fix: resilient standard.site publishing (retry + non-fatal)

STANDARD-SITE.md

@@ -32,7 +32,13 @@ and angular-buch.com); each parent repo supplies its own config.
 | `STANDARD_SITE_NAME` | no | Publication name. Default: the handle. |
 | `STANDARD_SITE_DESCRIPTION` | no | Publication description. |
 | `STANDARD_SITE_EXPECTED_DID` | no | Abort if the logged-in DID differs (guards against a wrong account). |
-| `STANDARD_SITE_ICON` | no | Publication profile image (square ≥256×256, png/jpg/webp). An http(s) URL or a local file path; uploaded as a blob and set as the publication `icon`. |
+| `STANDARD_SITE_ICON` | no | Publication profile image (square ≥256×256, png/jpg/webp). An http(s) URL or a local file path; uploaded as a blob and set as the publication `icon`. A missing/broken icon is non-fatal (warns and continues). |
+
+**Icon convention:** the icon is a website asset, not a shared-build asset. Put a
+square `standard-site-icon.png` in the consuming **website** repo's `public/` and
+point `STANDARD_SITE_ICON` at its deployed URL
+(e.g. `https://example.com/standard-site-icon.png`). Deploy the website before the
+article-data build runs so the URL resolves.
 | `STANDARD_SITE_SHOW_IN_DISCOVER` | no | `false` to opt out of the discovery feed. Default `true`. |
 | `STANDARD_SITE_DRY_RUN` | no | `true` logs the records that would be written/pruned without touching the PDS. |
 
@@ -68,6 +74,9 @@ and angular-buch.com); each parent repo supplies its own config.
   deleted.
 - `validate: false` is used on `putRecord` because a PDS does not natively know
   third-party lexicons like `site.standard.*`.
+- Resilient: transient PDS responses (429, 5xx) are retried with backoff, and a
+  publish failure is caught and logged rather than thrown — it never fails the
+  article-data build/deploy. Records catch up on the next successful build.
 
 ## Adding another website (e.g. angular-buch.com)
 

build.ts

@@ -75,11 +75,20 @@ async function build(): Promise<void> {
   console.log('\nValidating anchor links...');
   printValidationResults();
 
-  // Publish standard.site records (no-op unless configured via env)
-  await publishStandardSite([
-    { contentType: 'blog', entries: blogEntries },
-    { contentType: 'material', entries: materialEntries },
-  ]);
+  // Publish standard.site records (no-op unless configured via env). This is a
+  // secondary side-effect: a publish failure must never block the article-data
+  // build/deploy, so it is caught and logged rather than thrown.
+  try {
+    await publishStandardSite([
+      { contentType: 'blog', entries: blogEntries },
+      { contentType: 'material', entries: materialEntries },
+    ]);
+  } catch (error) {
+    console.error(
+      'standard.site: publishing failed (continuing build):',
+      error instanceof Error ? error.message : error,
+    );
+  }
 
   console.log('\nBuild complete!');
 }

standard-site/atproto.ts

@@ -10,11 +10,45 @@ export interface AtpSession {
   accessJwt: string;
 }
 
+/** How many attempts (1 initial + retries) for a transient failure. */
+const MAX_ATTEMPTS = 4;
+
+const sleep = (ms: number) => new Promise((resolve) => setTimeout(resolve, ms));
+
+/** 429 (rate limit) and 5xx (server) responses are worth retrying. */
+function isTransientStatus(status: number): boolean {
+  return status === 429 || status >= 500;
+}
+
+/**
+ * fetch with retries + exponential backoff for transient conditions (network
+ * errors, 429, 5xx). A PDS blip like a 502 should not fail the whole build.
+ */
+async function fetchWithRetry(url: string, init: RequestInit): Promise<Response> {
+  for (let attempt = 1; ; attempt++) {
+    let response: Response;
+    try {
+      response = await fetch(url, init);
+    } catch (error) {
+      if (attempt >= MAX_ATTEMPTS) {
+        throw error;
+      }
+      await sleep(250 * 2 ** (attempt - 1));
+      continue;
+    }
+    if (isTransientStatus(response.status) && attempt < MAX_ATTEMPTS) {
+      await sleep(250 * 2 ** (attempt - 1));
+      continue;
+    }
+    return response;
+  }
+}
+
 async function xrpc<T>(
   url: string,
   init: RequestInit,
 ): Promise<T> {
-  const response = await fetch(url, init);
+  const response = await fetchWithRetry(url, init);
   const text = await response.text();
   if (!response.ok) {
     throw new Error(`XRPC ${init.method} ${url} failed: ${response.status} ${text}`);
@@ -79,7 +113,7 @@ export async function uploadBlob(
   bytes: Uint8Array,
   mimeType: string,
 ): Promise<BlobRef> {
-  const response = await fetch(`${pds}/xrpc/com.atproto.repo.uploadBlob`, {
+  const response = await fetchWithRetry(`${pds}/xrpc/com.atproto.repo.uploadBlob`, {
     method: 'POST',
     headers: {
       'content-type': mimeType,