Merge pull request #96 from lwalden/fix/s5-004r-func-host-boot

fix: align ApplicationInsights packages so func host boots (S5-004r)

Author: lwalden

DECISIONS.md

@@ -346,6 +346,8 @@ Select which sleeve(s) activate first and final initial capital split/account ma
 | KD-002 | Claude API key not provisioned — NOT required under default posture: ADR-029/030 gateway-or-rules with `DirectApiFallbackEnabled=false` means zero metered spend (gateway bearer key provisioned via S5-004). Becomes relevant only if the owner explicitly enables the metered fallback | Conditional — regime integration shipped, degrades to deterministic rules | 2026-02-16 |
 | KD-003 | Backtest pipeline paths use Python (not .NET) | Must install Python + deps separately for backtesting | 2026-04-07 |
 | KD-004 | ✅ RESOLVED 2026-05-29 (S3-001, PR #70). ~~CachingMarketDataService.cs ~360 lines, exceeds 300-line architecture-fitness threshold (grew across S2-001/003/005/006)~~ — decomposed into a 125-line facade + new 280-line MarketRegimeProvider (internal composition, no DI change, zero behavior change) per ADR-017 | Maintainability — closed | 2026-05-29 |
+| KD-005 | **ApplicationInsights 2.x→3.x upgrade gate (S5-004r).** `Microsoft.ApplicationInsights.WorkerService` is pinned to the classic 2.x line (2.23.0): AI 3.x is OTel-based and removed classic API types (`ITelemetryInitializer`) that `Microsoft.Azure.Functions.Worker.ApplicationInsights` 2.50.0 is compiled against — pairing them crashes the isolated worker at DI bootstrap with `TypeLoadException`. Upgrade is BLOCKED until the Functions Worker AI package ships a 3.x-compatible release; when unblocking, re-evaluate the OpenTelemetry.Api pin removed in S5-004r (see comment in `TradingSystem.Functions.csproj`) and clear KD-006 in the same change | Stuck on classic 2.x telemetry SDK; no OTel-native pipeline. Boot-blocking if upgraded prematurely | 2026-06-10 |
+| KD-006 | **discord.com URI-redaction gate before enabling App Insights (S5-004r).** AI 2.x *dependency* telemetry records the full request URL of outbound HTTP calls — for the Discord named clients that URL IS the token-bearing webhook URL. This path bypasses the `System.Net.Http.HttpClient` ILogger filter in `Program.cs` (that filter only covers log telemetry) and is dormant solely because `APPLICATIONINSIGHTS_CONNECTION_STRING` is unset everywhere. Do NOT set that variable until a telemetry processor/initializer that redacts or drops discord.com dependency URLs is in place (or AI 3.x with OTel redaction supersedes it — KD-005) | Secret-leak risk into App Insights if telemetry is ever enabled without redaction | 2026-06-10 |
 
 ---
 

docs/paper-validation-runbook.md

@@ -43,7 +43,13 @@ Run this check each trading morning **before 5:00 AM PT in winter (PST) / 6:00 A
    classification (ADR-029/ADR-030) — **this is expected behavior, not an incident**.
    Restart the gateway (`D:\Source\claude-gateway`) when convenient.
 3. **Functions worker** started (`func start` per section 1).
-4. **Startup-log smoke check:** the worker banner lists both timer functions —
+4. **Host boot smoke:** after `func start`, the function-list banner must appear with **no
+   `TypeLoadException`** (or repeated "worker process" restarts) in the startup output.
+   The banner proves the isolated worker process actually loaded its DI graph — a
+   package-version mismatch (S5-004r) can pass the full unit suite yet crash the worker
+   at bootstrap, because tests never load the Functions host. Run this check first: the
+   function list in step 5 is meaningless until the worker has provably booted.
+5. **Startup-log smoke check:** the worker banner lists both timer functions —
    `DailyOrchestrator_PreMarket` and `DailyOrchestrator_EndOfDay`. If either is missing,
    the build is broken or the wrong project started; do not assume the day will run.
 

src/TradingSystem.Functions/Program.cs

@@ -40,6 +40,12 @@
             context.Configuration.GetSection("TradingSystem"));
         services.Configure<TacticalConfig>(
             context.Configuration.GetSection("TradingSystem:Tactical"));
+        // S5-004r: OptionsSleeveManager takes the bare TacticalConfig (its siblings take
+        // IOptions<TacticalConfig>), so the bare type must be resolvable or the worker dies
+        // at DI validation on boot. This was masked by the AI-package TypeLoadException —
+        // the unit suite never builds the full host graph, only the real `func start` does.
+        services.AddSingleton(sp =>
+            sp.GetRequiredService<Microsoft.Extensions.Options.IOptions<TacticalConfig>>().Value);
         services.Configure<IBKRConfig>(
             context.Configuration.GetSection("IBKR"));
         services.Configure<LocalStorageConfig>(
@@ -53,7 +59,12 @@
         services.Configure<ReportingConfig>(
             context.Configuration.GetSection("Reporting"));
 
-        // Application Insights
+        // Application Insights — registration is inert until APPLICATIONINSIGHTS_CONNECTION_STRING
+        // is set (it is NOT set in local.settings.json or any deployed config today): with no
+        // connection string the SDK builds a disabled TelemetryConfiguration and sends nothing.
+        // Before ever setting that variable, the discord.com URI-redaction gate in DECISIONS.md
+        // (Known Debt KD-005/KD-006) must be satisfied — AI 2.x dependency telemetry records full
+        // request URLs, which for the Discord named clients are token-bearing webhook URLs.
         services.AddApplicationInsightsTelemetryWorkerService();
         services.ConfigureFunctionsApplicationInsights();
 
@@ -124,8 +135,14 @@
             builder.SetMinimumLevel(LogLevel.Information);
             // S5-004 (review): IHttpClientFactory's LogicalHandler logs the full request URI at
             // Information on every send — for Discord named clients that URI IS the token-bearing
-            // webhook URL. Warning+ only for the HttpClient categories keeps the secret out of
-            // console/App Insights logs while preserving error visibility.
+            // webhook URL. Warning+ for every "System.Net.Http.HttpClient.*" ILogger category
+            // keeps the secret out of console logs and out of App Insights *log* telemetry while
+            // preserving error visibility. Scope note (S5-004r review): this filter governs only
+            // ILogger output — it does NOT cover App Insights *dependency* telemetry, which the
+            // AI 2.x SDK emits with the full request URL outside the logging pipeline. That path
+            // is dormant solely because APPLICATIONINSIGHTS_CONNECTION_STRING is unset (see the
+            // enablement-gate comment at the App Insights registration above and DECISIONS.md
+            // Known Debt).
             builder.AddFilter("System.Net.Http.HttpClient", LogLevel.Warning);
         });
     })

src/TradingSystem.Functions/TradingSystem.Functions.csproj

@@ -17,17 +17,21 @@
     <PackageReference Include="Microsoft.Azure.Functions.Worker.Extensions.Http" Version="3.3.0" />
     <PackageReference Include="Microsoft.Azure.Functions.Worker.Extensions.Timer" Version="4.3.1" />
     <PackageReference Include="Microsoft.Azure.Functions.Worker.Sdk" Version="1.16.2" />
-    <PackageReference Include="Microsoft.ApplicationInsights.WorkerService" Version="3.0.0" />
+    <!-- S5-004r: MUST stay on the classic 2.x line. AI 3.x is OTel-based and REMOVED classic
+         API types (e.g. ITelemetryInitializer) that Functions.Worker.ApplicationInsights is
+         built against — pairing 3.x with it makes the isolated worker crash at DI bootstrap
+         with TypeLoadException and the host recycles until the worker-restart cap. -->
+    <PackageReference Include="Microsoft.ApplicationInsights.WorkerService" Version="2.23.0" />
     <PackageReference Include="Microsoft.Azure.Functions.Worker.ApplicationInsights" Version="2.50.0" />
     <PackageReference Include="Azure.Identity" Version="1.18.0" />
     <PackageReference Include="Azure.Extensions.AspNetCore.Configuration.Secrets" Version="1.5.0" />
-    <!-- Transitive lift: GHSA-g94r-2vxg-569j (OpenTelemetry.Api < 1.15.3, excessive memory
-         allocation parsing propagation headers). Pulled in at 1.15.0 via
-         Microsoft.ApplicationInsights.WorkerService 3.0.0 -> Azure.Monitor.OpenTelemetry.Exporter.
-         Pinned to the first patched version; remove once ApplicationInsights.WorkerService ships
-         a patched chain. Tests and SmokeTest inherit this pin via their project reference to
-         Functions — if either drops that reference, it needs its own pin. -->
-    <PackageReference Include="OpenTelemetry.Api" Version="1.15.3" />
+    <!-- S5-004r: the S5-005 OpenTelemetry.Api 1.15.3 pin that used to live here was REMOVED,
+         not lost. The vulnerable transitive chain it patched (GHSA-g94r-2vxg-569j, via
+         ApplicationInsights.WorkerService 3.0.0 -> Azure.Monitor.OpenTelemetry.Exporter) does
+         not exist on the classic 2.23.0 line above — a transitive vulnerability scan
+         (dotnet list package, vulnerable + include-transitive) is clean without it. If
+         WorkerService is ever upgraded back to 3.x (see DECISIONS.md Known Debt),
+         re-evaluate whether an OTel pin is needed. -->
   </ItemGroup>
 
   <ItemGroup>