tell the user the submit button is working, and stop claiming a format check (#15)

- tell the user the submit button is working, and stop claiming a format check (34c4bfc4f1e6082f070348534e53df3a7dae5c31)

app.py

@@ -621,7 +621,7 @@ stronger, upload both and submit only the one you want scored.
 
 We evaluate the exact image recorded at registration. A later push to the same repository
 does **not** change a submission already made — submit again if you want the newer image
-evaluated. Use **Validate format (no submit)** to check your fields without spending one.
+evaluated.
 
 ### After you submit
 
@@ -879,7 +879,19 @@ def test_submit_tab():
                             interactive=config.TEST_PHASE_OPEN)
         out = gr.Markdown()
 
+        # Three steps, not one. The call takes about 20 seconds and said nothing
+        # while it ran, so a second click looked like the reasonable thing to
+        # do: two identical submissions landed 8 seconds apart on 2026-08-08.
+        # Disabling the button is the half that prevents it; the message is the
+        # half that stops the participant wanting to.
         btn.click(
+            lambda: (
+                gr.update(value="Submitting...", interactive=False),
+                "⏳ Submitting. This takes up to a minute, please do not click "
+                "again or reload.",
+            ),
+            outputs=[btn, out],
+        ).then(
             do_test_submit,
             inputs=[
                 reg_id, model_name, license_str, open_weight,
@@ -889,6 +901,11 @@ def test_submit_tab():
             # Serialize so the per-track cap check cannot be raced by one team firing
             # two registrations at once. The worker enforces the cap again anyway.
             concurrency_limit=1,
+        ).then(
+            # Re-enabled whether the submission succeeded or failed: a refusal a
+            # team cannot retry is worse than the refusal itself.
+            lambda: gr.update(value="Submit for evaluation", interactive=True),
+            outputs=btn,
         )
 
 
@@ -989,9 +1006,7 @@ def _subs_view(profile: gr.OAuthProfile | None):
 def my_submissions_tab():
     gr.Markdown("## My Submissions")
     gr.Markdown(
-        "Track your own submissions through the pipeline. **Format is checked "
-        "instantly when you submit**, so a *Pending* entry has already passed "
-        "format validation and is just awaiting scoring; only *Scored* entries "
+        "Track your own submissions through the pipeline. Only *Scored* entries "
         "appear on the leaderboard."
     )
     login = gr.LoginButton(scale=0)

tests/test_val_phase_closed.py

@@ -389,3 +389,73 @@ class CredentialsPanelShapeTest(unittest.TestCase):
     def test_the_secret_is_shown_once_and_only_in_the_login_block(self) -> None:
         body = self.panel()
         self.assertEqual(body.count("not-a-real-secret"), 2, "table row plus export")
+
+
+class SubmitButtonFeedbackTest(unittest.TestCase):
+    """The submit button must say something before the 20-second call returns.
+
+    Silence reads as "the click did not register", so the reasonable next move
+    is to click again. That is what produced two identical test submissions 8
+    seconds apart on 2026-08-08, and a FAILED submission consumes one of a
+    team's three slots.
+
+    Read as source: app.py needs gradio, which is not installed everywhere the
+    tests run, and a test that skips itself on a missing import reports nothing
+    while looking green.
+    """
+
+    def setUp(self) -> None:
+        path = os.path.join(
+            os.path.dirname(os.path.abspath(__file__)), os.pardir, "app.py"
+        )
+        with open(path) as fh:
+            self.src = fh.read()
+        block = self.src[self.src.index("def test_submit_tab("):]
+        self.block = block[: block.index("\ndef ", 1)]
+
+    def test_the_click_chain_gives_immediate_feedback(self) -> None:
+        self.assertIn("Submitting...", self.block)
+        self.assertIn(".then(", self.block, "feedback then work, not one handler")
+
+    def test_the_button_is_disabled_while_it_runs(self) -> None:
+        i = self.block.index("Submitting...")
+        self.assertIn("interactive=False", self.block[i - 200 : i + 200])
+
+    def test_it_is_re_enabled_afterwards(self) -> None:
+        """Including after a refusal: a team must be able to fix and retry."""
+        tail = self.block[self.block.rindex(".then(") :]
+        self.assertIn("interactive=True", tail)
+        self.assertIn("Submit for evaluation", tail)
+
+
+class NoFormatCheckClaimedForTheTestPhaseTest(unittest.TestCase):
+    """Nothing format-checks a test submission, so nothing may say it does.
+
+    A test submission is an image digest, not a predictions file. Claiming a
+    check that does not run tells a team its submission is further along than it
+    is, and right now no worker processes test requests at all.
+    """
+
+    def setUp(self) -> None:
+        path = os.path.join(
+            os.path.dirname(os.path.abspath(__file__)), os.pardir, "app.py"
+        )
+        with open(path) as fh:
+            self.src = fh.read()
+
+    def test_my_submissions_does_not_claim_a_format_check(self) -> None:
+        block = self.src[self.src.index("def my_submissions_tab("):]
+        # The function may be the last one in the file, so slicing to the next
+        # `def` has to tolerate there not being one. Getting this wrong makes
+        # the test error rather than fail, which reads as a broken test rather
+        # than as a broken claim.
+        nxt = block.find("\ndef ", 1)
+        if nxt != -1:
+            block = block[:nxt]
+        self.assertNotIn("Format is checked", block)
+        self.assertNotIn("passed format validation", block)
+
+    def test_the_test_guide_does_not_offer_a_control_it_lacks(self) -> None:
+        guide = self.src[self.src.index("_TEST_GUIDE_MD = "):]
+        guide = guide[: guide.index('"""', guide.index('"""') + 3)]
+        self.assertNotIn("Validate format", guide)