diff --git a/TELEMETRY.md b/TELEMETRY.md
index f27e546e1a..92b56e26cb 100644
--- a/TELEMETRY.md
+++ b/TELEMETRY.md
@@ -189,6 +189,7 @@ Attributes: `gen_ai.operation.name`, `gen_ai.request.model`,
 `gen_ai.usage.output_tokens`, `gen_ai.usage.input_tokens.cached`,
 `gen_ai.usage.input_tokens.cache_write`, `app.ai.reasoning_tokens`,
 `app.ai.empty_output.attempt`, `app.ai.provider_error.kind`,
+`app.ai.provider_error.summary`, `app.ai.provider_error.retry_attempt`,
 `app.guardian.review_attempt`,
 `app.ai.cost.input_usd`, `app.ai.cost.output_usd`,
 `app.ai.cost.cache_read_usd`, `app.ai.cost.cache_write_usd`,
diff --git a/packages/junior-evals/tests/integration/eval-ai-gateway-dispatcher.test.ts b/packages/junior-evals/tests/integration/eval-ai-gateway-dispatcher.test.ts
index 740ffc1eff..af44285339 100644
--- a/packages/junior-evals/tests/integration/eval-ai-gateway-dispatcher.test.ts
+++ b/packages/junior-evals/tests/integration/eval-ai-gateway-dispatcher.test.ts
@@ -89,7 +89,7 @@ describe("eval AI Gateway dispatcher", () => {
           failure: failedAssistant,
           messages: [userMessage as PiMessage, failedAssistant as PiMessage],
         }),
-      ).toEqual({ delayMs: 2_000, messages: [userMessage as PiMessage] });
+      ).toMatchObject({ delayMs: 2_000, messages: [userMessage as PiMessage] });
     } finally {
       await restore();
     }
diff --git a/packages/junior/src/chat/agent/index.ts b/packages/junior/src/chat/agent/index.ts
index 4015ad7537..4f3fe0c68d 100644
--- a/packages/junior/src/chat/agent/index.ts
+++ b/packages/junior/src/chat/agent/index.ts
@@ -1604,7 +1604,10 @@ async function executeAgentRunInPrivacyContext(
 
               providerRetryAttempt += 1;
               await prepareRetry(providerRetry.messages);
-              logWarn("agent.turn.provider.retrying");
+              logWarn("agent.turn.provider.retrying", {
+                ...getProviderErrorAttributes(providerRetry.providerError),
+                "app.ai.provider_error.retry_attempt": providerRetryAttempt,
+              });
               await sleep(providerRetry.delayMs, signal);
               run = agent!.continue();
             }
diff --git a/packages/junior/src/chat/pi/traced-stream.ts b/packages/junior/src/chat/pi/traced-stream.ts
index c58bb6abd3..a7907b7ad6 100644
--- a/packages/junior/src/chat/pi/traced-stream.ts
+++ b/packages/junior/src/chat/pi/traced-stream.ts
@@ -25,6 +25,10 @@ import {
   toGenAiMessagesTraceAttributes,
 } from "@/chat/conversation-privacy";
 import { hasCompactedConversationContext } from "@/chat/services/context-compaction-marker";
+import {
+  createProviderError,
+  getProviderErrorAttributes,
+} from "@/chat/services/provider-error";
 
 type GenAiAttributeMode = "content" | "metadata";
 type TraceAttributeValue = string | number | boolean | string[];
@@ -126,6 +130,12 @@ function buildChatEndAttributes(
     attributes["gen_ai.response.model"] = message.model;
   }
 
+  if (message.stopReason === "error") {
+    const providerError = createProviderError(message.errorMessage ?? "");
+    Object.assign(attributes, getProviderErrorAttributes(providerError));
+    attributes["error.type"] = providerError.kind;
+  }
+
   return attributes;
 }
 
@@ -184,14 +194,17 @@ export function createTracedStreamFn(
         .then(
           (finalMessage: AssistantMessage) => {
             try {
-              for (const [key, value] of Object.entries(
-                buildChatEndAttributes(finalMessage, mode),
-              )) {
+              const endAttributes = buildChatEndAttributes(finalMessage, mode);
+              for (const [key, value] of Object.entries(endAttributes)) {
                 span.setAttribute(key, value);
               }
               if (finalMessage.stopReason === "error") {
-                span.setAttribute("error.type", "provider_error");
-                span.setStatus({ code: 2, message: "LLM stream failed" });
+                const summary = endAttributes["app.ai.provider_error.summary"];
+                span.setStatus({
+                  code: 2,
+                  message:
+                    typeof summary === "string" ? summary : "LLM stream failed",
+                });
               }
             } finally {
               span.end();
diff --git a/packages/junior/src/chat/services/provider-error.ts b/packages/junior/src/chat/services/provider-error.ts
index 146b7a913b..337b9f7a71 100644
--- a/packages/junior/src/chat/services/provider-error.ts
+++ b/packages/junior/src/chat/services/provider-error.ts
@@ -116,6 +116,24 @@ function providerMessage(error: unknown): string {
   return (error instanceof Error ? error.message : String(error)).trim();
 }
 
+const PROVIDER_ERROR_SUMMARY_MAX_CHARS = 240;
+
+/** Short provider-boundary text for telemetry. Drop trailing JSON bodies. */
+function summarizeProviderErrorMessage(message: string): string | undefined {
+  const normalized = message.trim().replace(/\s+/g, " ");
+  if (!normalized) return undefined;
+
+  // Gateway errors often look like `503 {"error":...}`. Keep the lead text.
+  const jsonStart = normalized.indexOf("{");
+  const summary = (
+    jsonStart >= 0 ? normalized.slice(0, jsonStart) : normalized
+  ).trim();
+  if (!summary) return undefined;
+  return summary.length > PROVIDER_ERROR_SUMMARY_MAX_CHARS
+    ? `${summary.slice(0, PROVIDER_ERROR_SUMMARY_MAX_CHARS)}...`
+    : summary;
+}
+
 function extractTransportKind(
   error: unknown,
   depth = 0,
@@ -284,6 +302,16 @@ export function getProviderErrorUserMessage(error: ProviderError): string {
 export function getProviderErrorAttributes(
   error: ProviderError,
 ): Record<string, unknown> {
+  const cause = error.cause;
+  const causeMessage =
+    cause instanceof Error
+      ? cause.message
+      : typeof cause === "string"
+        ? cause
+        : undefined;
+  const summary = causeMessage
+    ? summarizeProviderErrorMessage(causeMessage)
+    : undefined;
   return {
     "app.ai.provider_error.kind": error.kind,
     "app.ai.provider_error.retryable": error.retryable,
@@ -293,6 +321,7 @@ export function getProviderErrorAttributes(
     ...(error.retryAfterMs !== undefined
       ? { "app.ai.provider_error.retry_after_ms": error.retryAfterMs }
       : undefined),
+    ...(summary ? { "app.ai.provider_error.summary": summary } : undefined),
     ...(error.modelId ? { "gen_ai.request.model": error.modelId } : undefined),
   };
 }
diff --git a/packages/junior/src/chat/services/provider-retry.ts b/packages/junior/src/chat/services/provider-retry.ts
index 1c0098fd7f..e68b674c32 100644
--- a/packages/junior/src/chat/services/provider-retry.ts
+++ b/packages/junior/src/chat/services/provider-retry.ts
@@ -3,7 +3,10 @@ import {
   type AssistantMessage,
 } from "@earendil-works/pi-ai";
 import type { PiMessage } from "@/chat/pi/messages";
-import { createProviderError } from "@/chat/services/provider-error";
+import {
+  createProviderError,
+  type ProviderError,
+} from "@/chat/services/provider-error";
 import {
   getPiMessageRole,
   trimTrailingAssistantMessages,
@@ -17,7 +20,13 @@ export function nextProviderRetry(args: {
   attempt: number;
   failure?: AssistantMessage;
   messages: PiMessage[];
-}): { delayMs: number; messages: PiMessage[] } | undefined {
+}):
+  | {
+      delayMs: number;
+      messages: PiMessage[];
+      providerError: ProviderError;
+    }
+  | undefined {
   const backoffMs = PROVIDER_RETRY_DELAYS_MS[args.attempt];
   const errorMessage = args.failure?.errorMessage;
   if (backoffMs === undefined || !args.failure || !errorMessage) {
@@ -47,5 +56,5 @@ export function nextProviderRetry(args: {
     return undefined;
   }
 
-  return { delayMs, messages };
+  return { delayMs, messages, providerError };
 }
diff --git a/packages/junior/tests/unit/services/provider-retry.test.ts b/packages/junior/tests/unit/services/provider-retry.test.ts
index 7a8164da10..559f9645ed 100644
--- a/packages/junior/tests/unit/services/provider-retry.test.ts
+++ b/packages/junior/tests/unit/services/provider-retry.test.ts
@@ -5,6 +5,7 @@ import type { PiMessage } from "@/chat/pi/messages";
 import {
   createProviderError,
   findProviderError,
+  getProviderErrorAttributes,
   getProviderErrorUserMessage,
   isProviderRetryError,
   ProviderError,
@@ -34,6 +35,11 @@ describe("provider retry helpers", () => {
       kind: "network",
       retryable: true,
     });
+    expect(getProviderErrorAttributes(error)).toMatchObject({
+      "app.ai.provider_error.kind": "network",
+      "app.ai.provider_error.summary":
+        "Anthropic stream ended before message_stop",
+    });
     expect(isProviderRetryError(error)).toBe(true);
     expect(isProviderRetryError(createProviderError("invalid_api_key"))).toBe(
       false,
@@ -42,6 +48,21 @@ describe("provider retry helpers", () => {
     expect(isProviderRetryError(new Error(error.message))).toBe(false);
   });
 
+  it("keeps a bounded provider summary without gateway JSON payloads", () => {
+    expect(
+      getProviderErrorAttributes(
+        createProviderError(XAI_SERVICE_UNAVAILABLE, {
+          modelId: "xai/grok-4.5",
+        }),
+      ),
+    ).toMatchObject({
+      "app.ai.provider_error.kind": "server",
+      "app.ai.provider_error.status": 503,
+      "app.ai.provider_error.summary": "503",
+      "gen_ai.request.model": "xai/grok-4.5",
+    });
+  });
+
   it("finds provider errors preserved by domain wrappers", () => {
     const providerError = createProviderError("No object generated", {
       kind: "invalid_response",
@@ -114,13 +135,21 @@ describe("provider retry helpers", () => {
       "Anthropic stream ended before message_stop",
     );
 
-    expect(
-      nextProviderRetry({
-        attempt: 0,
-        failure: failedAssistant,
-        messages: [user, failedAssistant],
-      }),
-    ).toEqual({ delayMs: 2_000, messages: [user] });
+    const retry = nextProviderRetry({
+      attempt: 0,
+      failure: failedAssistant,
+      messages: [user, failedAssistant],
+    });
+    expect(retry).toMatchObject({
+      delayMs: 2_000,
+      messages: [user],
+      providerError: { kind: "network", retryable: true },
+    });
+    expect(getProviderErrorAttributes(retry!.providerError)).toMatchObject({
+      "app.ai.provider_error.kind": "network",
+      "app.ai.provider_error.summary":
+        "Anthropic stream ended before message_stop",
+    });
   });
 
   it("retries a structured xAI 503 despite gateway credential metadata", () => {
@@ -147,7 +176,7 @@ describe("provider retry helpers", () => {
         failure: failedAssistant,
         messages: [user, failedAssistant],
       }),
-    ).toEqual({ delayMs: 2_000, messages: [user] });
+    ).toMatchObject({ delayMs: 2_000, messages: [user] });
   });
 
   it("honors bounded rate-limit hints", () => {
@@ -175,7 +204,7 @@ describe("provider retry helpers", () => {
         failure: failedAssistant,
         messages: [user, failedAssistant],
       }),
-    ).toEqual({ delayMs: 60_000, messages: [user] });
+    ).toMatchObject({ delayMs: 60_000, messages: [user] });
   });
 
   it("classifies HTTP request timeouts without overriding permanent signals", () => {
@@ -232,7 +261,7 @@ describe("provider retry helpers", () => {
         failure,
         messages: [user, failure],
       }),
-    ).toEqual({ delayMs: 2_000, messages: [user] });
+    ).toMatchObject({ delayMs: 2_000, messages: [user] });
   });
 
   it("keeps explicit permanent request failures terminal", () => {
diff --git a/packages/junior/tests/unit/services/turn-failure-response.test.ts b/packages/junior/tests/unit/services/turn-failure-response.test.ts
index 4346fab2a6..86132ca9aa 100644
--- a/packages/junior/tests/unit/services/turn-failure-response.test.ts
+++ b/packages/junior/tests/unit/services/turn-failure-response.test.ts
@@ -72,10 +72,14 @@ describe("finalizeFailedTurnReply", () => {
       "app.ai.provider_error.kind": "server",
       "app.ai.provider_error.retryable": true,
       "app.ai.provider_error.status": 503,
+      "app.ai.provider_error.summary": "503",
       "app.ai.failure_reason": "server",
       "gen_ai.request.model": "xai/grok-4.5",
     });
     expect(attributes).not.toHaveProperty("exception.message");
+    expect(String(attributes?.["app.ai.provider_error.summary"] ?? "")).not.toContain(
+      "providerMetadata",
+    );
   });
 
   it("classifies empty execution failures without raw exception text", () => {