diff --git a/scripts/rebuild_records_state.py b/scripts/rebuild_records_state.py
new file mode 100644
index 00000000..94bdabed
--- /dev/null
+++ b/scripts/rebuild_records_state.py
@@ -0,0 +1,258 @@
+"""Regenerate ``rdm_records_state.json`` straight from the CDS-RDM database.
+
+Root cause - previously, the code had a bug:
+
+    for version in versions.keys():
+        draft = self._pre_publish(...)
+        published_record = current_rdm_records_service.publish(...)
+        self._after_publish(...)
+    records.append(published_record._record)   # left outside the loop
+
+It ran exactly once per record, for only the latest version. Now fixed in code already.
+
+``invenio migration stats run`` uses only the ``versions[]`` in that state
+file to attribute downloads, so any multi-version record migrated in that
+window lost every download event belonging to any version except the latest.
+
+This script rebuilds the state file from scratch, entirely from CDS-RDM,
+one stream/collection at a time - the same collection name used everywhere
+else (``invenio migration run --collection <name>``, and the folder name
+under ``cds_migrator_kit/rdm/data/`` and ``<CDS_MIGRATOR_KIT_LOGS_PATH>/``):
+
+1. Resolve the collection's community id(s) from ``streams.yaml`` /
+   ``streams_done.yaml`` / ``streams_shelved.yaml`` (same as
+   cds_migrator_kit/runner/runner.py).
+2. Resolve those community ids to every parent record in them using ``RDMParentCommunity``.
+3. Resolve each parent to its legacy recid via the ``lrecid`` PID minted at migration time.
+4. Fetch every published version for that parent.
+5. Rebuild the state entry field-for-field like ``_load_record_state`` does,
+   reading current file metadata straight off each version.
+
+Output is written next to the original, as ``<collection>/rdm_records_state.fixed.json``.
+
+On restart the script reads its own log file for DONE lines and skips any
+legacy recid already completed.
+
+Usage:
+    invenio shell scripts/rebuild_records_state.py
+
+    main(collection="it", dry_run=True)
+    main(collection="it", dry_run=False)
+"""
+
+import json
+import traceback
+from pathlib import Path
+
+import yaml
+from flask import current_app
+from invenio_pidstore.models import PersistentIdentifier
+from invenio_rdm_records.records.api import RDMRecord
+from invenio_rdm_records.records.models import RDMParentCommunity, RDMRecordMetadata
+
+
+log_fp = None
+
+
+def log(msg):
+    print(msg)
+    if log_fp is not None:
+        log_fp.write(msg + "\n")
+        log_fp.flush()
+
+
+def load_completed_recids(log_path):
+    """Return the set of legacy recids already marked DONE in a previous run."""
+    completed = set()
+    path = Path(log_path)
+    if not path.exists():
+        return completed
+    with open(log_path, "r") as f:
+        for line in f:
+            if line.startswith("DONE: legacy_recid="):
+                completed.add(line.strip().split("=")[1])
+    return completed
+
+
+def find_legacy_recids(community_ids):
+    """Return {legacy_recid: parent_object_uuid} for every migrated record in
+    any of the given communities.
+
+    Two plain queries, nothing read from disk: community ids -> parent
+    uuids (``RDMParentCommunity``), then parent uuids -> legacy recid via
+    the ``lrecid`` PID (mirrors the lookup
+    ``CDSMigrationEntryLoad._have_migrated_recid`` does one recid at a time,
+    cds_migrator_kit/rdm/records/load/load.py).
+    """
+    parent_uuids = [
+        row.record_id
+        for row in RDMParentCommunity.query.filter(
+            RDMParentCommunity.community_id.in_(community_ids)
+        )
+    ]
+    pids = PersistentIdentifier.query.filter(
+        PersistentIdentifier.pid_type == "lrecid",
+        PersistentIdentifier.object_uuid.in_(parent_uuids),
+    )
+    return {pid.pid_value: str(pid.object_uuid) for pid in pids}
+
+
+def get_rdm_versions(parent_object_uuid):
+    """Return {version_index: RDMRecord} for every published version of a parent.
+
+    Soft-deleted records are excluded by ``RDMRecord.get_record()`` default.
+    """
+    version_models = (
+        RDMRecordMetadata.query.filter_by(parent_id=parent_object_uuid)
+        .order_by(RDMRecordMetadata.created)
+        .all()
+    )
+    records = {}
+    for m in version_models:
+        try:
+            rdm_record = RDMRecord.get_record(str(m.id))
+            records[rdm_record.versions.index] = rdm_record
+        except Exception as exc:
+            log(f"could not load record {m.id}: {exc}")
+    return records
+
+
+def convert_file_format(file_entries, bucket_id):
+    """Mirror ``RecordLoad._load_record_state.convert_file_format``."""
+    return [
+        {
+            "legacy_file_id": entry["metadata"]["legacy_file_id"],
+            "bucket_id": bucket_id,
+            "file_key": entry["key"],
+            "file_id": entry["file_id"],
+            "size": str(entry["size"]),
+        }
+        for entry in file_entries.values()
+    ]
+
+
+def extract_record_version(record):
+    """Mirror ``RecordLoad._load_record_state.extract_record_version``."""
+    bucket_id = str(record.files.bucket_id)
+    files = record.__class__.files.dump(
+        record, record.files, include_entries=True
+    ).get("entries", {})
+    return {
+        "new_recid": record.pid.pid_value,
+        "version": record.versions.index,
+        "files": convert_file_format(files, bucket_id),
+    }
+
+
+def build_state_entry(legacy_recid, rdm_versions):
+    """Rebuild one ``rdm_records_state.json`` entry from live RDM versions.
+
+    Mirrors ``RecordLoad._load_record_state``
+    (cds_migrator_kit/rdm/records/load/entities/record.py) field-for-field,
+    reading current per-version file metadata straight off each version
+    instead of relying on the state the buggy loop bookkeeping produced.
+    """
+    recid_state = {"legacy_recid": str(legacy_recid), "versions": []}
+    parent_recid = None
+
+    for version_index in sorted(rdm_versions):
+        record = rdm_versions[version_index]
+
+        if parent_recid is None:
+            parent_recid = record.parent.pid.pid_value
+            recid_state["parent_recid"] = parent_recid
+            recid_state["parent_object_uuid"] = str(record.parent.id)
+
+        recid_state["versions"].append(extract_record_version(record))
+
+        if "latest_version" not in recid_state:
+            latest = record.get_latest_by_parent(record.parent)
+            recid_state["latest_version"] = latest["id"]
+            recid_state["latest_version_object_uuid"] = str(latest.id)
+
+    return recid_state
+
+
+def write_state_file(filepath, entries):
+    """Write entries in the same JSON-list-of-compact-objects format
+    ``RecordStateLogger.finalise()`` uses (cds_migrator_kit/reports/log.py),
+    which is what ``invenio migration stats run --filepath`` expects."""
+    Path(filepath).parent.mkdir(parents=True, exist_ok=True)
+    with open(filepath, "w", encoding="utf-8") as f:
+        f.write("[\n")
+        for i, entry in enumerate(entries):
+            json_str = json.dumps(entry, ensure_ascii=False, separators=(",", ":"))
+            comma = "," if i < len(entries) - 1 else ""
+            f.write(f"{json_str}{comma}\n")
+        f.write("]")
+
+
+def main(community_ids, output_path, log_file, dry_run=True):
+    """Rebuild ``rdm_records_state.json`` for one stream/collection.
+
+    :param community_ids: the community UUIDs passed inside the streams.yaml,
+        as used by ``invenio migration run --collection``.
+    :param output_path: should be `<CDS_MIGRATOR_KIT_LOGS_PATH>/<collection>/rdm_records_state.fixed.json`
+        next to the original state file (no overwriting).
+    :param log_file: Log file.
+    :param dry_run: Pass dfry_run False to write generate state json data to file.
+    """
+    global log_fp
+
+    completed_recids = load_completed_recids(log_file)
+    log_fp = open(log_file, "a")
+
+    if completed_recids:
+        log(f"resuming — {len(completed_recids)} recid(s) already completed, skipping them")
+
+    try:
+        legacy_recids = find_legacy_recids(community_ids)
+        log(
+            f"starting, community_ids={str(community_ids)}, "
+            f"dry_run={dry_run}, found legacy_recids={len(legacy_recids)}]"
+        )
+
+        stats = {"checked": 0, "skipped_done": 0, "no_versions": 0, "fixed": 0, "errors": 0}
+        entries = []
+
+        for i, (legacy_recid, parent_object_uuid) in enumerate(legacy_recids.items(), start=1):
+            stats["checked"] += 1
+
+            if legacy_recid in completed_recids:
+                stats["skipped_done"] += 1
+                continue
+
+            log(f"[{i}/{len(legacy_recids)}] legacy_recid={legacy_recid}")
+
+            try:
+                rdm_versions = get_rdm_versions(parent_object_uuid)
+                if not rdm_versions:
+                    log(f"legacy_recid={legacy_recid} - no published versions found, skipping")
+                    stats["no_versions"] += 1
+                    continue
+
+                entries.append(build_state_entry(legacy_recid, rdm_versions))
+                stats["fixed"] += 1
+
+                if not dry_run:
+                    log(f"DONE: legacy_recid={legacy_recid}")
+
+            except Exception as exc:
+                log(f"unexpected error for legacy_recid={legacy_recid}: {exc}")
+                log(traceback.format_exc())
+                stats["errors"] += 1
+
+        if not dry_run:
+            write_state_file(output_path, entries)
+            log(f"wrote {output_path}")
+        else:
+            log(f"dry run — would write {output_path}")
+
+        log(
+            f"\nsummary:\nchecked={stats['checked']}\nalready_done={stats['skipped_done']}\n"
+            f"no_versions={stats['no_versions']}\nfixed={stats['fixed']}\n"
+            f"errors={stats['errors']}"
+        )
+    finally:
+        log_fp.close()
+        log_fp = None