docs(sql): chrome/AdminUI gate PASSED; records the Test-Connect provider-gate defect it found

Also adds the two fixture connection strings to both docker-dev centrals, so
the rig can author + browse the Postgres and Oracle fixtures on 10.100.0.35.

docker-dev/docker-compose.yml

@@ -286,6 +286,12 @@ services:
       # the rig's own `sql` container, a SEPARATE `DevSql` database seeded with dbo.TagValues (KeyValue)
       # + dbo.LatestStatus (WideRow). Committed-dev-secret exception, same as ConfigDb above.
       Sql__ConnectionStrings__DevSql: "Server=sql,1433;Database=DevSql;User Id=sa;Password=OtOpcUa!Dev123;TrustServerCertificate=True;"
+      # The Postgres + Oracle counterparts, pointing at the two DISPOSABLE fixture containers on the
+      # shared docker host (tests/.../Driver.Sql.IntegrationTests/Docker/{postgres,oracle}/). Same
+      # committed-dev-secret exception; the passwords are each fixture compose's documented default.
+      # Unlike DevSql these are NOT on the rig's own network, so they carry the host's IP.
+      Sql__ConnectionStrings__FixturePg: "Host=10.100.0.35;Port=15432;Database=sqlpollfixture;Username=otopcua;Password=OtOpcUaPg2026x"
+      Sql__ConnectionStrings__FixtureOracle: "User Id=otopcua;Password=OtOpcUaOra2026x;Data Source=10.100.0.35:11521/FREEPDB1"
       Cluster__Hostname: "0.0.0.0"
       Cluster__Port: "4053"
       Cluster__PublicHostname: "central-1"
@@ -426,6 +432,10 @@ services:
       # the rig's own `sql` container, a SEPARATE `DevSql` database seeded with dbo.TagValues (KeyValue)
       # + dbo.LatestStatus (WideRow). Committed-dev-secret exception, same as ConfigDb above.
       Sql__ConnectionStrings__DevSql: "Server=sql,1433;Database=DevSql;User Id=sa;Password=OtOpcUa!Dev123;TrustServerCertificate=True;"
+      # Postgres + Oracle fixture connection strings — see central-1. Both centrals must carry them:
+      # :9200 round-robins, so the AdminUI browse can land on either node.
+      Sql__ConnectionStrings__FixturePg: "Host=10.100.0.35;Port=15432;Database=sqlpollfixture;Username=otopcua;Password=OtOpcUaPg2026x"
+      Sql__ConnectionStrings__FixtureOracle: "User Id=otopcua;Password=OtOpcUaOra2026x;Data Source=10.100.0.35:11521/FREEPDB1"
       Cluster__Hostname: "0.0.0.0"
       Cluster__Port: "4053"
       Cluster__PublicHostname: "central-2"

docs/plans/2026-08-05-sql-postgres-oracle-design.md

@@ -17,9 +17,41 @@ WideRow `topByTimestamp`, catalog case-folding, browse walk). Env recipe used:
 `SQL_ORACLE_ENDPOINT=10.100.0.35:11521 SQL_ORACLE_USER=otopcua SQL_ORACLE_PASSWORD=<compose-up
 value>` (service defaults `FREEPDB1`).
 
-**§4.3's chrome/AdminUI gate remains with the operator** — provider dropdown authoring,
-Test-connect, `/raw` browse + browse-commit against both fixtures at `http://localhost:9200`. The
-docker-dev rig must be rebuilt from a master image at/after `d54f3c83` first.
+**§4.3's chrome/AdminUI gate PASSED (2026-08-05) — and it found a real defect.** Run against a
+docker-dev rig rebuilt from master (both centrals on one image, provider assemblies confirmed
+present in `/app`) with `Sql__ConnectionStrings__FixturePg` / `__FixtureOracle` added to central-1
+**and** central-2 (`:9200` round-robins, so a browse can land on either node).
+
+⚠️ **The gate's first finding: Test Connect was broken for BOTH new providers.**
+`SqlDriverProbe.DefaultDialectFor` was never widened — it still constructed `SqlServerDialect`
+only and threw `InvalidOperationException` for Postgres/Oracle, so a config that **deployed and
+browsed fine failed Test Connect**. There are *three* dialect-selection sites and Task 4's `Files:`
+block named two (`SqlDriverFactoryExtensions`, `SqlDriverBrowser`) plus the enum. No test caught
+it because **every case in `SqlDriverProbeTests` builds its probe through `ForTest(...)`**, which
+injects `_ => dialect` and never reaches the production switch — the parameterless constructor had
+zero coverage. Fixed in `75a057f2`: switch widened arm-for-arm with the factory, plus an internal
+seam covering the gate directly, a reserved-provider (MySql/Odbc) falsification control, and an
+end-to-end production-constructor case. Red-first confirmed (3 failures) before the widening;
+Driver.Sql.Tests then 361/361.
+
+Legs, after the fix:
+
+1. **Provider dropdown authoring ✅** — dropdown offers SqlServer/Postgres/Oracle; both saved and
+   persisted as `{"provider":"Postgres",…}` / `{"provider":"Oracle",…}`, i.e. serialized by **name**,
+   not ordinal (the driver-enum-serialization defect class does not recur here).
+2. **Test-connect ✅** — Postgres `OK · 174 ms`, Oracle `OK · 386 ms`. **Negative control:** the
+   Oracle driver cross-wired to the Postgres connection string returns
+   `FAILED · … (ARGUMENTEXCEPTION)` — the provider itself rejecting the string, and *not*
+   `InvalidOperationException`, which proves the dialect was constructed and the green came from a
+   real round trip rather than an unconditional OK.
+3. **`/raw` browse + browse-commit ✅** — Postgres browse discovered schema `sqlpoll` → `tag_values`
+   / `latest_status` → per-column LEAFs; **the schema is owned by `sqlpoll_owner` while the poller
+   connects as `otopcua`**, so this is the live confirmation that `PostgresDialect.ListSchemasSql`
+   sources from `information_schema.tables` rather than `.schemata`. Oracle browse listed the PDB's
+   schemas → `OTOPCUA` → `TAG_VALUES`/`LATEST_STATUS` → LEAFs. One leaf committed per provider,
+   both persisting schema-qualified with the right type:
+   `{"model":"WideRow","table":"sqlpoll.tag_values","columnName":"num_value"}` [Double] and
+   `{"model":"WideRow","table":"OTOPCUA.TAG_VALUES","columnName":"NUM_VALUE"}` [Double].
 
 ## 1. Problem