: "validate_changelogs"
+    command: ".buildkite/scripts/steps/validate-changelogs.sh"
+    agents:
+      image: "python:3.11-slim"
+    soft_fail: true
+    notify:
+      - github_commit_status:
+          context: "Validate changelog entries"
 EOL
diff --git a/.buildkite/scripts/steps/validate-changelogs.sh b/.buildkite/scripts/steps/validate-changelogs.sh
new file mode 100755
index 0000000000..797f001381
--- /dev/null
+++ b/.buildkite/scripts/steps/validate-changelogs.sh
@@ -0,0 +1,62 @@
+#!/bin/bash
+# Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+# or more contributor license agreements. Licensed under the Elastic License
+# 2.0 and the following additional limitation. Functionality enabled by the
+# files subject to the Elastic License 2.0 may only be used in production when
+# invoked by an Elasticsearch process with a license key installed that permits
+# use of machine learning features. You may not use this file except in
+# compliance with the Elastic License 2.0 and the foregoing additional
+# limitation.
+
+set -euo pipefail
+
+SKIP_LABELS=">test >refactoring >docs >build >non-issue"
+
+# On PR builds, check if the PR has a label that skips changelog validation.
+# BUILDKITE_PULL_REQUEST_LABELS is a comma-separated list set by Buildkite.
+if [[ -n "${BUILDKITE_PULL_REQUEST_LABELS:-}" ]]; then
+  IFS=',' read -ra LABELS <<< "${BUILDKITE_PULL_REQUEST_LABELS}"
+  for label in "${LABELS[@]}"; do
+    label="$(echo "${label}" | xargs)"  # trim whitespace
+    for skip in ${SKIP_LABELS}; do
+      if [[ "${label}" == "${skip}" ]]; then
+        echo "Skipping changelog validation: PR has label '${label}'"
+        exit 0
+      fi
+    done
+  done
+fi
+
+# Install system and Python dependencies
+if ! command -v git &>/dev/null; then
+  apt-get update -qq && apt-get install -y -qq git >/dev/null 2>&1
+fi
+python3 -m pip install --quiet --break-system-packages pyyaml jsonschema 2>/dev/null \
+  || python3 -m pip install --quiet pyyaml jsonschema
+
+# Find changelog files changed in this PR (compared to main/target branch)
+TARGET_BRANCH="${BUILDKITE_PULL_REQUEST_BASE_BRANCH:-main}"
+
+# Fetch the target branch so we can diff against it
+git fetch origin "${TARGET_BRANCH}" --depth=1 2>/dev/null || true
+
+CHANGED_CHANGELOGS=$(git diff --name-only --diff-filter=ACM "origin/${TARGET_BRANCH}"...HEAD -- 'docs/changelog/*.yaml' || true)
+
+if [[ -z "${CHANGED_CHANGELOGS}" ]]; then
+  echo "No changelog files found in this PR."
+  echo "If this PR changes user-visible behaviour, please add a changelog entry."
+  echo "See docs/changelog/README.md for details."
+  echo "To skip this check, add one of these labels: ${SKIP_LABELS}"
+
+  # Soft warning rather than hard failure during rollout
+  if [[ "${CHANGELOG_REQUIRED:-false}" == "true" ]]; then
+    exit 1
+  fi
+  exit 0
+fi
+
+echo "Validating changelog files:"
+echo "${CHANGED_CHANGELOGS}"
+echo ""
+
+python3 dev-tools/validate_changelogs.py ${CHANGED_CHANGELOGS}
diff --git a/build.gradle b/build.gradle
index 843e8718d7..080714884e 100644
--- a/build.gradle
+++ b/build.gradle
@@ -169,6 +169,20 @@ task format(type: Exec) {
   workingDir "${projectDir}"
 }
 
+task validateChangelogs(type: Exec) {
+  commandLine 'python3', 'dev-tools/validate_changelogs.py'
+  workingDir "${projectDir}"
+  description = 'Validate changelog YAML entries against the schema'
+  group = 'verification'
+}
+
+task bundleChangelogs(type: Exec) {
+  commandLine 'python3', 'dev-tools/bundle_changelogs.py', '--version', project.version
+  workingDir "${projectDir}"
+  description = 'Generate consolidated changelog from per-PR YAML entries'
+  group = 'documentation'
+}
+
 task precommit(type: Exec) {
   commandLine shell
   workingDir "${projectDir}"
diff --git a/dev-tools/bundle_changelogs.py b/dev-tools/bundle_changelogs.py
new file mode 100755
index 0000000000..9614b536a8
--- /dev/null
+++ b/dev-tools/bundle_changelogs.py
@@ -0,0 +1,147 @@
+#!/usr/bin/env python3
+"""
+Bundle per-PR changelog YAML files into a consolidated changelog for release.
+
+Usage:
+    python3 bundle_changelogs.py [--dir DIR] [--version VERSION] [--format FORMAT]
+
+Outputs a formatted changelog grouped by type and area, suitable for inclusion
+in release notes.
+
+Formats:
+    markdown (default) - Markdown suitable for GitHub releases
+    asciidoc          - AsciiDoc suitable for Elastic docs
+"""
+
+import argparse
+import sys
+from collections import defaultdict
+from pathlib import Path
+
+try:
+    import yaml
+except ImportError:
+    print("Missing pyyaml. Install with: pip3 install pyyaml", file=sys.stderr)
+    sys.exit(2)
+
+
+TYPE_ORDER = [
+    ("known-issue", "Known issues"),
+    ("security", "Security fixes"),
+    ("breaking", "Breaking changes"),
+    ("breaking-java", "Breaking Java changes"),
+    ("deprecation", "Deprecations"),
+    ("feature", "New features"),
+    ("new-aggregation", "New aggregations"),
+    ("enhancement", "Enhancements"),
+    ("bug", "Bug fixes"),
+    ("regression", "Regression fixes"),
+    ("upgrade", "Upgrades"),
+]
+
+ML_CPP_PULL_URL = "https://github.com/elastic/ml-cpp/pull"
+ML_CPP_ISSUE_URL = "https://github.com/elastic/ml-cpp/issues"
+
+
+def load_entries(changelog_dir):
+    entries = []
+    for path in sorted(changelog_dir.glob("*.yaml")):
+        with open(path) as f:
+            data = yaml.safe_load(f)
+            if data and isinstance(data, dict):
+                data["_file"] = path.name
+                entries.append(data)
+    return entries
+
+
+def format_markdown(entries, version=None):
+    lines = []
+    if version:
+        lines.append(f"## {version}\n")
+
+    grouped = defaultdict(lambda: defaultdict(list))
+    for entry in entries:
+        area = entry.get("area", "General")
+        grouped[entry["type"]][area].append(entry)
+
+    for type_key, type_label in TYPE_ORDER:
+        if type_key not in grouped:
+            continue
+        lines.append(f"### {type_label}\n")
+        for area in sorted(grouped[type_key].keys()):
+            lines.append(f"**{area}**")
+            for entry in sorted(grouped[type_key][area], key=lambda e: e.get("pr", 0)):
+                pr = entry.get("pr")
+                summary = entry["summary"]
+                issues = entry.get("issues", [])
+                issue_refs = ", ".join(f"#{i}" for i in issues)
+                if pr:
+                    line = f"- {summary} [#{pr}]({ML_CPP_PULL_URL}/{pr})"
+                else:
+                    line = f"- {summary}"
+                if issue_refs:
+                    line += f" ({issue_refs})"
+                lines.append(line)
+            lines.append("")
+
+    return "\n".join(lines)
+
+
+def format_asciidoc(entries, version=None):
+    lines = []
+    if version:
+        lines.append(f"== {version}\n")
+
+    grouped = defaultdict(lambda: defaultdict(list))
+    for entry in entries:
+        area = entry.get("area", "General")
+        grouped[entry["type"]][area].append(entry)
+
+    for type_key, type_label in TYPE_ORDER:
+        if type_key not in grouped:
+            continue
+        lines.append(f"=== {type_label}\n")
+        for area in sorted(grouped[type_key].keys()):
+            lines.append(f"*{area}*")
+            for entry in sorted(grouped[type_key][area], key=lambda e: e.get("pr", 0)):
+                pr = entry.get("pr")
+                summary = entry["summary"]
+                issues = entry.get("issues", [])
+                issue_refs = ", ".join(
+                    f"{ML_CPP_ISSUE_URL}/{i}[#{i}]" for i in issues
+                )
+                if pr:
+                    line = f"* {summary} {{ml-pull}}{pr}[#{pr}]"
+                else:
+                    line = f"* {summary}"
+                if issue_refs:
+                    line += f" ({issue_refs})"
+                lines.append(line)
+            lines.append("")
+
+    return "\n".join(lines)
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Bundle changelog YAML files")
+    parser.add_argument("--dir", default=None, help="Changelog directory")
+    parser.add_argument("--version", default=None, help="Version string for heading")
+    parser.add_argument("--format", default="markdown", choices=["markdown", "asciidoc"])
+    args = parser.parse_args()
+
+    repo_root = Path(__file__).resolve().parent.parent
+    changelog_dir = Path(args.dir) if args.dir else repo_root / "docs" / "changelog"
+
+    entries = load_entries(changelog_dir)
+    if not entries:
+        print("No changelog entries found.", file=sys.stderr)
+        sys.exit(0)
+
+    if args.format == "asciidoc":
+        print(format_asciidoc(entries, args.version))
+    else:
+        print(format_markdown(entries, args.version))
+
+
+if __name__ == "__main__":
+    main()
diff --git a/dev-tools/export_changelogs.py b/dev-tools/export_changelogs.py
new file mode 100755
index 0000000000..5d264eceeb
--- /dev/null
+++ b/dev-tools/export_changelogs.py
@@ -0,0 +1,320 @@
+#!/usr/bin/env python3
+"""
+Export ml-cpp changelog entries for inclusion in Elasticsearch release notes.
+
+Copies changelog YAML files from docs/changelog/ to a target directory
+(typically elastic/elasticsearch's docs/changelog/) with a 'ml-cpp-' filename
+prefix to avoid PR number collisions with ES-native entries.
+
+Usage:
+    # Preview what would be exported
+    python3 dev-tools/export_changelogs.py --dry-run
+
+    # Export to a local ES checkout
+    python3 dev-tools/export_changelogs.py --target ~/src/elasticsearch/docs/changelog
+
+    # Export and create a PR in the ES repo
+    python3 dev-tools/export_changelogs.py --target ~/src/elasticsearch/docs/changelog --create-pr
+
+    # Export specific files only
+    python3 dev-tools/export_changelogs.py --target /tmp/out docs/changelog/3008.yaml
+"""
+
+import argparse
+import difflib
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+try:
+    import yaml
+except ImportError:
+    print("Missing pyyaml. Install with: pip3 install pyyaml", file=sys.stderr)
+    sys.exit(2)
+
+try:
+    import jsonschema
+except ImportError:
+    print("Missing jsonschema. Install with: pip3 install jsonschema", file=sys.stderr)
+    sys.exit(2)
+
+
+PREFIX = "ml-cpp-"
+SOURCE_REPO = "elastic/ml-cpp"
+
+
+def validate_entries(entries, schema_path):
+    """Validate all entries against the JSON schema. Returns list of errors."""
+    with open(schema_path) as f:
+        schema = json.load(f)
+
+    validator = jsonschema.Draft7Validator(schema)
+    errors = []
+    for source_path, _, data in entries:
+        for error in validator.iter_errors(data):
+            path = ".".join(str(p) for p in error.absolute_path) or "(root)"
+            errors.append(f"{source_path.name}: {path}: {error.message}")
+    return errors
+
+
+def collect_entries(changelog_dir, specific_files=None):
+    """Collect changelog YAML files, returning (source_path, target_name, data) tuples."""
+    if specific_files:
+        paths = [Path(f) for f in specific_files]
+    else:
+        paths = sorted(changelog_dir.glob("*.yaml"))
+
+    entries = []
+    for path in paths:
+        if not path.exists():
+            print(f"Warning: {path} not found, skipping", file=sys.stderr)
+            continue
+        with open(path) as f:
+            data = yaml.safe_load(f)
+        if not data or not isinstance(data, dict):
+            continue
+
+        target_name = PREFIX + path.name
+        entries.append((path, target_name, data))
+
+    return entries
+
+
+def resolve_conflict(source_path, dest, target_name):
+    """Handle a pre-existing file at the destination. Returns the action taken."""
+    source_lines = source_path.read_text().splitlines(keepends=True)
+    dest_lines = dest.read_text().splitlines(keepends=True)
+
+    if source_lines == dest_lines:
+        print(f"  {target_name}: identical to existing file, skipping")
+        return "skip"
+
+    print(f"\n  {target_name}: file already exists with different content.\n")
+    diff = difflib.unified_diff(
+        dest_lines, source_lines,
+        fromfile=f"existing: {dest.name}",
+        tofile=f"incoming: {source_path.name}",
+    )
+    sys.stdout.writelines("    " + line for line in diff)
+    print()
+
+    while True:
+        choice = input(f"  [{target_name}] (o)verwrite / (s)kip / (a)bort export? ").strip().lower()
+        if choice in ("o", "overwrite"):
+            write_entry_with_source_repo(source_path, dest)
+            print(f"  {target_name}: overwritten")
+            return "overwrite"
+        elif choice in ("s", "skip"):
+            print(f"  {target_name}: skipped")
+            return "skip"
+        elif choice in ("a", "abort"):
+            print("\nExport aborted.")
+            sys.exit(1)
+        else:
+            print("  Please enter 'o' (overwrite), 's' (skip), or 'a' (abort).")
+
+
+def verify_es_repo(target_dir):
+    """Verify that the target looks like an ES docs/changelog directory."""
+    target = Path(target_dir).resolve()
+
+    if not target.is_dir():
+        print(f"Error: target directory does not exist: {target}", file=sys.stderr)
+        sys.exit(1)
+
+    es_repo_root = target.parent.parent
+    markers = [
+        es_repo_root / "build.gradle",
+        es_repo_root / "settings.gradle",
+        es_repo_root / "docs" / "changelog",
+    ]
+    if not all(m.exists() for m in markers):
+        print(
+            f"Warning: {es_repo_root} does not look like an Elasticsearch checkout.\n"
+            f"  Expected to find build.gradle, settings.gradle, and docs/changelog/\n"
+            f"  at the repo root (two levels above --target).\n",
+            file=sys.stderr,
+        )
+        choice = input("  Continue anyway? (y/n) ").strip().lower()
+        if choice not in ("y", "yes"):
+            print("Export aborted.")
+            sys.exit(1)
+
+    return es_repo_root
+
+
+def write_entry_with_source_repo(source_path, dest):
+    """Write a changelog entry to dest, injecting source_repo if not already present."""
+    with open(source_path) as f:
+        data = yaml.safe_load(f)
+    if "source_repo" not in data:
+        data["source_repo"] = SOURCE_REPO
+    with open(dest, "w") as f:
+        yaml.dump(data, f, default_flow_style=False, sort_keys=False)
+
+
+def export_entries(entries, target_dir, dry_run=False):
+    """Export entries to the target directory with prefixed filenames and source_repo."""
+    target = Path(target_dir)
+
+    exported = []
+    skipped = 0
+    for source_path, target_name, data in entries:
+        dest = target / target_name
+        pr = data.get("pr", "n/a")
+        summary = data.get("summary", "")[:60]
+        if dry_run:
+            flag = " [EXISTS]" if dest.exists() else ""
+            print(f"  {target_name}  (PR #{pr}: {summary}){flag}")
+            exported.append(dest)
+        elif dest.exists():
+            action = resolve_conflict(source_path, dest, target_name)
+            if action == "overwrite":
+                exported.append(dest)
+            else:
+                skipped += 1
+        else:
+            write_entry_with_source_repo(source_path, dest)
+            print(f"  Copied {source_path.name} -> {target_name}")
+            exported.append(dest)
+
+    if skipped > 0 and not dry_run:
+        print(f"\n  ({skipped} file(s) skipped due to conflicts)")
+
+    return exported
+
+
+def create_pr(es_repo_dir, exported_files, version=None):
+    """Create a git branch and PR in the ES repo with the exported entries."""
+    es_repo = Path(es_repo_dir).resolve()
+    branch_name = "ml-cpp-changelog-export"
+    if version:
+        branch_name += f"-{version}"
+
+    try:
+        subprocess.run(["git", "checkout", "-b", branch_name], cwd=es_repo, check=True)
+        subprocess.run(["git", "add"] + [str(f) for f in exported_files], cwd=es_repo, check=True)
+
+        msg = "[ML] Add ml-cpp changelog entries"
+        if version:
+            msg += f" for {version}"
+        subprocess.run(["git", "commit", "-m", msg], cwd=es_repo, check=True)
+        subprocess.run(["git", "push", "-u", "origin", branch_name], cwd=es_repo, check=True)
+
+        pr_body = (
+            "Adds ml-cpp changelog entries to the ES release notes.\n\n"
+            "Source: elastic/ml-cpp docs/changelog/"
+        )
+        if version:
+            pr_body += f"\nVersion: {version}"
+        result = subprocess.run(
+            ["gh", "pr", "create", "--title", msg, "--body", pr_body],
+            cwd=es_repo, capture_output=True, text=True,
+        )
+        if result.returncode == 0:
+            print(f"\nPR created: {result.stdout.strip()}")
+        else:
+            print(f"\nFailed to create PR: {result.stderr}", file=sys.stderr)
+            sys.exit(1)
+    except subprocess.CalledProcessError as e:
+        print(f"Git error: {e}", file=sys.stderr)
+        sys.exit(1)
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Export ml-cpp changelog entries for ES release notes",
+    )
+    parser.add_argument(
+        "--target",
+        help="Target directory (e.g. ~/src/elasticsearch/docs/changelog)",
+    )
+    parser.add_argument(
+        "--dir",
+        default=None,
+        help="Source changelog directory (default: docs/changelog/)",
+    )
+    parser.add_argument(
+        "--version",
+        default=None,
+        help="Version label (used in PR title/branch if --create-pr)",
+    )
+    parser.add_argument(
+        "--dry-run",
+        action="store_true",
+        help="Show what would be exported without copying files",
+    )
+    parser.add_argument(
+        "--create-pr",
+        action="store_true",
+        help="Create a PR in the ES repo (requires --target to be inside an ES checkout)",
+    )
+    parser.add_argument(
+        "--prune",
+        action="store_true",
+        help="Delete source YAML files after successful export (use after release)",
+    )
+    parser.add_argument(
+        "files",
+        nargs="*",
+        help="Specific changelog files to export (default: all *.yaml in --dir)",
+    )
+    args = parser.parse_args()
+
+    repo_root = Path(__file__).resolve().parent.parent
+    changelog_dir = Path(args.dir) if args.dir else repo_root / "docs" / "changelog"
+    schema_path = repo_root / "docs" / "changelog" / "changelog-schema.json"
+
+    entries = collect_entries(changelog_dir, args.files if args.files else None)
+    if not entries:
+        print("No changelog entries found.")
+        return
+
+    print(f"Found {len(entries)} changelog entry(ies).")
+
+    # Validate all entries before exporting
+    if schema_path.exists():
+        print("Validating entries against schema... ", end="", flush=True)
+        errors = validate_entries(entries, schema_path)
+        if errors:
+            print(f"FAILED ({len(errors)} error(s)):\n")
+            for error in errors:
+                print(f"  - {error}")
+            print("\nFix validation errors before exporting.")
+            sys.exit(1)
+        print("OK")
+    else:
+        print(f"Warning: schema not found at {schema_path}, skipping validation",
+              file=sys.stderr)
+
+    print()
+
+    if args.dry_run or not args.target:
+        export_entries(entries, args.target or "/dev/null", dry_run=True)
+        if not args.target:
+            print("\nUse --target to export, or --dry-run to preview.")
+        return
+
+    # Verify the target is a real ES checkout
+    es_repo_root = verify_es_repo(args.target)
+
+    exported = export_entries(entries, args.target)
+    if not exported:
+        print("\nNo files exported.")
+        return
+
+    print(f"\nExported {len(exported)} file(s) to {args.target}")
+
+    if args.create_pr:
+        create_pr(es_repo_root, exported, args.version)
+
+    if args.prune:
+        for source_path, _, _ in entries:
+            source_path.unlink()
+            print(f"  Pruned {source_path}")
+        print(f"\nPruned {len(entries)} source file(s)")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/dev-tools/validate_changelogs.py b/dev-tools/validate_changelogs.py
new file mode 100755
index 0000000000..1bc8437c97
--- /dev/null
+++ b/dev-tools/validate_changelogs.py
@@ -0,0 +1,156 @@
+#!/usr/bin/env python3
+"""
+Validate changelog YAML files against the changelog JSON schema.
+
+Usage:
+    python3 validate_changelogs.py [--schema SCHEMA] [--dir DIR] [FILES...]
+
+If FILES are given, only those files are validated.
+Otherwise all *.yaml files in DIR (default: docs/changelog/) are validated.
+
+Exit codes:
+    0  All files valid (or no files to validate)
+    1  One or more validation errors
+    2  Missing dependencies or bad arguments
+"""
+
+import argparse
+import json
+import os
+import re
+import sys
+from pathlib import Path
+
+
+def check_dependencies():
+    """Check that required Python packages are available."""
+    missing = []
+    try:
+        import yaml  # noqa: F401
+    except ImportError:
+        missing.append("pyyaml")
+    try:
+        import jsonschema  # noqa: F401
+    except ImportError:
+        missing.append("jsonschema")
+    if missing:
+        print(
+            f"Missing Python packages: {', '.join(missing)}\n"
+            f"Install with: pip3 install {' '.join(missing)}",
+            file=sys.stderr,
+        )
+        sys.exit(2)
+
+
+def load_schema(schema_path):
+    with open(schema_path) as f:
+        return json.load(f)
+
+
+def validate_file(filepath, schema):
+    """Validate a single YAML file. Returns a list of error strings."""
+    import jsonschema
+    import yaml
+
+    errors = []
+    filename = os.path.basename(filepath)
+    stem = Path(filepath).stem
+
+    try:
+        with open(filepath) as f:
+            data = yaml.safe_load(f)
+    except yaml.YAMLError as e:
+        errors.append(f"{filename}: invalid YAML: {e}")
+        return errors
+
+    if data is None:
+        errors.append(f"{filename}: file is empty")
+        return errors
+
+    if not isinstance(data, dict):
+        errors.append(f"{filename}: expected a YAML mapping, got {type(data).__name__}")
+        return errors
+
+    # Validate against JSON schema
+    validator = jsonschema.Draft7Validator(schema)
+    for error in sorted(validator.iter_errors(data), key=lambda e: list(e.path)):
+        path = ".".join(str(p) for p in error.absolute_path) or "(root)"
+        errors.append(f"{filename}: {path}: {error.message}")
+
+    # Filename convention: numeric filenames must match the pr field.
+    # Types without a pr field (known-issue, security) may use descriptive names.
+    if re.match(r"^\d+$", stem):
+        if "pr" in data and data["pr"] != int(stem):
+            errors.append(
+                f"{filename}: pr field ({data['pr']}) does not match filename ({stem})"
+            )
+    elif "pr" in data:
+        errors.append(
+            f"{filename}: file has a pr field ({data['pr']}), "
+            f"so filename should be {data['pr']}.yaml"
+        )
+
+    return errors
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Validate changelog YAML files")
+    parser.add_argument(
+        "--schema",
+        default=None,
+        help="Path to the JSON schema (default: docs/changelog/changelog-schema.json)",
+    )
+    parser.add_argument(
+        "--dir",
+        default=None,
+        help="Directory containing changelog YAML files (default: docs/changelog/)",
+    )
+    parser.add_argument(
+        "files",
+        nargs="*",
+        help="Specific files to validate (overrides --dir)",
+    )
+    args = parser.parse_args()
+
+    check_dependencies()
+
+    # Resolve paths relative to repo root
+    repo_root = Path(__file__).resolve().parent.parent
+    schema_path = Path(args.schema) if args.schema else repo_root / "docs" / "changelog" / "changelog-schema.json"
+    changelog_dir = Path(args.dir) if args.dir else repo_root / "docs" / "changelog"
+
+    if not schema_path.exists():
+        print(f"Schema not found: {schema_path}", file=sys.stderr)
+        sys.exit(2)
+
+    schema = load_schema(schema_path)
+
+    # Collect files to validate
+    if args.files:
+        yaml_files = [Path(f) for f in args.files]
+    else:
+        yaml_files = sorted(changelog_dir.glob("*.yaml"))
+
+    if not yaml_files:
+        print("No changelog files to validate.")
+        return
+
+    all_errors = []
+    for filepath in yaml_files:
+        if not filepath.exists():
+            all_errors.append(f"{filepath}: file not found")
+            continue
+        errors = validate_file(filepath, schema)
+        all_errors.extend(errors)
+
+    if all_errors:
+        print(f"Changelog validation failed ({len(all_errors)} error(s)):\n")
+        for error in all_errors:
+            print(f"  - {error}")
+        sys.exit(1)
+    else:
+        print(f"Validated {len(yaml_files)} changelog file(s) successfully.")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/docs/changelog/.gitkeep b/docs/changelog/.gitkeep
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/docs/changelog/0000.yaml.sample b/docs/changelog/0000.yaml.sample
new file mode 100644
index 0000000000..bd1d40314a
--- /dev/null
+++ b/docs/changelog/0000.yaml.sample
@@ -0,0 +1,6 @@
+pr: 1234
+summary: Fix anomaly detection model state persistence for large jobs
+area: Machine Learning
+type: bug
+issues:
+  - 1230
diff --git a/docs/changelog/2841.yaml b/docs/changelog/2841.yaml
new file mode 100644
index 0000000000..a7d97b7041
--- /dev/null
+++ b/docs/changelog/2841.yaml
@@ -0,0 +1,5 @@
+pr: 2841
+summary: "Better messaging regarding OOM process termination"
+area: Machine Learning
+type: enhancement
+issues: []
diff --git a/docs/changelog/2846.yaml b/docs/changelog/2846.yaml
new file mode 100644
index 0000000000..d225e9a1cb
--- /dev/null
+++ b/docs/changelog/2846.yaml
@@ -0,0 +1,5 @@
+pr: 2846
+summary: "Report the actual memory usage of the autodetect process"
+area: Machine Learning
+type: enhancement
+issues: []
diff --git a/docs/changelog/2848.yaml b/docs/changelog/2848.yaml
new file mode 100644
index 0000000000..ccca2a64c6
--- /dev/null
+++ b/docs/changelog/2848.yaml
@@ -0,0 +1,5 @@
+pr: 2848
+summary: "Improve adherence to memory limits for the bucket gatherer"
+area: Machine Learning
+type: enhancement
+issues: []
diff --git a/docs/changelog/2863.yaml b/docs/changelog/2863.yaml
new file mode 100644
index 0000000000..7042ed337a
--- /dev/null
+++ b/docs/changelog/2863.yaml
@@ -0,0 +1,5 @@
+pr: 2863
+summary: "Update the PyTorch library to version 2.7.1"
+area: Machine Learning
+type: upgrade
+issues: []
diff --git a/docs/changelog/2889.yaml b/docs/changelog/2889.yaml
new file mode 100644
index 0000000000..7206fb8e5e
--- /dev/null
+++ b/docs/changelog/2889.yaml
@@ -0,0 +1,5 @@
+pr: 2889
+summary: "Downgrade log severity for a batch of recoverable errors"
+area: Machine Learning
+type: enhancement
+issues: []
diff --git a/docs/changelog/2894.yaml b/docs/changelog/2894.yaml
new file mode 100644
index 0000000000..528edff646
--- /dev/null
+++ b/docs/changelog/2894.yaml
@@ -0,0 +1,5 @@
+pr: 2894
+summary: "Better error handling regarding quantiles state documents"
+area: Machine Learning
+type: enhancement
+issues: []
diff --git a/docs/changelog/2895.yaml b/docs/changelog/2895.yaml
new file mode 100644
index 0000000000..6d5e810b39
--- /dev/null
+++ b/docs/changelog/2895.yaml
@@ -0,0 +1,5 @@
+pr: 2895
+summary: "Better handling of invalid JSON state documents"
+area: Machine Learning
+type: enhancement
+issues: []
diff --git a/docs/changelog/3008.yaml b/docs/changelog/3008.yaml
new file mode 100644
index 0000000000..e851b35d57
--- /dev/null
+++ b/docs/changelog/3008.yaml
@@ -0,0 +1,5 @@
+pr: 3008
+summary: "Harden pytorch_inference with TorchScript model graph validation"
+area: Machine Learning
+type: enhancement
+issues: [2890]
diff --git a/docs/changelog/3015.yaml b/docs/changelog/3015.yaml
new file mode 100644
index 0000000000..9b9dee8485
--- /dev/null
+++ b/docs/changelog/3015.yaml
@@ -0,0 +1,5 @@
+pr: 3015
+summary: "Add EuroBERT and Jina v5 ops to graph validation allowlist"
+area: Machine Learning
+type: enhancement
+issues: []
diff --git a/docs/changelog/3017.yaml b/docs/changelog/3017.yaml
new file mode 100644
index 0000000000..ae0820ff1a
--- /dev/null
+++ b/docs/changelog/3017.yaml
@@ -0,0 +1,5 @@
+pr: 3017
+summary: "Fix flaky CIoManagerTest/testFileIoGood test"
+area: Machine Learning
+type: bug
+issues: []
diff --git a/docs/changelog/README.md b/docs/changelog/README.md
new file mode 100644
index 0000000000..6ab3193508
--- /dev/null
+++ b/docs/changelog/README.md
@@ -0,0 +1,102 @@
+# Changelog entries
+
+Each pull request that changes user-visible behaviour should include a changelog
+entry as a YAML file in this directory, named `<PR_NUMBER>.yaml`.
+
+The schema is aligned with the
+[Elasticsearch changelog schema](https://github.com/elastic/elasticsearch/blob/main/build-tools-internal/src/main/resources/changelog-schema.json)
+so that ml-cpp entries can be consumed directly by the ES release notes pipeline.
+
+## Format
+
+```yaml
+pr: 2914
+summary: Split build and test into separate pipeline steps
+area: Machine Learning
+type: enhancement
+issues: []
+```
+
+### Required fields
+
+| Field     | Description |
+|-----------|-------------|
+| `type`    | The type of change (see below). Always required. |
+| `summary` | A concise, user-facing description of the change. Always required. |
+| `pr`      | The pull request number (integer). Required unless type is `known-issue` or `security`. |
+| `area`    | The area of the codebase affected (see below). Required unless type is `known-issue` or `security`. |
+
+### Optional fields
+
+| Field         | Description |
+|---------------|-------------|
+| `issues`      | List of related GitHub issue numbers (integers). Default: `[]` |
+| `highlight`   | Release highlight object (see below). |
+| `breaking`    | Breaking change details. **Required** when type is `breaking` or `breaking-java`. |
+| `deprecation` | Deprecation details. **Required** when type is `deprecation`. |
+
+### Valid areas
+
+Most ml-cpp entries should use **Machine Learning**. Other valid areas from the
+ES schema (e.g. **Inference**) may be used when appropriate. The full list of
+valid areas is defined in `changelog-schema.json`.
+
+### Valid types
+
+| Type | Description |
+|------|-------------|
+| `breaking` | A change that breaks backwards compatibility (requires `breaking` object) |
+| `breaking-java` | A breaking change to the Java API (requires `breaking` object) |
+| `bug` | A fix for an existing defect |
+| `deprecation` | Deprecation of existing functionality (requires `deprecation` object) |
+| `enhancement` | An improvement to existing functionality |
+| `feature` | A wholly new feature |
+| `known-issue` | A known issue (`pr` and `area` not required) |
+| `new-aggregation` | A new aggregation type |
+| `regression` | A fix for a recently introduced defect |
+| `security` | A security fix (`pr` and `area` not required) |
+| `upgrade` | An upgrade-related change |
+
+### Highlight object
+
+For changes worthy of a release highlight:
+
+```yaml
+highlight:
+  notable: true
+  title: "Short title for the highlight"
+  body: "Longer description in AsciiDoc format (no triple-backtick code blocks)."
+```
+
+### Breaking / Deprecation object
+
+Required when `type` is `breaking`, `breaking-java`, or `deprecation`:
+
+```yaml
+breaking:
+  area: Machine Learning
+  title: "Short title describing the breaking change"
+  details: "Detailed description of what changed (AsciiDoc, no triple-backticks)."
+  impact: "What users need to do to adapt."
+  notable: true
+```
+
+Valid areas for breaking/deprecation changes are a subset of the main areas,
+defined in `changelog-schema.json` under `compatibilityChangeArea`.
+
+## When is a changelog entry required?
+
+A changelog entry is **required** for any PR that:
+- Fixes a bug
+- Adds or changes user-visible functionality
+- Changes the API or data formats
+- Deprecates or removes functionality
+
+A changelog entry is **not required** for:
+- Pure refactoring with no behaviour change
+- Test-only changes
+- CI/build infrastructure changes (unless they affect the shipped artefact)
+- Documentation-only changes
+
+PRs that do not require a changelog entry should be labelled with
+`>test`, `>refactoring`, `>docs`, `>build`, or `>non-issue` to skip validation.
diff --git a/docs/changelog/changelog-schema.json b/docs/changelog/changelog-schema.json
new file mode 100644
index 0000000000..2777aedb4a
--- /dev/null
+++ b/docs/changelog/changelog-schema.json
@@ -0,0 +1,311 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$id": "https://github.com/elastic/ml-cpp/tree/main/docs/changelog",
+  "$ref": "#/definitions/Changelog",
+  "definitions": {
+    "Changelog": {
+      "type": "object",
+      "properties": {
+        "pr": {
+          "type": "integer"
+        },
+        "issues": {
+          "type": "array",
+          "items": {
+            "type": "integer"
+          }
+        },
+        "area": {
+          "type": "string",
+          "enum": [
+            "Aggregations",
+            "Allocation",
+            "Analysis",
+            "Application",
+            "Audit",
+            "Authentication",
+            "Authorization",
+            "Autoscaling",
+            "CAT APIs",
+            "CCR",
+            "CCS",
+            "CRUD",
+            "Client",
+            "Cluster Coordination",
+            "Codec",
+            "Data streams",
+            "DLM",
+            "Discovery-Plugins",
+            "Distributed",
+            "Downsampling",
+            "EQL",
+            "ES|QL",
+            "Engine",
+            "Experiences",
+            "Extract&Transform",
+            "FIPS",
+            "Features",
+            "Geo",
+            "Graph",
+            "Health",
+            "Highlighting",
+            "ILM",
+            "IdentityProvider",
+            "Indices APIs",
+            "Inference",
+            "Infra/CLI",
+            "Infra/Circuit Breakers",
+            "Infra/Core",
+            "Infra/Logging",
+            "Infra/Node Lifecycle",
+            "Infra/Plugins",
+            "Infra/REST API",
+            "Infra/Resiliency",
+            "Infra/Scripting",
+            "Infra/Settings",
+            "Infra/Transport API",
+            "Infra/Metrics",
+            "Ingest",
+            "Ingest Node",
+            "Java High Level REST Client",
+            "Java Low Level REST Client",
+            "License",
+            "Logs",
+            "Machine Learning",
+            "Mapping",
+            "Monitoring",
+            "Network",
+            "Packaging",
+            "Percolator",
+            "Performance",
+            "PromQL",
+            "Query Languages",
+            "Ranking",
+            "Recovery",
+            "Reindex",
+            "Relevance",
+            "Rollup",
+            "SQL",
+            "Search",
+            "Searchable Snapshots",
+            "Security",
+            "SLM",
+            "Snapshot/Restore",
+            "Stats",
+            "Store",
+            "Suggesters",
+            "Task Management",
+            "TLS",
+            "Transform",
+            "TSDB",
+            "Vector Search",
+            "Watcher"
+          ]
+        },
+        "type": {
+          "type": "string",
+          "enum": [
+            "breaking",
+            "breaking-java",
+            "bug",
+            "deprecation",
+            "enhancement",
+            "feature",
+            "known-issue",
+            "new-aggregation",
+            "regression",
+            "security",
+            "upgrade"
+          ]
+        },
+        "summary": {
+          "type": "string",
+          "minLength": 1
+        },
+        "highlight": {
+          "$ref": "#/definitions/Highlight"
+        },
+        "breaking": {
+          "$ref": "#/definitions/CompatibilityChange"
+        },
+        "deprecation": {
+          "$ref": "#/definitions/CompatibilityChange"
+        },
+        "source_repo": {
+          "type": "string",
+          "description": "GitHub repository (owner/name) for entries from external repos, e.g. elastic/ml-cpp. Defaults to elastic/elasticsearch when absent."
+        }
+      },
+      "required": [
+        "type",
+        "summary"
+      ],
+      "anyOf": [
+        {
+          "$comment": "PR number and area fields not required for known-issue type",
+          "if": {
+            "not": {
+              "properties": {
+                "type": {
+                  "const": "known-issue"
+                }
+              }
+            }
+          },
+          "then": {
+            "required": [
+              "pr",
+              "area"
+            ]
+          }
+        },
+        {
+          "$comment": "PR number and area fields not required for security type",
+          "if": {
+            "not": {
+              "properties": {
+                "type": {
+                  "const": "security"
+                }
+              }
+            }
+          },
+          "then": {
+            "required": [
+              "pr",
+              "area"
+            ]
+          }
+        }
+      ],
+      "allOf": [
+        {
+          "if": {
+            "properties": {
+              "type": {
+                "const": "breaking"
+              }
+            }
+          },
+          "then": {
+            "required": [
+              "breaking"
+            ]
+          }
+        },
+        {
+          "if": {
+            "properties": {
+              "type": {
+                "const": "breaking-java"
+              }
+            }
+          },
+          "then": {
+            "required": [
+              "breaking"
+            ]
+          }
+        }
+      ],
+      "if": {
+        "properties": {
+          "type": {
+            "const": "deprecation"
+          }
+        }
+      },
+      "then": {
+        "required": [
+          "deprecation"
+        ]
+      },
+      "additionalProperties": false
+    },
+    "Highlight": {
+      "properties": {
+        "notable": {
+          "type": "boolean"
+        },
+        "title": {
+          "type": "string",
+          "minLength": 1
+        },
+        "body": {
+          "type": "string",
+          "pattern": "(?s)^((?!```).)*$",
+          "minLength": 1
+        }
+      },
+      "required": [
+        "title",
+        "body"
+      ],
+      "additionalProperties": false
+    },
+    "CompatibilityChange": {
+      "properties": {
+        "area": {
+          "$ref": "#/definitions/compatibilityChangeArea"
+        },
+        "title": {
+          "type": "string",
+          "minLength": 1
+        },
+        "details": {
+          "type": "string",
+          "pattern": "(?s)^((?!```).)*$",
+          "minLength": 1
+        },
+        "impact": {
+          "type": "string",
+          "pattern": "(?s)^((?!```).)*$",
+          "minLength": 1
+        },
+        "notable": {
+          "type": "boolean"
+        },
+        "ess_setting_change": {
+          "type": "boolean"
+        }
+      },
+      "required": [
+        "area",
+        "title",
+        "details",
+        "impact"
+      ],
+      "additionalProperties": false
+    },
+    "compatibilityChangeArea": {
+      "type": "string",
+      "enum": [
+        "Aggregations",
+        "Analysis",
+        "Authorization",
+        "Cluster and node setting",
+        "Command line tool",
+        "CRUD",
+        "ES|QL",
+        "ILM",
+        "Index setting",
+        "Ingest",
+        "JVM option",
+        "Java API",
+        "Logging",
+        "Logs",
+        "Machine Learning",
+        "Mapping",
+        "Metrics",
+        "Packaging",
+        "Painless",
+        "REST API",
+        "Rollup",
+        "Search",
+        "System requirement",
+        "Transform"
+      ]
+    },
+    "additionalProperties": false
+  }
+}