diff --git a/output_writer.py b/output_writer.py
index 13b43c1..180f476 100644
--- a/output_writer.py
+++ b/output_writer.py
@@ -394,6 +394,33 @@ def write_csv(rows: Sequence[Any], path: str, row_cls: Type = Product) -> None:
 EXIT_PARTIAL = 6
 
 
+# Exit code for a run that never GOT its pages: a navigation timeout, a dead
+# or unauthenticated proxy, a DNS failure, or an edge answering with
+# something that is not the page that was asked for.
+#
+# Distinct from EXIT_NO_PRODUCTS because those are opposite facts. Exit 4 is
+# a statement about the CATALOGUE — "we asked, and the answer was nothing" —
+# so handing it to a run that never reached the site tells a pipeline the
+# listing is empty when nothing was read at all.
+#
+# 5 rather than a new number, and 5 rather than EXIT_PARTIAL:
+#
+#   * this family's contract already reserves 5 for a transport failure
+#     (scraper_api_client has used it for a remote API error since it was
+#     written), so this needs no new code and no per-repo table for a caller
+#     driving more than one of these scrapers;
+#   * EXIT_PARTIAL (6) means "some rows were gathered and the output is
+#     incomplete". A run holding nothing writes no output at all, so a
+#     consumer that reads the file on a 6 finds either nothing or the
+#     PREVIOUS run's good data, which `save` deliberately does not
+#     overwrite. Exit 5 promises no file.
+#
+# Deliberately NOT applied when rows WERE gathered: a timeout on page 7 of
+# 10 is a partial run (exit 6, output written), which is already right. This
+# decides only what a run holding nothing reports.
+EXIT_FETCH_FAILED = 5
+
+
 def write_run_meta(out_prefix: str, meta: dict) -> str:
     """Write a run-metadata sidecar next to the output, return its path.
 
@@ -589,7 +616,7 @@ def finish_run(rows: Sequence[Any], out_prefix: str, fmt: str,
             print(f"[!] Failed run: 0 of {pages_requested} page(s) were "
                   f"fetched ({stop_reason}). This is NOT an empty result — "
                   f"nothing was read from the site at all.")
-            return EXIT_PARTIAL
+            return EXIT_FETCH_FAILED
         return rc
     if not complete:
         print(f"[!] Partial run: stopped after {pages_completed} of "
diff --git a/smoke_test.py b/smoke_test.py
index 1793bf1..c3b0435 100644
--- a/smoke_test.py
+++ b/smoke_test.py
@@ -73,7 +73,7 @@
 import page_flow
 import product_parser
 from diff_runs import diff_products
-from output_writer import (Market, Product, save, finish_run, write_csv,
+from output_writer import (EXIT_FETCH_FAILED, Market, Product, save, finish_run, write_csv,
                            run_meta, dedupe_by_key, dedupe_by_sku,
                            merge_pages, is_deeper,
                            ROW_CLASS_BY_MODE, UNIQUE_BY_SKU_MODES,
@@ -1213,13 +1213,13 @@ def run(rows, stop_reason, blocked=False, allow_empty=False):
         # pipeline branching on the exit code, which is what this family
         # says exit codes are for, would have recorded an empty catalogue.
         code, meta = run([], "page_load_timeout", allow_empty=True)
-        ok &= check("0 rows because nothing was FETCHED is exit 6, not 4",
-                    code == EXIT_PARTIAL)
+        ok &= check("0 rows because nothing was FETCHED is exit 5, not 4",
+                    code == EXIT_FETCH_FAILED)
         ok &= check("and the sidecar says failed, not complete",
                     meta is not None and meta["status"] == "failed")
         code, meta = run([], "next_batch_refused", allow_empty=True)
-        ok &= check("a refused batch with no rows is exit 6 too",
-                    code == EXIT_PARTIAL)
+        ok &= check("a refused batch with no rows is exit 5 too",
+                    code == EXIT_FETCH_FAILED)
         code, meta = run([], "blocked_cloudflare", blocked=True, allow_empty=True)
         ok &= check("but a BLOCK still outranks both, at exit 3",
                     code == EXIT_BLOCKED)