fix(confidence-engine): use configured model for chat detection

docs/current-handoff.md

@@ -63,6 +63,11 @@
 - Caught exceptions log their actual exception, message, and stack while the client response remains generic; successful responses are not logged. No reasoning, status, or response semantics changed.
 - Next restart point: one production-default manufacturing call while observing the Next.js dev-server terminal.
 
+## Chat capability detection
+
+- Direct POST `/api/chat` was proven supported. The false-negative cause was the capability probe using model `dummy-check`, which conflated model availability with endpoint capability.
+- The probe now uses the configured model; live behaviour after this fix remains untested.
+
 ## Current product architecture
 
 Three distinct routes, not a single page:

lib/llm/provider.js

@@ -60,7 +60,7 @@ function recoverJson(raw) {
 
 let _chatSupported = null;
 
-async function detectChatSupport(baseUrl) {
+async function detectChatSupport(baseUrl, modelName) {
   if (_chatSupported !== null) return _chatSupported;
   
   try {
@@ -68,7 +68,7 @@ async function detectChatSupport(baseUrl) {
       method: "POST",
       headers: { "Content-Type": "application/json" },
       body: JSON.stringify({
-        model: "dummy-check",
+        model: modelName,
         messages: [{ role: "user", content: "test" }],
         stream: false,
       }),
@@ -109,7 +109,7 @@ class OllamaLlmProvider {
     // Step 1: Detect whether /api/chat exists (cache result)
     // ================================================================
     try {
-      chatSupported = await detectChatSupport(baseUrl);
+      chatSupported = await detectChatSupport(baseUrl, modelName);
     } catch { /* failed silently — defaults to false */ }
 
     // ================================================================

tests/llm/provider.test.js

@@ -0,0 +1,32 @@
+import { describe, expect, it, vi } from "vitest";
+
+describe("OllamaLlmProvider chat capability detection", () => {
+  it("uses the configured model for the chat probe and keeps the chat path", async () => {
+    const originalBaseUrl = process.env.OLLAMA_BASE_URL;
+    const fetchSpy = vi.fn()
+      .mockResolvedValueOnce({ ok: true, body: { consume: vi.fn() } })
+      .mockResolvedValueOnce({
+        ok: true,
+        json: async () => ({ message: { content: "{}" } }),
+      });
+    vi.stubGlobal("fetch", fetchSpy);
+    process.env.OLLAMA_BASE_URL = "http://ollama.test";
+
+    try {
+      const { getProvider } = await import("@/lib/llm/provider.js");
+      await getProvider().generateReconstruction("prompt", "configured-model");
+
+      expect(JSON.parse(fetchSpy.mock.calls[0][1].body)).toMatchObject({
+        model: "configured-model",
+        stream: false,
+      });
+      expect(fetchSpy.mock.calls[0][0]).toBe("http://ollama.test/api/chat");
+      expect(fetchSpy.mock.calls[1][0]).toBe("http://ollama.test/api/chat");
+      expect(fetchSpy.mock.calls[1][0]).not.toContain("/api/generate");
+    } finally {
+      vi.unstubAllGlobals();
+      if (originalBaseUrl === undefined) delete process.env.OLLAMA_BASE_URL;
+      else process.env.OLLAMA_BASE_URL = originalBaseUrl;
+    }
+  });
+});