Move submissions to Space pull requests, drop GitHub links

README.md

@@ -22,12 +22,14 @@ web, desktop (Ubuntu / Windows / macOS) and mobile (Android / iOS). Every trajec
 success, and every step carries a human label.
 
 - **Data**: https://huggingface.co/datasets/ZJUSCL/CUAStepBench
-- **Submissions**: https://github.com/ZJUSCL/CUAStepBench
+- **Submissions**: the Community tab of this Space
 - **Paper logs**: https://huggingface.co/buckets/ZJUSCL/CUAStepBench_paperlogs
 
-## Updating the leaderboard
+## Submitting a result
 
-All rows live in `results.json`. Adding a submission means appending one entry there, nothing else.
+Submissions are pull requests against this Space. Open the **Community** tab and start a new pull request, or open
+`results.json`, edit it in the browser, and choose *Open as a pull request*. Append one entry to `entries` and change
+nothing else. The page rebuilds itself once the pull request is merged.
 
 ```json
 {
@@ -43,8 +45,20 @@ All rows live in `results.json`. Adding a submission means appending one entry t
 }
 ```
 
-`track` is `closed` or `open`. A metric that the framework does not produce is `null` and renders as a dash. `step_note`
-and `model_note` take `dagger` or `ddagger`, whose text lives under `notes` in the same file. `source` links to the raw
-LLM API calls and per-task `result.json` outputs behind the row, on any host that needs no login.
+`track` is `closed` or `open`. A framework that issues any closed-source API call is `closed`, even when the rest of the
+pipeline runs on open weights. All 8 metrics are percentages at one decimal place, and a level the framework does not
+produce is `null` rather than omitted, which renders as a dash. `step_note` and `model_note` take `dagger` or
+`ddagger`, whose text lives under `notes` in the same file.
 
-Sorting, the per-table best-value highlight and the footnotes are all derived from `results.json` by `index.html`.
+`source` holds every LLM API call behind the numbers together with the per-task `judge/<your-judge-name>/result.json`
+outputs, on any host that needs no login. Each call carries its `task_id`, model identifier, the full request (system
+prompt, messages, and which images were passed), the raw response, token counts, and a timestamp. Any layout works
+provided every call maps unambiguously to its task.
+
+The pull request description states whether step labels are native or come from a step extraction procedure, and how the
+run was produced. All 278 tasks must be covered, with no re-annotation and no subset selection.
+
+## Layout
+
+`results.json` is the only data file. Sorting, the per-table best-value highlight, the stat blocks and the footnotes are
+all derived from it by `index.html`, which has no dependencies and no build step.

index.html

@@ -128,14 +128,32 @@
     procedure was applied. Rows are ordered by trajectory F1 by default; click any column header to re-sort.</p>
 
     <h3>Submitting a result</h3>
-    <p>Open a pull request against
-    <a id="a-github" href="https://github.com/ZJUSCL/CUAStepBench">github.com/ZJUSCL/CUAStepBench</a> with the row and a
-    public link to your logs. The row carries all 8 metrics at one decimal place, the release month, and a
-    <code>Source</code> link.</p>
-    <pre><code>| 2026-09 | MyJudge | MyModel-7B | 75.0 | 74.1 | 78.2 | 76.1 | 90.1 | 46.0 | 35.2 | 39.9 | [logs](https://example.com/my-judge-logs) |</code></pre>
-    <p>The <code>Source</code> link holds every LLM API call behind the numbers together with the per-task
-    <code>judge/&lt;your-judge-name&gt;/result.json</code> outputs, on any host that needs no login. All rows currently on
-    this page point at the paper logs. All tasks must be covered, with no re-annotation and no subset selection.</p>
+    <p>Submissions are pull requests against this Space. Open the <b>Community</b> tab and start a new pull request, or
+    open <code>results.json</code>, edit it in the browser, and choose <i>Open as a pull request</i>. Append one entry to
+    <code>entries</code> and change nothing else. The page rebuilds itself once the pull request is merged.</p>
+    <pre><code>{
+  "track": "open",
+  "release": "2026-09",
+  "framework": "MyJudge",
+  "model": "MyModel-7B",
+  "model_note": null,
+  "traj": {"acc": 75.0, "prec": 74.1, "recall": 78.2, "f1": 76.1},
+  "step": {"acc": 90.1, "prec": 46.0, "recall": 35.2, "f1": 39.9},
+  "step_note": null,
+  "source": "https://example.com/my-judge-logs"
+}</code></pre>
+    <p><code>track</code> is <code>closed</code> or <code>open</code>. A framework that issues any closed-source API call
+    is <code>closed</code>, even when the rest of the pipeline runs on open weights. All 8 metrics are percentages at one
+    decimal place, and a level the framework does not produce is <code>null</code> rather than omitted.</p>
+    <p><code>source</code> holds every LLM API call behind the numbers together with the per-task
+    <code>judge/&lt;your-judge-name&gt;/result.json</code> outputs, on any host that needs no login. Each call carries its
+    <code>task_id</code>, model identifier, the full request (system prompt, messages, and which images were passed), the
+    raw response, token counts, and a timestamp. Any layout works provided every call maps unambiguously to its task. All
+    rows currently on this page point at the paper logs.</p>
+    <p>The pull request description states whether step labels are native or come from a step extraction procedure, and
+    how the run was produced. All <span id="n-tasks-2">278</span> tasks must be covered, with no re-annotation and no
+    subset selection.</p>
+    <p><a id="a-community" href="https://huggingface.co/spaces/ZJUSCL/CUAStepBench_Leaderboard/discussions" target="_blank" rel="noopener">Open the Community tab ↗</a></p>
   </div>
 </div>
 
@@ -221,7 +239,7 @@ fetch("results.json?v=" + Date.now()).then(r => r.json()).then(data => {
 
   document.getElementById("links").innerHTML = [
     ["Dataset", data.links.data],
-    ["GitHub / submissions", data.links.github],
+    ["Submit a result", data.links.submit],
     ["Paper logs", data.links.paper_logs]
   ].map(([t, u]) => `<a href="${u}" target="_blank" rel="noopener">${t} ↗</a>`).join("");
 
@@ -230,6 +248,7 @@ fetch("results.json?v=" + Date.now()).then(r => r.json()).then(data => {
   document.getElementById("s-entries").textContent = data.entries.length;
   document.getElementById("s-updated").textContent = data.updated;
   document.getElementById("n-tasks").textContent = data.total_tasks;
+  document.getElementById("n-tasks-2").textContent = data.total_tasks;
 
   renderTable(document.getElementById("t-closed"), data.entries.filter(e => e.track === "closed"), "traj.f1", "desc");
   renderTable(document.getElementById("t-open"), data.entries.filter(e => e.track === "open"), "traj.f1", "desc");

results.json

@@ -5,7 +5,7 @@
   "updated": "2026-08",
   "links": {
     "data": "https://huggingface.co/datasets/ZJUSCL/CUAStepBench",
-    "github": "https://github.com/ZJUSCL/CUAStepBench",
+    "submit": "https://huggingface.co/spaces/ZJUSCL/CUAStepBench_Leaderboard/discussions",
     "paper_logs": "https://huggingface.co/buckets/ZJUSCL/CUAStepBench_paperlogs"
   },
   "notes": {
@@ -19,8 +19,18 @@
       "framework": "Codex",
       "model": "GPT-5.5",
       "model_note": null,
-      "traj": {"acc": 79.1, "prec": 87.1, "recall": 70.1, "f1": 77.7},
-      "step": {"acc": null, "prec": null, "recall": null, "f1": null},
+      "traj": {
+        "acc": 79.1,
+        "prec": 87.1,
+        "recall": 70.1,
+        "f1": 77.7
+      },
+      "step": {
+        "acc": null,
+        "prec": null,
+        "recall": null,
+        "f1": null
+      },
       "step_note": null,
       "source": "https://huggingface.co/buckets/ZJUSCL/CUAStepBench_paperlogs"
     },
@@ -30,8 +40,18 @@
       "framework": "SeekJudge",
       "model": "DeepSeek-V4-Pro + Gemini-3.0-Flash",
       "model_note": "ddagger",
-      "traj": {"acc": null, "prec": null, "recall": null, "f1": 73.6},
-      "step": {"acc": null, "prec": null, "recall": null, "f1": 44.0},
+      "traj": {
+        "acc": null,
+        "prec": null,
+        "recall": null,
+        "f1": 73.6
+      },
+      "step": {
+        "acc": null,
+        "prec": null,
+        "recall": null,
+        "f1": 44.0
+      },
       "step_note": null,
       "source": "https://huggingface.co/buckets/ZJUSCL/CUAStepBench_paperlogs"
     },
@@ -41,8 +61,18 @@
       "framework": "CUAJudge",
       "model": "GPT-5-mini",
       "model_note": null,
-      "traj": {"acc": 71.1, "prec": 78.6, "recall": 61.1, "f1": 68.8},
-      "step": {"acc": 90.4, "prec": 49.7, "recall": 30.8, "f1": 38.0},
+      "traj": {
+        "acc": 71.1,
+        "prec": 78.6,
+        "recall": 61.1,
+        "f1": 68.8
+      },
+      "step": {
+        "acc": 90.4,
+        "prec": 49.7,
+        "recall": 30.8,
+        "f1": 38.0
+      },
       "step_note": "dagger",
       "source": "https://huggingface.co/buckets/ZJUSCL/CUAStepBench_paperlogs"
     },
@@ -52,8 +82,18 @@
       "framework": "SeekJudge",
       "model": "SeekJudge-9B",
       "model_note": null,
-      "traj": {"acc": 73.1, "prec": 73.0, "recall": 76.1, "f1": 74.5},
-      "step": {"acc": 89.6, "prec": 44.5, "recall": 33.3, "f1": 38.1},
+      "traj": {
+        "acc": 73.1,
+        "prec": 73.0,
+        "recall": 76.1,
+        "f1": 74.5
+      },
+      "step": {
+        "acc": 89.6,
+        "prec": 44.5,
+        "recall": 33.3,
+        "f1": 38.1
+      },
       "step_note": null,
       "source": "https://huggingface.co/buckets/ZJUSCL/CUAStepBench_paperlogs"
     },
@@ -63,8 +103,18 @@
       "framework": "SeekJudge",
       "model": "Qwen3VL-8B",
       "model_note": null,
-      "traj": {"acc": 61.7, "prec": 58.7, "recall": 89.4, "f1": 70.8},
-      "step": {"acc": 89.4, "prec": 41.4, "recall": 20.4, "f1": 27.3},
+      "traj": {
+        "acc": 61.7,
+        "prec": 58.7,
+        "recall": 89.4,
+        "f1": 70.8
+      },
+      "step": {
+        "acc": 89.4,
+        "prec": 41.4,
+        "recall": 20.4,
+        "f1": 27.3
+      },
       "step_note": null,
       "source": "https://huggingface.co/buckets/ZJUSCL/CUAStepBench_paperlogs"
     },
@@ -74,8 +124,18 @@
       "framework": "CUAJudge",
       "model": "Qwen3VL-8B",
       "model_note": null,
-      "traj": {"acc": 67.9, "prec": 72.0, "recall": 62.5, "f1": 66.9},
-      "step": {"acc": 90.9, "prec": 67.7, "recall": 12.7, "f1": 21.3},
+      "traj": {
+        "acc": 67.9,
+        "prec": 72.0,
+        "recall": 62.5,
+        "f1": 66.9
+      },
+      "step": {
+        "acc": 90.9,
+        "prec": 67.7,
+        "recall": 12.7,
+        "f1": 21.3
+      },
       "step_note": "dagger",
       "source": "https://huggingface.co/buckets/ZJUSCL/CUAStepBench_paperlogs"
     },
@@ -85,8 +145,18 @@
       "framework": "OSThemis",
       "model": "Qwen3VL-8B",
       "model_note": null,
-      "traj": {"acc": 63.8, "prec": 71.1, "recall": 51.2, "f1": 59.5},
-      "step": {"acc": 83.6, "prec": 17.0, "recall": 14.3, "f1": 15.6},
+      "traj": {
+        "acc": 63.8,
+        "prec": 71.1,
+        "recall": 51.2,
+        "f1": 59.5
+      },
+      "step": {
+        "acc": 83.6,
+        "prec": 17.0,
+        "recall": 14.3,
+        "f1": 15.6
+      },
       "step_note": "dagger",
       "source": "https://huggingface.co/buckets/ZJUSCL/CUAStepBench_paperlogs"
     }