Add Task 2 HERCULEAN training data hub and downloads

README.md

@@ -51,10 +51,12 @@ Select an asset pair and manage a zero-net-dollar position over time using
 point-in-time prices, news, and corporate filings. The task is designed to
 reward relative-value reasoning rather than unhedged directional exposure.
 
-The planned data are derived from
-[HERCULEAN](https://arxiv.org/abs/2605.14355). Exact market windows, eligible
-assets, execution assumptions, transaction costs, and position-validity rules
-will be published with the dataset and scorer.
+Training data is available from [HERCULEAN](https://huggingface.co/datasets/TheFinAI/Herculean/tree/main/data).
+Download `prices.parquet`, `news.parquet`, and `filings.parquet` from the
+[Task 2 training data hub](https://the-finai.github.io/IEEE-bigdata-cup/task2/).
+Development and private evaluation splits, exact market windows, eligible assets,
+execution assumptions, transaction costs, and position-validity rules will be
+published with the scorer.
 
 ### Task 3 · Financial Audit Verification
 
@@ -161,6 +163,7 @@ identifier.
 | [Task 1 step-by-step submission guide](https://the-finai.github.io/IEEE-bigdata-cup/task1/submit/#how-to-submit) | Live |
 | Task 1 validator, sample B0, and B1 baseline | Live in this repository |
 | Task 1 organizer baseline scores | Financial Rules and Organizer Baseline ranked on the 580-question development set |
+| [Task 2 training data](https://the-finai.github.io/IEEE-bigdata-cup/task2/) | Live: prices, news, and filings (Parquet) |
 | [Task 3 starter kit, validator, and scorer](finreason_task3/) | Live in this repository |
 | Task 3 practice phase (332 public FinMR cases) | Live — scored on upload, never ranked |
 | Task 3 development and test datasets | Coming soon |

app/page.tsx

@@ -35,7 +35,7 @@ const tasks = [
       "Select an asset pair and manage a zero-net-dollar position over time using point-in-time prices, news, and corporate filings. Final execution and position rules will be published with the scorer.",
     flow: ["Market context", "Paired actions", "Risk-aware return"],
     data:
-      "Planned HERCULEAN-derived development and private evaluation splits. Exact windows, asset policy, costs, and validity rules will be published with the dataset release.",
+      "Training data is available from HERCULEAN: prices, news, and corporate filings in Parquet format. Development and private evaluation splits, exact windows, asset policy, costs, and validity rules will be published with the scorer.",
     metrics: [
       "Cumulative return · CR",
       "Sharpe ratio · SR",
@@ -231,6 +231,7 @@ export default function Home() {
           </a>
           <nav className="primary-nav" aria-label="Primary navigation">
             <a href={`${basePath}/task1/`}>Task 1</a>
+            <a href={`${basePath}/task2/`}>Task 2</a>
             <a href={`${basePath}/task3/`}>Task 3</a>
             <a href="#tracks">Tracks</a>
             <a href="#evaluation">Evaluation</a>
@@ -447,6 +448,11 @@ export default function Home() {
                 <div>
                   <span className="meta-label">Data status</span>
                   <p>{task.data}</p>
+                  {task.number === "02" && (
+                    <a className="button button-primary" href={`${basePath}/task2/`}>
+                      Download Task 2 training data
+                    </a>
+                  )}
                 </div>
                 <div>
                   <span className="meta-label">Score signals</span>
@@ -523,7 +529,8 @@ export default function Home() {
               All dates below are in 2026. Competition submissions and Working Notes have
               separate deadlines. AoE means Anywhere on Earth (UTC−12); the 15 October
               competition cutoff passes at 12:00 UTC on 16 October. Task availability
-              and release notices remain published in the participant hubs.
+              and release notices remain published in the participant hubs. Task 2 training
+              data is available now.
             </p>
           </div>
         </div>

app/sitemap.ts

@@ -30,6 +30,12 @@ export default function sitemap(): MetadataRoute.Sitemap {
       changeFrequency: "daily",
       priority: 0.7,
     },
+    {
+      url: `${siteUrl}task2/`,
+      lastModified: new Date("2026-09-05"),
+      changeFrequency: "weekly",
+      priority: 0.9,
+    },
     {
       url: `${siteUrl}task1/leaderboard/test/`,
       lastModified: new Date("2026-09-09"),

app/task2/page.tsx

@@ -0,0 +1,84 @@
+import type { Metadata } from "next";
+import Link from "next/link";
+
+const datasetUrl = "https://huggingface.co/datasets/TheFinAI/Herculean";
+const trainingFiles = ["prices.parquet", "news.parquet", "filings.parquet"];
+
+export const metadata: Metadata = {
+  title: "Task 2 Training Data | FinReason Cup",
+  description: "HERCULEAN training data for Task 2 Market-Neutral Hedging: prices, news, and corporate filings.",
+  alternates: { canonical: "https://the-finai.github.io/IEEE-bigdata-cup/task2/" },
+};
+
+export default function Task2HubPage() {
+  return (
+    <main className="task-hub-page task1-page">
+      <nav className="task-hub-nav" aria-label="Task 2 navigation">
+        <Link className="task-hub-brand" href="/">
+          <span className="task-hub-brand-mark" aria-hidden="true">FR</span>
+          <span><strong>FinReason Cup</strong><small>IEEE Big Data 2026</small></span>
+        </Link>
+        <div className="task-hub-nav-links">
+          <Link href="/task1/">Task 1</Link>
+          <Link href="/task2/" aria-current="page">Task 2 data</Link>
+        </div>
+      </nav>
+
+      <header className="task-hub-heading">
+        <div className="task-hub-heading-content">
+          <div>
+            <p className="section-index">TASK 2 / PARTICIPANT HUB</p>
+            <h1>Market-Neutral Hedging training data</h1>
+            <p>
+              Training data for Task 2 is available from HERCULEAN on Hugging Face.
+              Download prices, news, and corporate filings below to prepare your system.
+            </p>
+          </div>
+          <dl className="task-hub-facts" aria-label="Task 2 quick facts">
+            <div><dt>Status</dt><dd>Training data available</dd></div>
+            <div><dt>Format</dt><dd>3 Parquet files</dd></div>
+            <div><dt>Submission</dt><dd>Publication pending</dd></div>
+          </dl>
+        </div>
+      </header>
+
+      <section className="task-platform-card" aria-labelledby="task2-downloads-title">
+        <div className="task-platform-copy">
+          <div>
+            <p className="section-index">DOWNLOADS</p>
+            <h2 id="task2-downloads-title">Task 2 files</h2>
+          </div>
+          <p>
+            These links download the training files directly from the
+            HERCULEAN dataset. You can also browse the full{" "}
+            <a href={`${datasetUrl}/tree/main/data`}>data directory on Hugging Face</a>.
+          </p>
+        </div>
+        <div className="task-mode-grid">
+          <article>
+            <span>TRAINING / 3 FILES</span>
+            <ul className="policy-list">
+              {trainingFiles.map((file) => (
+                <li key={file}>
+                  <a href={`${datasetUrl}/resolve/main/data/${file}?download=true`}>{file}</a>
+                </li>
+              ))}
+            </ul>
+          </article>
+        </div>
+        <p className="task-platform-footnote">
+          Data source and license: <a href={datasetUrl}>HERCULEAN dataset card</a> (CC BY 4.0).
+        </p>
+      </section>
+
+      <aside className="task-hub-note">
+        <strong>Evaluation and submission status</strong>
+        <p>
+          Development and private evaluation splits, exact market windows, eligible assets,
+          execution assumptions, transaction costs, position-validity rules, and the scorer
+          will be published separately. The Task 2 submission route is not yet open.
+        </p>
+      </aside>
+    </main>
+  );
+}