diff --git a/src/openai.ts b/src/openai.ts
index 022c197..679cc4a 100644
--- a/src/openai.ts
+++ b/src/openai.ts
@@ -352,7 +352,7 @@ const EMBEDDINGS_TIMEOUT_MS = 20_000;
 // We do not track "last read" to keep writes minimal.
 const EMBEDDINGS_CACHE_EVICT_BATCH = 512;
 const EMBEDDINGS_CACHE_EVICT_MAX_BATCH = 8192;
-const EMBEDDINGS_CACHE_QUOTA_RETRY_ATTEMPTS = 4;
+const EMBEDDINGS_CACHE_QUOTA_MAX_RETRIES = 4;
 const EMBEDDINGS_JOB_TTL_MS = 24 * 60 * 60_000;
 const EMBEDDINGS_JOB_LOCK_MS = 30_000;
 
@@ -435,13 +435,14 @@ const isEmbeddingsCacheQuotaError = (error: unknown): boolean => {
   const message = error instanceof Error ? error.message : typeof error === "string" ? error : "";
   const combined = `${name} ${message}`.toLowerCase();
   if (!combined) return false;
-  return combined.includes("quota") ||
-    combined.includes("insufficient") && combined.includes("storage") ||
-    combined.includes("insufficient") && combined.includes("space") ||
+  return (
+    combined.includes("quota") ||
+    (combined.includes("insufficient") && combined.includes("storage")) ||
+    (combined.includes("insufficient") && combined.includes("space")) ||
     combined.includes("no space") ||
     combined.includes("storage limit") ||
-    combined.includes("storage") && combined.includes("exceeded") ||
-    combined.includes("limit") && combined.includes("exceeded");
+    (combined.includes("storage") && combined.includes("exceeded"))
+  );
 };
 
 const writeEmbeddingsCacheEntry = async (
@@ -494,7 +495,8 @@ const writeEmbeddingsCacheEntryBestEffort = async (
   deadlineMs: number,
 ): Promise<{ isNew: boolean }> => {
   let evictBatch = EMBEDDINGS_CACHE_EVICT_BATCH;
-  for (let attempt = 0; attempt <= EMBEDDINGS_CACHE_QUOTA_RETRY_ATTEMPTS; attempt += 1) {
+  // Attempts = 1 (initial write) + max retries.
+  for (let attempt = 0; attempt <= EMBEDDINGS_CACHE_QUOTA_MAX_RETRIES; attempt += 1) {
     if (Date.now() >= deadlineMs) return { isNew: false };
     try {
       return await writeEmbeddingsCacheEntry(kv, cacheModelKey, hash, embedding, createdAtMs);
@@ -510,7 +512,7 @@ const writeEmbeddingsCacheEntryBestEffort = async (
         console.warn(
           `[ai.ubq.fi] embeddings_cache quota eviction model=${cacheModelKey} evicted=${evicted.evicted_embeddings} stale_index_deleted=${evicted.deleted_stale_index_keys} batch=${evictBatch}`,
         );
-        if (evicted.evicted_embeddings <= 0) return { isNew: false };
+        if (evicted.evicted_embeddings <= 0 && evicted.deleted_stale_index_keys <= 0) return { isNew: false };
       } catch (evictError) {
         console.warn("[ai.ubq.fi] embeddings_cache quota eviction failed:", evictError);
         return { isNew: false };
@@ -1604,13 +1606,12 @@ export const handleEmbeddings = async (req: Request, usageContext?: UsageContext
         });
       }
 
-      let wroteNew = 0;
       for (let i = 0; i < chunkItems.length; i += 1) {
         const item = chunkItems[i]!;
         const vec = vectors[i]!;
         for (const idx of item.indices) vectorsByIndex[idx] = vec;
         if (shouldCache && kv) {
-          const result = await writeEmbeddingsCacheEntryBestEffort(
+          await writeEmbeddingsCacheEntryBestEffort(
             kv,
             cacheModelKey,
             item.hash,
@@ -1618,7 +1619,6 @@ export const handleEmbeddings = async (req: Request, usageContext?: UsageContext
             Date.now(),
             deadlineMs,
           );
-          if (result.isNew) wroteNew += 1;
         }
       }
     }