updated submit form

app.py

@@ -426,18 +426,25 @@ def submit_predictions(
     code_url: str,
     split: str,
 ) -> str:
+    # Only the file and agent name are required. Everything else is optional
+    # and gets filled with sensible defaults / "Anonymous" so a curious user
+    # can try a submission without committing to a full metadata profile.
     if file_obj is None:
         return "❌ **Missing file.** Please attach a predictions JSON."
-    if not agent_name.strip():
-        return "❌ **Missing agent name.**"
-    if not base_model.strip():
-        return "❌ **Missing base model.**"
-    if not organization.strip():
-        return "❌ **Missing organization.**"
-    if not contact_email.strip() or "@" not in contact_email:
-        return "❌ **Invalid contact email.**"
-    if not code_url.strip() or not code_url.startswith(("http://", "https://")):
-        return "❌ **Missing or invalid code URL.** A public `code_url` that reproduces your results is required."
+    if not agent_name or not agent_name.strip():
+        return "❌ **Missing agent name.** Please give your submission a short display name."
+
+    # Soft-validate optional fields — warn but don't block.
+    soft_warnings: list[str] = []
+    if contact_email.strip() and "@" not in contact_email:
+        soft_warnings.append("contact email looks malformed")
+    if code_url.strip() and not code_url.startswith(("http://", "https://")):
+        soft_warnings.append("code URL must start with http:// or https://")
+
+    # Normalise optionals with placeholders so downstream never sees empty strings.
+    base_model   = base_model.strip()   or "unspecified"
+    organization = organization.strip() or "Anonymous"
+    scaffold     = scaffold             or "none"
 
     try:
         src_path = Path(file_obj.name if hasattr(file_obj, "name") else file_obj)
@@ -501,16 +508,27 @@ def submit_predictions(
         else "📬 No notification channels configured on this Space — "
              "if you don't hear back in 10 days, please email the paper authors.\n\n"
     )
+    soft_line = ""
+    if soft_warnings:
+        soft_line = "⚠️ **Note:** " + "; ".join(soft_warnings) + ".\n\n"
+
+    email_line = (
+        f"We may email `{bundle['metadata']['contact_email']}` if we need to verify "
+        f"reproducibility via your `code_url`."
+        if bundle["metadata"]["contact_email"]
+        else "**Tip:** add a contact email next time so we can follow up about "
+             "reproducibility or questions."
+    )
 
     return (
         (warning_prefix + "\n\n" if warning_prefix else "")
+        + soft_line
         + f"✅ **Submission received** as `{fname}` for the **{split}** split "
         f"(**{len(predictions)}** entries).\n\n"
         + storage_line
         + notify_line
-        + f"A maintainer will score it against the private {split}-set answers and merge it to the "
-        f"leaderboard within ~1 week. We may email `{contact_email}` if we need to verify "
-        f"reproducibility via your `code_url`.\n\n"
+        + f"A maintainer will score it against the {split}-set gold answers and merge it to the "
+        f"leaderboard within ~1 week. " + email_line + "\n\n"
         f"**For a permanent public record,** please also open a PR to the "
         f"[benchmark repo]({REPO_URL}) with your predictions file under `submissions/`."
     )
@@ -576,11 +594,13 @@ def build_app() -> gr.Blocks:
             with gr.Tab("📤 Submit"):
                 gr.Markdown("## Submit your agent's predictions")
                 gr.Markdown(
-                    "Upload a JSON file containing **your agent's output** on DEEPSYNTH. "
-                    "The uploaded file must be the *predictions file* produced by running "
-                    "your agent on the split's questions — not your agent's source code, "
-                    "and not a raw transcript. We then score it against the private gold "
-                    "answers and add your row to the leaderboard."
+                    "Upload a JSON file containing **your agent's output** on either the "
+                    "**dev** (40 public tasks) or **test** (80 held-out tasks) split of DEEPSYNTH. "
+                    "For dev submissions, the evaluator scores against the publicly released gold "
+                    "answers. For test submissions, we score against our private gold answers and "
+                    "add your row to the held-out leaderboard. The uploaded file must be the "
+                    "*predictions file* produced by running your agent on the split's questions — "
+                    "not your agent's source code, and not a raw transcript."
                 )
 
                 gr.Markdown(
@@ -608,12 +628,21 @@ def build_app() -> gr.Blocks:
 
                 with gr.Row():
                     with gr.Column():
-                        agent_name_in    = gr.Textbox(label="Agent name", placeholder="e.g. ReAct-GPT5")
-                        base_model_in    = gr.Textbox(label="Base model", placeholder="e.g. gpt-5.2-pro (2026-02)")
+                        agent_name_in    = gr.Textbox(
+                            label="Agent name (required)",
+                            placeholder="e.g. ReAct-GPT5",
+                            info="Short display name shown on the leaderboard.",
+                        )
+                        base_model_in    = gr.Textbox(
+                            label="Base model",
+                            placeholder="e.g. gpt-5.2-pro (2026-02)",
+                            info="Optional. Defaults to 'unspecified'.",
+                        )
                         scaffold_in      = gr.Dropdown(
                             choices=["none", "ReAct", "CodeAct", "Plan-and-Execute", "Reflexion", "MCTS", "Custom"],
                             label="Scaffold",
-                            value="ReAct",
+                            value="none",
+                            info="Optional. Select the agent scaffold you used.",
                         )
                         split_in         = gr.Dropdown(
                             choices=["dev", "test"],
@@ -621,15 +650,24 @@ def build_app() -> gr.Blocks:
                             value="test",
                         )
                     with gr.Column():
-                        organization_in  = gr.Textbox(label="Organization", placeholder="e.g. MSR, Stanford, Google, etc.")
-                        contact_email_in = gr.Textbox(label="Contact email", placeholder="you@org.edu")
+                        organization_in  = gr.Textbox(
+                            label="Organization",
+                            placeholder="e.g. MSR India",
+                            info="Optional. Defaults to 'Anonymous'.",
+                        )
+                        contact_email_in = gr.Textbox(
+                            label="Contact email",
+                            placeholder="you@org.edu",
+                            info="Optional but recommended — we may email you about reproducibility.",
+                        )
                         code_url_in      = gr.Textbox(
-                            label="Code URL (required)",
+                            label="Code URL",
                             placeholder="https://github.com/you/your-agent",
+                            info="Optional. A public URL helps us accept your submission faster.",
                         )
 
                 predictions_in = gr.File(
-                    label="Predictions JSON (the output file produced by your agent)",
+                    label="Predictions JSON (required) — the output file produced by your agent",
                     file_types=[".json"],
                 )
                 submit_btn = gr.Button("Submit for review", variant="primary")