docs: record reviewed SDK2 ownership regressions and diagnostic design

docs/plans/2026-11-10-opcua-sdk-2-evidence.md

@@ -549,3 +549,53 @@ configured-store replacement semantics and was not adopted.
 These results leave the original cold RSS failures and #799/G4 unchanged. The
 [prospective resolution procedure](../../scripts/validation/opcua-lifecycle-memory/RESOLUTION.md)
 is explicitly a design draft, not approval to launch a warmed acceptance run.
+
+### Baseline ownership repair and SDK2 follow-through
+
+The [original ownership controls](../../scripts/validation/opcua-lifecycle-memory/baseline-canonical/OWNERSHIP.md)
+retain both ordinary exit-1 outcomes. Closed-session/point objects collect while
+the same original host remains alive, and old authentication and continuation
+tokens are rejected. After normal stop, five original server graph objects
+remain. One separate public terminal Dispose releases all five and reduces active
+timers from seven to two. The intervention does not rewrite the ordinary failure
+or identify a particular measured timer root.
+
+The SDK1 host omitted terminal server disposal. This was repaired independently
+on master `efb9197a`; [#807](https://gitea.dohertylan.com/dohertj2/lmxopcua/issues/807)
+is closed. Five causal RED cases pass after the fix; 1,032 owning tests pass with
+one existing skip, and the Release solution builds without warnings. The frozen
+`83ed880e` publications and aggregate oracles remain unchanged. The repaired
+master has not been selected as a resource denominator.
+
+SDK2 already disposes adopted servers through ApplicationInstance. Follow-on
+review found that a second host disposal could return while the owner was still
+closing a real session, and configuration/certificate failure before adoption
+could strand the host's raw server. One owner now provides shared completion,
+retains server/authentication state through shutdown, and directly awaits terminal
+disposal only for an unadopted server. A terminally disposed host rejects new
+startup. This does not claim general concurrent asynchronous Start/Dispose support.
+
+At integrated source `ce21f5db`, the whole solution builds with **zero warnings
+or errors**. The server unit suite has **1,122 passes** and one existing skip;
+full wire has **193 passes** and three existing opt-in diagnostic skips. The
+seven focused ownership cases, actual encrypted concurrent shutdown and aligned
+weak-root control pass; trusted-channel/certificate and same-PKI two-boot checks
+are included. [Source hashes and guarded RED/GREEN receipts](evidence/2026-11-10-opcua2/disposal-ownership-integrated.json)
+preserve the three intended unit RED failures, the real-session RED, and the
+earlier test compilation/assertion attempts separately. Astra independently
+verified source, causal failures and integrated results. Earlier full CI, publish
+and soak evidence retain their own source revisions and outcomes.
+
+The [pure Apple empty-keychain control](../../scripts/validation/opcua-lifecycle-memory/native-retention/PURE_APPLE.md)
+kept AppleDatabase inventory flat across 200 successful create/delete/release
+operations with unchanged default/search-list fingerprints. Empty-keychain
+lifecycle alone is insufficient to reproduce the native retention. The first
+[installed bridge import attempt](../../scripts/validation/opcua-lifecycle-memory/native-retention/NATIVE_IMPORT_ATTEMPT.md)
+failed after one partial operation on a diagnostic identity-association
+assumption; all owners and private files were cleaned. It is not a repeated
+import result or evidence of native retention resolution.
+
+#799 and G4 remain open. The [proposed Linux late-slope diagnostic](../../scripts/validation/opcua-lifecycle-memory/LATE_SLOPE.md)
+has implementation design approval only. It retains the cold baseline and first
+resource rejection and cannot emit acceptance. No new long run or release is
+approved; Windows/live interoperability gates remain separate.