feat(task3): open the development and test phases

app/task3/page.tsx

@@ -13,6 +13,7 @@ export const metadata: Metadata = {
 
 const finmrDataset = "https://huggingface.co/datasets/TheFinAI/FinMR";
 const starterKit = "https://github.com/The-FinAI/IEEE-bigdata-cup/tree/main/finreason_task3";
+const hiddenDataset = "https://huggingface.co/datasets/YanAdjeNole/FinReason-Task3";
 
 type PhaseCard = {
   name: string;
@@ -31,6 +32,8 @@ export default function Task3HubPage() {
   const config = getTask3PublicConfig();
   const scoringIsLive =
     config.siteMode === "final" && config.scoringSpace.state === "ready";
+  const testIsLive =
+    config.siteMode === "final" && config.testSpace.state === "ready";
 
   const phases: PhaseCard[] = [
     {
@@ -48,16 +51,19 @@ export default function Task3HubPage() {
     {
       name: "Development",
       slug: "development",
-      live: false,
-      gold: "Held by the organizers",
+      // Development runs on the same scoring workspace as practice, so it is
+      // live under exactly the same condition. Hard-coding this would let a
+      // build with no Task 3 configuration announce an open phase.
+      live: scoringIsLive,
+      gold: "680 held-out cases, answers withheld",
       feedback: "Score, rank, and a public leaderboard",
       ranked: "Ranked",
     },
     {
       name: "Test",
       slug: "test",
-      live: false,
-      gold: "Held by the organizers",
+      live: testIsLive,
+      gold: "680 held-out cases, answers withheld",
       feedback: "An acceptance receipt only",
       ranked: "Decides the final result",
     },
@@ -145,7 +151,7 @@ export default function Task3HubPage() {
         <div className="task-platform-copy">
           <div>
             <p className="section-index">DATA</p>
-            <h2 id="task3-data-title">Where the practice data comes from</h2>
+            <h2 id="task3-data-title">Where the data comes from</h2>
           </div>
           <p>
             This site hosts no Task 3 files. The 332 public practice cases are the Financial
@@ -154,6 +160,13 @@ export default function Task3HubPage() {
             <a href={starterKit}>Task 3 starter kit</a> downloads them for you, and ships the
             validator, the scorer, and one baseline per way of running a model.
           </p>
+          <p>
+            The development and test questions are published separately, without their
+            answers, as{" "}
+            <a href={hiddenDataset}>YanAdjeNole/FinReason-Task3</a> — 680 cases each. They
+            carry no rule label: for one of the three rules the label alone gives away the
+            answer&rsquo;s shape.
+          </p>
         </div>
         <p className="task-platform-footnote">
           Every one of the 332 practice cases is built around one flagged Data Quality

app/task3/submission-guide.tsx

@@ -1,5 +1,6 @@
 const basePath = process.env.NEXT_PUBLIC_BASE_PATH ?? "";
 const starterKit = "https://github.com/The-FinAI/IEEE-bigdata-cup/tree/main/finreason_task3";
+const hiddenDataset = "https://huggingface.co/datasets/YanAdjeNole/FinReason-Task3";
 
 type SubmissionGuideProps = {
   scoringSpaceUrl: string | null;
@@ -51,12 +52,18 @@ export function SubmissionGuide({
         <li>
           <span className="submission-step-number" aria-hidden="true">01</span>
           <div>
-            <h3><span className="sr-only">Step 1 of 5: </span>Get the practice data.</h3>
+            <h3><span className="sr-only">Step 1 of 5: </span>Get the data.</h3>
             <p>
               Clone the <a href={starterKit}>Task 3 starter kit</a> and run its prepare script.
               It downloads the 332 public cases and writes both the answered and unanswered
               copies.
             </p>
+            <p>
+              For the two ranked phases, download{" "}
+              <code>development_inputs.jsonl</code> and <code>test_inputs.jsonl</code> from{" "}
+              <a href={hiddenDataset}>YanAdjeNole/FinReason-Task3</a>. Each holds 680
+              questions with the answers withheld, and no rule label.
+            </p>
             <pre tabIndex={0} aria-label="Prepare the practice data"><code>{prepareCommand}</code></pre>
           </div>
         </li>

tests/rendered-html.test.mjs

@@ -279,10 +279,20 @@ test("publishes the Task 3 participant hub with honest phase status", async () =
     assert.notEqual(rowOf(slug), "", `${slug} row is missing`);
   }
 
-  // Development and test have no datasets, in any build.
-  for (const slug of ["development", "test"]) {
-    assert.match(rowOf(slug), /data-state="pending"/, `${slug} must be pending`);
-    assert.doesNotMatch(rowOf(slug), /data-state="ready"/, `${slug} must not claim to be live`);
+  // The hidden datasets exist now, so these rows are no longer pinned to
+  // pending. What must still hold is that no row claims to be open in a build
+  // that has no workspace to send people to -- the original defect. Each row
+  // tracks its own workspace: practice and development share the scoring one,
+  // test has its own.
+  const panelPending = /<dd>Link under verification<\/dd>/.test(hub);
+  for (const slug of ["practice", "development", "test"]) {
+    if (panelPending) {
+      assert.match(
+        rowOf(slug),
+        /data-state="pending"/,
+        `${slug} must be pending while the site has no verified Space`,
+      );
+    }
   }
 
   // Practice tracks the configuration, and this assertion works in BOTH site
@@ -295,6 +305,15 @@ test("publishes the Task 3 participant hub with honest phase status", async () =
     "the practice row disagrees with the quick-facts panel",
   );
 
+  // Development runs on the same workspace as practice, so the two rows must
+  // never disagree: one of them being open while the other is not would mean
+  // the page is reporting a workspace state that does not exist.
+  assert.equal(
+    /data-state="ready"/.test(rowOf("development")),
+    /data-state="ready"/.test(rowOf("practice")),
+    "development and practice share a workspace but report different states",
+  );
+
   // The page must never contradict itself: if the panel says the link is still
   // being verified, no phase row may simultaneously claim to be open. This holds
   // in both site modes, which is why it is the assertion that would have caught
@@ -443,6 +462,27 @@ test("a Task 3 phase the hub calls open has a working upload link", async () =>
   );
 });
 
+test("an open ranked phase says where to get its questions", async () => {
+  // "Development: Live" is an empty claim if the questions are nowhere to be
+  // found. The gold stays private, but the inputs must be reachable from both
+  // the hub and the guide before either page calls those phases open.
+  const [hub, submit] = await Promise.all([
+    text("out/task3/index.html"),
+    text("out/task3/submit/index.html"),
+  ]);
+  const devRow = hub.match(/<tr[^>]*data-phase="development"[\s\S]*?<\/tr>/)?.[0] ?? "";
+  const ranked = /data-state="ready"/.test(devRow);
+  const dataset = /huggingface\.co\/datasets\/[\w.-]+\/FinReason-Task3/;
+  if (!ranked) return;
+  for (const [name, page] of [["hub", hub], ["submit guide", submit]]) {
+    assert.match(
+      page,
+      dataset,
+      `${name} calls development open but never says where its questions are`,
+    );
+  }
+});
+
 test("the home page links to all three task hubs", async () => {
   // A participant site nobody can navigate to does not do its job. The Task 3
   // route existed in the sitemap but no page linked to it.