Merge v1.25.3: v1253-briefing-timeout

src/components/insights/daily-briefing.tsx

@@ -36,6 +36,7 @@ import type {
   DailyBriefingKeyFinding,
   DailyBriefingSignal,
 } from "@/lib/ai/schema";
+import type { BriefingFailureClass } from "@/lib/insights/briefing-failure-marker";
 
 /**
  * v1.4.27 MB7 / CF-68 — per-metric routing target for the briefing
@@ -127,6 +128,13 @@ interface DailyBriefingProps {
    * generic empty state for a "couldn't generate" one whose CTA retries.
    */
   generationFailed?: boolean;
+  /**
+   * v1.25.3 — coarse class of the most recent failure. When the empty state is
+   * shown ("couldn't generate"), it lets the hint point at the right lever:
+   * `timeout` → raise the AI response timeout, `auth` → re-check the provider.
+   * Absent / null → the generic failed-description holds.
+   */
+  generationFailureClass?: BriefingFailureClass | null;
   /**
    * Optional slot for a meta control mounted in the card header — the
    * comparison toggle migrates here from the hero in commit 5.
@@ -358,6 +366,7 @@ export function DailyBriefing({
   noProvider = false,
   noProviderStale = false,
   generationFailed = false,
+  generationFailureClass = null,
   metaSlot,
   memory = null,
 }: DailyBriefingProps) {
@@ -365,6 +374,20 @@ export function DailyBriefing({
   const fmt = useFormatters();
   const { user } = useAuth();
 
+  // v1.25.3 — when there is no last good text and the last attempt failed,
+  // point the hint at the lever that matches the failure class. A timeout (the
+  // dominant failure on a slow self-hosted model) → raise the AI response
+  // timeout; an auth / config rejection → re-check the provider in Settings.
+  // Anything else keeps the calmer generic line. Non-alarming by design.
+  const failedDescriptionKey =
+    generationFailureClass === "timeout"
+      ? "insights.dailyBriefing.failedDescriptionTimeout"
+      : generationFailureClass === "auth"
+        ? "insights.dailyBriefing.failedDescriptionProvider"
+        : generationFailureClass === "rate-limit"
+          ? "insights.dailyBriefing.failedDescriptionRateLimit"
+          : "insights.dailyBriefing.failedDescription";
+
   return (
     // v1.13.1 — heading-above-card pattern. The "Tagesbriefing" title
     // moves OUT of the card into an `<h2>` above it (mirroring the
@@ -588,7 +611,7 @@ export function DailyBriefing({
               }
               description={
                 generationFailed
-                  ? t("insights.dailyBriefing.failedDescription")
+                  ? t(failedDescriptionKey)
                   : t("insights.dailyBriefing.emptyDescription")
               }
               action={

src/components/insights/use-insights-advisor.ts

@@ -16,6 +16,7 @@ import type {
 } from "@/lib/ai/schema";
 import { queryKeys } from "@/lib/query-keys";
 import { apiFetchRaw } from "@/lib/api/api-fetch";
+import type { BriefingFailureClass } from "@/lib/insights/briefing-failure-marker";
 
 /**
  * v1.4.16 phase D reconcile (CRITICAL C1 + C2) — shared TanStack Query
@@ -80,6 +81,13 @@ export interface InsightAdvisorPayload {
    * Absent on a pre-field cached payload → treated as "not failed" (no hint).
    */
   generationFailed?: boolean;
+  /**
+   * v1.25.3 — coarse class of the most recent failure (`timeout`, `auth`,
+   * `rate-limit`, `provider`, `format`, `unknown`), so the empty state can
+   * point its hint at the right lever. Null / absent → no specific hint, the
+   * generic failed-description holds.
+   */
+  generationFailureClass?: BriefingFailureClass | null;
 }
 
 /**
@@ -99,8 +107,7 @@ const ADVISOR_TIMEOUT_MS = 8_000;
  * discarded a slow-but-successful generation — the abort returned a null
  * payload, the mutation only invalidated (re-reading the OLD cache), yet
  * the server had often already written the fresh briefing. Give the force
- * branch the same 45 s budget the out-of-band warm job uses
- * (`COMPREHENSIVE_WARM_TIMEOUT_MS`) so a slow success is kept, not
+ * branch a generous 45 s client budget so a slow success is kept, not
  * dropped. The READ path stays fast at 8 s — only the explicit user tap
  * may wait. Per `.planning/v1.15.18-daily-briefing-audit.md` Fix 1b.
  */
@@ -291,6 +298,12 @@ export interface UseInsightsAdvisorResult {
    * when there is no last good text. Defaults false (no hint) when absent.
    */
   generationFailed: boolean;
+  /**
+   * v1.25.3 — coarse class of the most recent failure, so the empty state can
+   * point its hint at the right lever. Null when the last attempt succeeded or
+   * the payload predates the field.
+   */
+  generationFailureClass: BriefingFailureClass | null;
 }
 
 /**
@@ -379,5 +392,7 @@ export function useInsightsAdvisorQuery(
     // Absent (pre-field cached payload or unsettled query) → not failed, so a
     // transient unknown never flashes a false "couldn't refresh" hint.
     generationFailed: query.data?.payload?.generationFailed ?? false,
+    // Absent → no specific lever hint; the generic failed-description holds.
+    generationFailureClass: query.data?.payload?.generationFailureClass ?? null,
   };
 }

src/components/settings/ai/response-timeout-card.tsx

@@ -7,7 +7,8 @@
  * call (`CompletionParams.timeoutMs`). Surfaced here mainly for local /
  * self-hosted backends: an MLX/exo server can take >60 s on the first request
  * while it loads the model, which the legacy 60 s default timed out before the
- * first token landed. Empty = the generous built-in default (~180 s).
+ * first token landed. Empty = the built-in comprehensive-briefing default
+ * (~120 s, `AI_BUDGETS.comprehensive.timeoutMs`).
  * ──────────────────────────────────────────────────────────────── */
 
 import { useState } from "react";