@@ -0,0 +1,21 @@
+---
+'@objectstack/service-storage': patch
+---
+
+fix(storage)!: a file its record lets go keeps that record's read rule
+
+ADR-0104 D3 promises one file, one owning record and one access rule. That rule now holds after the owning record lets the file go: until a writer who may read that record claims the file again, or the platform collects it, the file is read by that record's rule. The download door and the record-write claim path ask the same verdict, so both follow it. A file no record has owned is read as before, within its organization. The uploader still reads their own file.
+
+`sys_file` gains two nullable fields, `released_ref_object` and `released_ref_id`, naming the record that last owned the file. The platform writes them when a record lets a file go and clears them when a record claims it. Existing rows hold NULL and are read as before.
+
+Clause-②: no (narrowing) — a file read that a record's rule refuses is now refused after the record lets the file go. Nothing that was refused becomes accepted.
+
+Metadata-protocol: none (a file read rule and two platform-managed bookkeeping fields; nothing a metadata author writes changes)
+
+<!-- adr-0087: not-required (no-migration-prescription) No metadata author's surface moves: no spec key or authorable spelling is added, removed, renamed or re-shaped, and no exported declaration is removed or narrowed, so there is nothing for `objectstack migrate meta` to rewrite. The platform object `sys_file` gains two nullable bookkeeping fields that only the platform writes; schema sync adds the columns and existing rows read NULL, which is the shape every row had before. What changes is which callers a file read admits after the file's owning record lets it go. The remedy is reading the file as a reader of that record, never a rewrite of a package source or a stored row. The other categories are closed on facts: the package publishes (not unpublished); no ADR-0087 id is named or touched (not registered or already-registered); and nothing exported is removed or narrowed (not runtime-interface-only, type-surface-only or runtime-export-retired). -->
+
+**BREAKING** for a reader the owning record's rule refuses, shipped as `patch` because the metadata protocol does not move.
+
+- **Before.** The owning record's rule did not hold for the file once the record let it go.
+- **Now.** It is read by the rule of the record that let it go, until it is claimed again or collected.
+- **Fix.** Read or claim such a file as a reader of that record.
@@ -1459,9 +1459,11 @@ exists in the acting organization. One file had two access rules.
 - **One judgement, two callers.** The door's verdict is one function, and the
   claim path calls it under the write's caller context; neither restates it.
   It admits the uploader; a reader of the record that owns the file, through
-  its object's `fileAccessDelegate` where one is declared; a member of the
-  organization for a file no record owns; a reader of any parent for an
-  attachments-scope file; anyone for a `public_read` file.
+  its object's `fileAccessDelegate` where one is declared; for a file a record
+  let go, a reader of that record, until the file is claimed again or
+  collected (#26046); a member of the organization for a file no record has
+  owned; a reader of any parent for an attachments-scope file; anyone for a
+  `public_read` file.
 - **The quiet answer.** At the write door, a file the writer may not read is
   answered as one the organization-scoped lookup cannot see: the id is kept
   as written, nothing is claimed or copied, and constraint verdicts do not run
 
@@ -65,6 +65,17 @@ import { callerContext } from './attachment-access-hooks.js';
  * and keep any other id as written; neither is a file the anonymous caller
  * may provide. An empty or absent file field is written as before.
  *
+ * ### A released file keeps its record's rule
+ *
+ * [#26046] Releasing a file clears its ownership columns, and the download
+ * door reads a file no record owns by the organization's rule. That rule is
+ * for a file no record ever held; a file a record let go is still the one that
+ * record held. So release keeps the record ({@link releaseOwnership}), the
+ * door's verdict judges the file by that record's read access until it is
+ * claimed again or collected, and a claim, which asks the same verdict, is
+ * open only to a writer who may read that record. A deleted record is read by
+ * nobody, so its released files are left to their uploader.
+ *
  * ### The one write that cannot honour it, and is refused
  *
  * [#7102] A predicate (`multi: true`) update has ONE payload for N matched
@@ -803,6 +814,12 @@ async function claimFile(
     patch.status = 'committed';
     patch.deleted_at = null;
   }
+  // [#26046] The claiming record's rule replaces the one the file kept from
+  // the record that let it go.
+  if (row.released_ref_object != null || row.released_ref_id != null) {
+    patch.released_ref_object = null;
+    patch.released_ref_id = null;
+  }
   await engine.update('sys_file', patch, { context: { ...SYSTEM_CTX } });
 }
 
@@ -848,6 +865,13 @@ async function collectionOpen(engine: FileReferenceEngine): Promise<boolean> {
  * Only a `committed` file is tombstoned: `pending` rows already carry their
  * own never-completed reap policy, and re-tombstoning a `deleted` row would
  * reset its grace clock.
+ *
+ * [#26046] Every release keeps the record that let the file go
+ * (`released_ref_object` / `released_ref_id`), whether or not it tombstones:
+ * the file is read by that record's rule until a writer who may read it
+ * claims it again ({@link claimFile} clears them), or it is collected. Without
+ * them a released file would read as one no record ever held, whose rule is
+ * the organization's, which is weaker than the record's was.
  */
 async function releaseOwnership(
   engine: FileReferenceEngine,
@@ -861,7 +885,14 @@ async function releaseOwnership(
     if (id == null) continue;
     const tombstone = collect && row.status === 'committed';
     try {
-      const patch: Record<string, unknown> = { id, ref_object: null, ref_id: null, ref_field: null };
+      const patch: Record<string, unknown> = {
+        id,
+        ref_object: null,
+        ref_id: null,
+        ref_field: null,
+        released_ref_object: row.ref_object ?? null,
+        released_ref_id: row.ref_id ?? null,
+      };
       if (tombstone) {
         patch.status = 'deleted';
         patch.deleted_at = new Date().toISOString();