fix: fetch platform originals without exporting

--fetch also exported and stopped on the absent dist/truth before any
download, so CI skipped every exporter fidelity case and tolerated the
step. --refresh-cache only downloads: the step can no longer fail
unseen.

.github/workflows/validate.yml

@@ -278,12 +278,12 @@ jobs:
       - run: npm ci --prefix website
 
       # The exporter fidelity cases compare what we hand back against the
-      # file each platform publishes, so they need those files. They skip
-      # without them rather than fail, which is why an unreachable upstream
-      # must not stop the run: the rest of the suite still has to report.
+      # file each platform publishes, so they need those files. --fetch
+      # exported as well and stopped on the absent dist/truth before any
+      # download, so the cases skipped on every run: --refresh-cache only
+      # downloads them.
       - name: Fetch the platforms' own files
-        continue-on-error: true
-        run: python scripts/export_native.py --all --fetch --output-dir "$RUNNER_TEMP/native"
+        run: python scripts/export_native.py --all --refresh-cache
 
       # A pack built without its data directories exits non-zero: it is not
       # the pack a user gets. Tests that build packs need the same trees a

scripts/pr_checks.py

@@ -50,8 +50,8 @@ STEPS: tuple[Step, ...] = (
          needs="vermin", provider="the pip line of CONTRIBUTING.md"),
     Step("shellcheck --version", "the launcher test runs shellcheck", needs="shellcheck",
          provider="the shellcheck package"),
-    Step("python scripts/export_native.py --all --fetch --output-dir tmp/native",
-         "files the exporter tests compare with", may_fail=True),
+    Step("python scripts/export_native.py --all --refresh-cache",
+         "files the exporter tests compare with"),
     Step("python scripts/refresh_data_dirs.py", "data directories the pack-building tests read"),
     Step("python scripts/generate_site.py", "the pages the site tests read"),
     Step("npm ci --prefix website", "the site's packages, Node 24",

tests/test_contributing_text.py

@@ -240,7 +240,6 @@ class OneCommandRunsTheChecks(unittest.TestCase):
 
         workflow = (REPO_ROOT / ".github" / "workflows" / "validate.yml").read_text(encoding="utf-8")
         tolerated = set(re.findall(r"continue-on-error: true\n\s*run: (\S+ \S+)", workflow))
-        self.assertTrue(tolerated)
         for step in pr_checks.STEPS:
             with self.subTest(step=step.command):
                 self.assertEqual(step.may_fail, " ".join(step.command.split()[:2]) in tolerated)

wiki/release-process.md

@@ -144,7 +144,7 @@ upstream. An edited profile is not held to it.
 
 **run-tests.** On Python 3.8, 3.9 and 3.12: holds `install.py` to 3.8 with
 `vermin`, sets up bubblewrap for the sandbox tests, fetches the platforms' own
-files (allowed to fail, the fidelity tests then skip) and the data
+files (`export_native.py --all --refresh-cache`, for the fidelity tests) and the data
 directories, generates the site, then runs
 `python -m unittest discover -s tests -t . -v` with `RETROBIOS_OFFLINE=1`, the
 full-screen installer tests again under 3.8, and the site build

wiki/testing-guide.md

@@ -540,9 +540,9 @@ of main passes, not every commit under it.
 Modules that need real artifacts skip themselves when those artifacts are absent,
 so `test_pack_integrity` is a no-op in CI (no `dist/`) and a real check locally.
 The exporter fidelity cases go the other way: they compare with each platform's
-own file, which `python scripts/export_native.py --all --fetch --output-dir
-tmp/native` downloads into `.cache/upstream-native/`. CI runs it before the
-suite, and a clone without that cache skips them.
+own file, which `python scripts/export_native.py --all --refresh-cache`
+downloads into `.cache/upstream-native/`. CI runs it before the suite, and a
+clone without that cache skips them.
 Run `python -m unittest discover -s tests -t .` before pushing.
 
 Tests must pass before merge. If a test fails in CI, reproduce locally with: