Some research

tests/rag.py

@@ -108,7 +108,13 @@ class RagIndexAndSearchTest(TestCase):
 
     name = "rag-index-and-search"
     mode = "pipe"
-    timeout = 60.0 * 8  # two LLM round trips per sampled file now (save + per-file search)
+    # Per file: one save round trip (up to SAVE_TIMEOUT, observed free-tier backends taking
+    # 30-60s even on success) + one search round trip (no LLM call, local FTS, cheap) + process
+    # wait/spawn overhead. This must scale with SAMPLE_SIZE (env-configurable, default 5) -- a
+    # flat constant sized for an earlier, smaller default silently became too tight and made a
+    # slow-but-successful free-tier response get killed as a false "timeout" failure.
+    SAVE_TIMEOUT = 180.0
+    timeout = SAMPLE_SIZE * (SAVE_TIMEOUT + 10.0 + 15.0 + 15.0 + 10.0) + 60.0
 
     async def setup(self):
         self.tmp_home = tempfile.mkdtemp(prefix="tai-test-home-")
@@ -131,7 +137,7 @@ class RagIndexAndSearchTest(TestCase):
                 % (title, sent_content)
             )
             proc = await self.spawn(extra_args=["--yes", "--profile", "ragtest", prompt])
-            await proc.drain_until_eof(timeout=90.0)
+            await proc.drain_until_eof(timeout=self.SAVE_TIMEOUT)
             await proc.wait(timeout=10.0)
 
             record = self._find_record_by_title(title)