From 0adfb244b28815d5d29770ecde82c857f191a8ff Mon Sep 17 00:00:00 2001
From: Joseph Doherty <dohejw01@gmail.com>
Date: Sun, 11 Oct 2026 10:18:36 -0400
Subject: [PATCH] docs(index): OtOpcUa batch #667-#673 (retention, Part 11
 modified values, event CPs, OpcUaClient session loss, time-weighted
 aggregates, unreachable-Primary capture)

---
 .../skills/sister-project-history/otopcua.md  | 29 +++++++++++++++++++
 1 file changed, 29 insertions(+)

diff --git a/.claude/skills/sister-project-history/otopcua.md b/.claude/skills/sister-project-history/otopcua.md
index b954605..25466db 100644
--- a/.claude/skills/sister-project-history/otopcua.md
+++ b/.claude/skills/sister-project-history/otopcua.md
@@ -192,3 +192,32 @@ AddRoute of the dev Mac (`oee-tagdump-mac`, `10.181.206.245.1.1`) on the two tes
   about 30 s with no historization, because #537 Primary-gated capture waits for Akka downing. It is
   backend-independent; clean failovers are gap-free. **ScadaBridge impact: none by default**; the same as SqlServer
   when selected.
+- **2026-10-11 batch #667–#673 (all closed; OtOpcUa `bc03adb1`).** Four docker-dev live gates passed (one R2 failure
+  fixed and re-gated). Follow-ups are #674–#678.
+  - **#667:** opt-in SqlServer retention, `ServerHistorian:SqlServer:RetentionDays`. Old samples and events are purged
+    in batches. The purge is Primary-gated, with one purger per database via `sp_getapplock`.
+  - **#668:** Part 11 modified-value history.
+    - Both SQL backends keep an audit trail of every HistoryUpdate, recording the API-key principal and the prior value.
+    - SqlServer schema is v2: `hist.SampleModified`, written in the same statement as the change. QuestDB uses an
+      append-only `sample_modified` table.
+    - `IsReadModified=true` reads support continuation points. Gateway returns `BadHistoryOperationUnsupported`.
+    - ⚠️ With `AutoCreateSchema=false`, apply `002_sample_modified.sql` before upgrading, or every historian call
+      fails.
+  - **#669:** event HistoryRead now uses continuation points. Open-ended reads follow Part 11: start-only reads
+    forward, end-only raw reads backward.
+  - **#670:** OpcUaClient now goes `Degraded` on upstream session loss.
+    - It publishes one `BadNoCommunication` per tag per outage, and the #527 watchdog does the recovery.
+    - The SDK `SessionReconnectHandler` is deleted.
+    - A post-recovery replay guard stops late, older notifications from regressing values.
+  - **#671:** Part 13 time-weighted aggregates on both SQL backends: TimeAverage, TimeAverage2, Total, Total2 and
+    Interpolative. ⚠️ `Total` is now the time-weighted integral, no longer `SUM`. A window over
+    `MaxTimeWeightedRawRows` returns `BadTooManyOperations`.
+  - **#672:** a standby starts capturing as soon as its recording Primary is unreachable, controlled by
+    `CaptureWhilePrimaryUnreachable`, default true. The crash-failover gap drops from about 31 s to about 13 s.
+  - **#673:** FOCAS test flake fixed.
+  - **ScadaBridge impact:**
+    - Tags backed by an OpcUaClient driver now go Bad (`BadNoCommunication`, value retained server-side) while their
+      upstream is lost, where they used to stay Good. This is the quality contract working.
+    - On SqlServer/QuestDB nodes, `Total` aggregate values change meaning.
+    - Event HistoryRead may now return continuation points, which a client must follow.
+    - Otherwise none by default.