feat(web): auto-retry transient "Cannot reach Maple API" errors

apps/web/src/components/common/error-state.tsx

@@ -1,4 +1,5 @@
 import { Button } from "@maple/ui/components/ui/button"
+import { useNetworkAutoRetry } from "@/hooks/use-network-auto-retry"
 import { formatBackendError } from "@/lib/error-messages"
 import { cn } from "@maple/ui/lib/utils"
 
@@ -33,6 +34,13 @@ interface ErrorStateProps {
 export function ErrorState({ error, title, onRetry, variant = "panel", className }: ErrorStateProps) {
 	const formatted = formatBackendError(error)
 	const heading = title ?? formatted.title
+	// Connectivity blips self-heal: probe on a backoff poll + the `online`
+	// event, so recovery doesn't require the user to click or reload.
+	const autoRetrying = formatted.kind === "network" && onRetry !== undefined
+	useNetworkAutoRetry(autoRetrying, onRetry)
+	const description = autoRetrying
+		? `${formatted.description} Retrying automatically…`
+		: formatted.description
 
 	if (variant === "inline") {
 		return (
@@ -46,7 +54,7 @@ export function ErrorState({ error, title, onRetry, variant = "panel", className
 					</span>
 					<p className="text-xs font-medium text-foreground">{heading}</p>
 				</div>
-				<p className="text-xs whitespace-pre-wrap text-muted-foreground">{formatted.description}</p>
+				<p className="text-xs whitespace-pre-wrap text-muted-foreground">{description}</p>
 				{onRetry && (
 					<Button
 						size="sm"
@@ -72,7 +80,7 @@ export function ErrorState({ error, title, onRetry, variant = "panel", className
 				<DroppedSignalGlyph compact />
 				<div className="min-w-0 flex-1 basis-56 space-y-0.5">
 					<p className="text-sm font-medium text-foreground">{heading}</p>
-					<p className="line-clamp-2 text-xs text-muted-foreground">{formatted.description}</p>
+					<p className="line-clamp-2 text-xs text-muted-foreground">{description}</p>
 				</div>
 				{onRetry && (
 					<Button size="sm" variant="outline" className="shrink-0" onClick={onRetry}>
@@ -95,7 +103,7 @@ export function ErrorState({ error, title, onRetry, variant = "panel", className
 			</div>
 			<div className="max-w-xs space-y-1">
 				<p className="text-sm font-medium text-foreground">{heading}</p>
-				<p className="text-xs whitespace-pre-wrap text-muted-foreground">{formatted.description}</p>
+				<p className="text-xs whitespace-pre-wrap text-muted-foreground">{description}</p>
 			</div>
 			{onRetry && (
 				<Button size="sm" variant="outline" onClick={onRetry}>

apps/web/src/components/route-error.tsx

@@ -4,6 +4,7 @@ import { useEffect } from "react"
 import { AlertWarningIcon, CircleQuestionIcon, HouseIcon } from "@/components/icons"
 import { Button, buttonVariants } from "@maple/ui/components/ui/button"
 import { Empty, EmptyDescription, EmptyHeader, EmptyMedia, EmptyTitle } from "@maple/ui/components/ui/empty"
+import { useNetworkAutoRetry } from "@/hooks/use-network-auto-retry"
 import { formatBackendError } from "@/lib/error-messages"
 import { isChunkLoadError, shouldAttemptChunkReload } from "@/lib/chunk-reload"
 
@@ -17,9 +18,17 @@ function RouteError({ error, reset }: ErrorComponentProps) {
 		}
 	}, [isStaleChunk])
 
-	const { title, description } = formatBackendError(error)
+	const formatted = formatBackendError(error)
+	const { title, description } = formatted
 	const stack = error instanceof Error ? error.stack : undefined
 
+	const retry = () => {
+		reset()
+		router.invalidate()
+	}
+	// Route-loader transport failures self-heal without a manual reload.
+	useNetworkAutoRetry(formatted.kind === "network" && !isStaleChunk, retry)
+
 	return (
 		<Empty className="min-h-[60vh]">
 			<EmptyHeader>
@@ -38,8 +47,7 @@ function RouteError({ error, reset }: ErrorComponentProps) {
 							window.location.reload()
 							return
 						}
-						reset()
-						router.invalidate()
+						retry()
 					}}
 				>
 					{isStaleChunk ? "Reload" : "Try again"}

apps/web/src/hooks/use-network-auto-retry.ts

@@ -0,0 +1,62 @@
+import * as React from "react"
+
+import { useMountEffect } from "@/hooks/use-mount-effect"
+
+/** Poll backoff: 5s → 10s → 20s, then every 30s (±10% jitter). */
+const POLL_DELAYS_MS = [5_000, 10_000, 20_000, 30_000]
+
+/**
+ * While `enabled`, invokes `onRetry` automatically:
+ *
+ * - immediately when the browser fires the window `online` event (or the tab
+ *   becomes visible again),
+ * - on a backoff poll per {@link POLL_DELAYS_MS}, with jitter so several error
+ *   panels on one screen don't probe in lockstep.
+ *
+ * Poll ticks are skipped while the tab is hidden or the browser reports
+ * offline (the `online`/`visibilitychange` listeners fire the probe instead).
+ * The timer chain is mount-scoped; `enabled` and `onRetry` are read fresh at
+ * fire time. If a probe unmounts the host (error → loading → error remount),
+ * the backoff resets to 5s — effectively a steady poll during a long outage,
+ * which the 5s floor keeps cheap.
+ */
+export function useNetworkAutoRetry(enabled: boolean, onRetry: (() => void) | undefined): void {
+	const fire = React.useEffectEvent(() => {
+		if (enabled) onRetry?.()
+	})
+
+	useMountEffect(() => {
+		// React Doctor cannot infer that useMountEffect is an Effect; this is the
+		// canonical Effect Event pattern for a mount-scoped external subscription.
+		// oxlint-disable-next-line react-doctor/rules-of-hooks
+		const probe = () => fire()
+
+		let attempt = 0
+		let timeout: ReturnType<typeof setTimeout> | undefined
+
+		const schedule = () => {
+			const base = POLL_DELAYS_MS[Math.min(attempt, POLL_DELAYS_MS.length - 1)]
+			const delay = base * (0.9 + Math.random() * 0.2)
+			timeout = setTimeout(() => {
+				attempt += 1
+				if (!document.hidden && navigator.onLine !== false) probe()
+				schedule()
+			}, delay)
+		}
+
+		const onOnline = () => probe()
+		const onVisibilityChange = () => {
+			if (!document.hidden && navigator.onLine !== false) probe()
+		}
+
+		window.addEventListener("online", onOnline)
+		document.addEventListener("visibilitychange", onVisibilityChange)
+		schedule()
+
+		return () => {
+			window.removeEventListener("online", onOnline)
+			document.removeEventListener("visibilitychange", onVisibilityChange)
+			if (timeout !== undefined) clearTimeout(timeout)
+		}
+	})
+}

apps/web/src/lib/error-messages.test.ts

@@ -1,3 +1,4 @@
+import { HttpClientError, HttpClientRequest } from "effect/unstable/http"
 import { describe, expect, it } from "vitest"
 import { formatBackendError } from "./error-messages"
 
@@ -168,6 +169,27 @@ describe("formatBackendError", () => {
 		expect(result.title).toBe("Not authorized")
 	})
 
+	it("tags transport HttpClientError as a network error", () => {
+		const error = new HttpClientError.HttpClientError({
+			reason: new HttpClientError.TransportError({
+				request: HttpClientRequest.get("https://api.maple.dev/v1/services"),
+			}),
+		})
+		const result = formatBackendError(error)
+		expect(result.title).toBe("Cannot reach Maple API")
+		expect(result.kind).toBe("network")
+	})
+
+	it("tags fetch-failure Error messages as network errors", () => {
+		const result = formatBackendError(new Error("Failed to fetch"))
+		expect(result.title).toBe("Cannot reach Maple API")
+		expect(result.kind).toBe("network")
+	})
+
+	it("does not tag non-network errors", () => {
+		expect(formatBackendError(new Error("boom")).kind).toBeUndefined()
+	})
+
 	it("falls back for plain Error", () => {
 		const result = formatBackendError(new Error("boom"))
 		expect(result.title).toBe("Something went wrong")

apps/web/src/lib/error-messages.ts

@@ -5,6 +5,8 @@ import { isChunkLoadError } from "./chunk-reload"
 export interface FormattedError {
 	readonly title: string
 	readonly description: string
+	/** `"network"` = transient connectivity failure; safe to auto-retry. */
+	readonly kind?: "network"
 }
 
 const QUOTA_DESCRIPTIONS: Record<string, string> = {
@@ -250,6 +252,8 @@ export const formatBackendError = (input: unknown): FormattedError => {
 			return {
 				title: "Cannot reach Maple API",
 				description: "Check your connection — data will resume once the API is reachable.",
+				// A bad URL never self-heals — only transport failures auto-retry.
+				...(reasonTag === "TransportError" ? { kind: "network" as const } : {}),
 			}
 		}
 		if (status !== undefined && status >= 500) {
@@ -274,6 +278,7 @@ export const formatBackendError = (input: unknown): FormattedError => {
 			return {
 				title: "Cannot reach Maple API",
 				description: "Check your connection — data will resume once the API is reachable.",
+				kind: "network",
 			}
 		}
 		return {

apps/web/src/lib/services/common/atom-client.ts

@@ -4,6 +4,7 @@ import { Effect } from "effect"
 import { HttpClient, HttpClientError } from "effect/unstable/http"
 import { apiBaseUrl } from "./api-base-url"
 import { MapleFetchHttpClientLive } from "./http-client"
+import { isRetryableTransportError, mapleRetrySchedule } from "./retry-policy"
 
 export class MapleApiAtomClient extends AtomHttpApi.Service<MapleApiAtomClient>()(
 	"@maple/web/services/common/MapleApiAtomClient",
@@ -27,11 +28,15 @@ export class MapleApiAtomClient extends AtomHttpApi.Service<MapleApiAtomClient>(
 					),
 				HttpClient.retry({
 					times: 3,
+					schedule: mapleRetrySchedule,
 					while: (error) => {
+						// Transient network failures (idempotent requests only) self-heal
+						// with backoff instead of failing fast to the error UI.
+						if (isRetryableTransportError(error)) return true
 						if (!HttpClientError.isHttpClientError(error)) return false
 						const status = error.response?.status
 						if (status === undefined) return false
-						// Only retry on 500/502/503 — not 504 (timeout) or undefined (network failure)
+						// Retry on 500/502/503 — not 504 (query timeout, won't get faster)
 						if (status >= 500 && status < 600 && status !== 504) return true
 						// Billing reads (customer/usage/plans) can fire during the Clerk
 						// token-settle window where getToken() is transiently null → the

apps/web/src/lib/services/common/retry-policy.test.ts

@@ -0,0 +1,40 @@
+import { HttpClientError, HttpClientRequest } from "effect/unstable/http"
+import { describe, expect, it } from "vitest"
+import { isRetryableTransportError } from "./retry-policy"
+
+const transportError = (request: HttpClientRequest.HttpClientRequest, cause?: unknown) =>
+	new HttpClientError.HttpClientError({
+		reason: new HttpClientError.TransportError({ request, cause }),
+	})
+
+describe("isRetryableTransportError", () => {
+	it("retries transport failures on idempotent requests", () => {
+		const error = transportError(HttpClientRequest.get("https://api.maple.dev/v1/services"))
+		expect(isRetryableTransportError(error)).toBe(true)
+	})
+
+	it("never replays mutations", () => {
+		const error = transportError(HttpClientRequest.post("https://api.maple.dev/v1/dashboards"))
+		expect(isRetryableTransportError(error)).toBe(false)
+	})
+
+	it("does not multiply the client timeout", () => {
+		const error = transportError(
+			HttpClientRequest.get("https://api.maple.dev/v1/services"),
+			new DOMException("timed out", "TimeoutError"),
+		)
+		expect(isRetryableTransportError(error)).toBe(false)
+	})
+
+	it("does not replay aborted requests", () => {
+		const error = transportError(
+			HttpClientRequest.get("https://api.maple.dev/v1/services"),
+			new DOMException("aborted", "AbortError"),
+		)
+		expect(isRetryableTransportError(error)).toBe(false)
+	})
+
+	it("ignores non-HttpClientError values", () => {
+		expect(isRetryableTransportError(new Error("Failed to fetch"))).toBe(false)
+	})
+})

apps/web/src/lib/services/common/retry-policy.ts

@@ -0,0 +1,24 @@
+import { Schedule } from "effect"
+import { HttpClientError } from "effect/unstable/http"
+
+const IDEMPOTENT_METHODS = new Set(["GET", "HEAD", "OPTIONS"])
+
+/**
+ * Transient network failure worth replaying: a `TransportError` (fetch failed,
+ * DNS, connection reset) on an idempotent request. Transport failures can occur
+ * after the request reached the server, so mutations are never replayed. The
+ * 45s client timeout (`AbortSignal.timeout` in http-client.ts) also surfaces as
+ * a TransportError — excluded so one hang doesn't multiply into several.
+ */
+export const isRetryableTransportError = (error: unknown): boolean => {
+	if (!HttpClientError.isHttpClientError(error)) return false
+	if (error.reason._tag !== "TransportError") return false
+	const cause = error.reason.cause
+	if (cause instanceof DOMException && (cause.name === "TimeoutError" || cause.name === "AbortError")) {
+		return false
+	}
+	return IDEMPOTENT_METHODS.has(error.request.method.toUpperCase())
+}
+
+/** Backoff between HTTP-layer retry attempts: 300ms → 600ms → 1.2s. */
+export const mapleRetrySchedule = Schedule.exponential("300 millis")

apps/web/src/lib/services/common/v2-atom-client.ts

@@ -4,6 +4,7 @@ import { Effect } from "effect"
 import { HttpClient, HttpClientError } from "effect/unstable/http"
 import { apiBaseUrl } from "./api-base-url"
 import { MapleFetchHttpClientLive } from "./http-client"
+import { isRetryableTransportError, mapleRetrySchedule } from "./retry-policy"
 
 /** Typed dashboard client for the public, stability-committed v2 API. */
 export class MapleApiV2AtomClient extends AtomHttpApi.Service<MapleApiV2AtomClient>()(
@@ -22,7 +23,11 @@ export class MapleApiV2AtomClient extends AtomHttpApi.Service<MapleApiV2AtomClie
 					),
 				HttpClient.retry({
 					times: 3,
+					schedule: mapleRetrySchedule,
 					while: (error) => {
+						// Transient network failures (idempotent requests only) self-heal
+						// with backoff instead of failing fast to the error UI.
+						if (isRetryableTransportError(error)) return true
 						if (!HttpClientError.isHttpClientError(error)) return false
 						const status = error.response?.status
 						return status !== undefined && status >= 500 && status < 600 && status !== 504