@@ -250,9 +250,9 @@ See `ui/AGENTS.md` for the app-id / localStorage / auth gotchas.
     "be liberal in what you accept" does **not** apply. Change the **spec** only when the spec itself is genuinely
     wrong, and then deliberately (edit the Zod schema + migrate). When an alias must be tolerated at all, declare it
     as an **ADR-0087 conversion-layer entry** (never a bare `??`, no executor shims) so it is declared, loud, tested,
-    and *removable on a schedule*. Stored `sys_metadata` rows are covered from the other side: every rehydration seam
-    replays the **full** conversion chain — retired entries included — via `applyConversionsToStoredItem` (ADR-0087
-    addendum), so a consumer never needs its own accommodation for a legacy stored shape either. Strengthens #5.
+    and *removable on a schedule*. Stored rows: `applyConversionsToStoredItem` replays the chain the installed release
+    carries, retired entries included; kits age out one major after retiring (ADR-0087 as amended, ruling 6434038284),
+    so run `os migrate meta --stored --apply` before leaving a release that still converts the row. Strengthens #5.
 13. **An accepted ADR binds until a superseding ADR says otherwise.** Reversing a recorded decision is itself a
     decision: it needs a **new ADR** (or an amended status line on the old one), not a changeset that quietly does the
     opposite. Before changing behaviour in `docs/adr/`-governed territory, **grep the ADRs for the surface you are
 
@@ -5,9 +5,15 @@
  * (#4327), the follow-up to #3903's read-path guarantee.
  *
  * #4317 closed the correctness gap from the read side: every stored-row
- * rehydration seam replays the full ADR-0087 conversion chain
- * (`applyConversionsToStoredItem`, retired entries included), so a row written
- * under any past protocol is *served* canonical forever. What it deliberately
+ * rehydration seam replays the ADR-0087 conversion chain the installed release
+ * carries (`applyConversionsToStoredItem`, retired entries included), so a row
+ * is *served* canonical while that release still carries its conversions. A
+ * retirement kit ages out one major after it retires (ADR-0087 as amended,
+ * ruling 6434038284), so this pass must run on a release that still carries a
+ * row's conversions, before the deployment moves to the release that drops
+ * them. Past that, the seams still read and serve the row's legacy bytes, but
+ * no conversion recognizes its shape, so this pass counts it `canonical`: a
+ * false clean, not evidence. What it deliberately
  * did not do is make the rows themselves canonical — a pre-17 row keeps its
  * legacy bytes, the chain re-lowers it on every load, and each affected row
  * emits one conversion notice per process.
@@ -19,10 +25,13 @@
  * the types rather than in the CLI so an admin route renders the same run the
  * same way.
  *
- * **Not load-bearing.** #3855's conclusion still holds — an operator-run
- * migration cannot be relied on, so the read path stays the guarantee and
- * nothing gates on this having run. No `sys_migration` flag is recorded for
- * exactly that reason: a flag row would advertise a gate that does not exist.
+ * **Not load-bearing while the release carries a row's conversions.** #3855's
+ * conclusion holds within that window — an operator-run migration cannot be
+ * relied on, so the read path stays the guarantee there and nothing gates on
+ * this having run; for a row whose conversions the next release drops, this
+ * pass is the required step above, not an optional one. No `sys_migration`
+ * flag is recorded for exactly that reason: a flag row would advertise a gate
+ * that does not exist.
  * The verifiable statement operators wanted is the *re-run* — a second pass
  * that reports every row canonical is the evidence, and it costs one command.
  */
 
@@ -52,7 +52,9 @@ export interface ApplyConversionsOptions {
    * have no author to teach, so each replays the FULL chain, retired entries
    * included — ADR-0087's `## Addendum (2026-07-31)` for the first two, the
    * #12772 ruling for the third. The fixture CI sets it as well, so graduated
-   * transforms stay covered forever (ADR-0087 D3).
+   * transforms stay covered while the installed release carries them; a
+   * retirement kit ages out one major after it retires (ADR-0087 as amended,
+   * ruling 6434038284), and its transform leaves coverage with it.
    *
    * `objectstack migrate meta` is NOT one of these callers: `applyMetaMigrations`
    * (`../migrations/chain.js`) looks each step's conversion up in
 
@@ -14,13 +14,18 @@
  * {@link applyConversionsToStoredItem} is the one primitive every stored-row
  * rehydration seam calls, and it encodes the policy for data at rest:
  *
- * - **The full chain replays, including `retiredFromLoadPath` entries.**
+ * - **The chain the installed release carries replays, including
+ *   `retiredFromLoadPath` entries.**
  *   Retirement is an *authoring-surface* event — the schema tombstones the
  *   key so a live author is taught the canonical spelling. A stored row has
  *   no author; refusing its historical shape would not teach anyone, it
- *   would only break data that once worked. D3 keeps every conversion
- *   forever precisely so any past major can replay forward — a row at rest
- *   is the perpetual "consumer arriving late".
+ *   would only break data that once worked. But a retirement kit ages out
+ *   one major after it retires (ADR-0087 as amended, ruling 6434038284), and
+ *   its conversion leaves with it: a row still in that shape is no longer
+ *   recognized here, so it must be canonicalized by the stored pass
+ *   (`os migrate meta --stored --apply`) on a release that still carries
+ *   the row's conversions, before the deployment moves to the release that
+ *   drops them.
  * - **Idempotent by construction.** Conversions only rewrite shapes they
  *   positively recognize, so re-converting an already-canonical item is a
  *   no-op — seams may safely stack (e.g. a row converted here and again at