ls3-lab · radinhamidi · Jun 24, 2026 · Jun 24, 2026
diff --git a/reproducibility/site/scripts/build-data.ts b/reproducibility/site/scripts/build-data.ts
@@ -11,9 +11,9 @@
  *
  * Outputs (gitignored, written to src/data/):
  *   - overview.json    summary used on /
- *   - datasets.json    [{id, name, run_count, eval_metrics}]
- *   - methods.json     [{id, run_count}]
- *   - models.json      [{id, run_count}]
+ *   - datasets.json    [{id, name, run_count, eval_metrics, hf_url?}]
+ *   - methods.json     [{id, run_count, display, paper?, paper_url?}]
+ *   - models.json      [{id, run_count, display, provider, slug}]
  *   - retrievers.json  [{id, display_name, paradigm, run_count}]
  *   - matrix.json      flat matrix: rows = (method, model, retriever), values per dataset
  *   - runs.json        full run index, keyed by run_id
@@ -45,6 +45,17 @@ const RETRIEVER_REGISTRY_YAML = path.join(
   "reproducibility",
   "retriever_registry.yaml",
 );
+// Curated method→paper citations (single source of truth, shared with the
+// marketing site). Keyed by logical method id; variant ids (e.g. query2doc-cot)
+// fall back to their base id (query2doc).
+const METHOD_CITATIONS_JSON = path.join(
+  REPO_ROOT,
+  "web",
+  "site",
+  "src",
+  "data",
+  "methods.json",
+);
 
 const OUT_DIR = path.join(SITE_ROOT, "src", "data");
 const VIEWS_DIR = path.join(OUT_DIR, "views");
@@ -80,8 +91,46 @@ interface DatasetEntry {
   bm25_weights?: { k1: number; b: number };
   eval_metrics: string[];
   run_count: number;
+  hf_url?: string;
 }
 
+// Canonical HuggingFace dataset/subset pages, keyed by dataset id. BEIR repos
+// are first-party (BeIR org); BRIGHT links target the per-domain split in the
+// `examples` config of the single xlangai/BRIGHT repo; the MS MARCO / TREC-DL
+// links point to the cleanest public HF mirror of each passage task.
+const DATASET_HF_URLS: Record<string, string> = {
+  "msmarco-v1-passage.trecdl2019": "https://huggingface.co/datasets/whybe-choi/trec-dl-2019",
+  "msmarco-v1-passage.trecdl2020": "https://huggingface.co/datasets/whybe-choi/trec-dl-2020",
+  "msmarco-v1-passage.dlhard": "https://huggingface.co/datasets/irds/msmarco-passage_trec-dl-hard",
+  "beir-v1.0.0-trec-covid": "https://huggingface.co/datasets/BeIR/trec-covid",
+  "beir-v1.0.0-fiqa": "https://huggingface.co/datasets/BeIR/fiqa",
+  "beir-v1.0.0-trec-news": "https://huggingface.co/datasets/BeIR/trec-news-generated-queries",
+  "beir-v1.0.0-arguana": "https://huggingface.co/datasets/BeIR/arguana",
+  "beir-v1.0.0-dbpedia-entity": "https://huggingface.co/datasets/BeIR/dbpedia-entity",
+  "beir-v1.0.0-scifact": "https://huggingface.co/datasets/BeIR/scifact",
+  "bright-biology": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/biology",
+  "bright-earth-science": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/earth_science",
+  "bright-economics": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/economics",
+  "bright-psychology": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/psychology",
+  "bright-robotics": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/robotics",
+  "bright-stackoverflow": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/stackoverflow",
+  "bright-sustainable-living": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/sustainable_living",
+  "bright-pony": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/pony",
+  "bright-leetcode": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/leetcode",
+  "bright-aops": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/aops",
+  "bright-theoremqa-theorems": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/theoremqa_theorems",
+  "bright-theoremqa-questions": "https://huggingface.co/datasets/xlangai/BRIGHT/viewer/examples/theoremqa_questions",
+};
+
+// Polished display name + provider slug per model id. Provider drives the logo
+// icon shown next to the model in the UI; falls back to the org prefix.
+const MODEL_META: Record<string, { display: string; provider: string }> = {
+  "openai/gpt-4.1": { display: "GPT-4.1", provider: "openai" },
+  "openai/gpt-4.1-nano": { display: "GPT-4.1 nano", provider: "openai" },
+  "Qwen/Qwen2.5-72B-Instruct": { display: "Qwen2.5 72B Instruct", provider: "qwen" },
+  "Qwen/Qwen2.5-7B-Instruct": { display: "Qwen2.5 7B Instruct", provider: "qwen" },
+};
+
 interface RetrieverEntry {
   id: string;
   display_name: string;
@@ -173,6 +222,29 @@ function readRetrieverRegistry(): Record<string, { display_name: string; paradig
   return doc.retrievers ?? {};
 }
 
+function readMethodCitations(): Record<string, { paper: string; paper_url: string }> {
+  if (!fs.existsSync(METHOD_CITATIONS_JSON)) return {};
+  const list = JSON.parse(fs.readFileSync(METHOD_CITATIONS_JSON, "utf-8")) as Array<{
+    id: string;
+    paper?: string;
+    paper_url?: string;
+  }>;
+  const map: Record<string, { paper: string; paper_url: string }> = {};
+  for (const m of list) {
+    if (m.paper && m.paper_url) map[m.id] = { paper: m.paper, paper_url: m.paper_url };
+  }
+  return map;
+}
+
+// Resolve a (possibly variant) method id to its citation, falling back to the
+// base id before the first "-" (e.g. query2doc-cot → query2doc).
+function citationFor(
+  id: string,
+  cites: Record<string, { paper: string; paper_url: string }>,
+): { paper: string; paper_url: string } | null {
+  return cites[id] ?? cites[id.split("-")[0]] ?? null;
+}
+
 function* iterRunFiles(): Generator<string> {
   if (!fs.existsSync(RUNS_DIR)) return;
   const stack = [RUNS_DIR];
@@ -534,10 +606,19 @@ function encodePathSegment(s: string): string {
 // Strip provider prefix from a model id for display: "openai/gpt-4.1" → "gpt-4.1".
 // The canonical id stays in the data; this is purely cosmetic.
 function displayModel(s: string): string {
+  if (MODEL_META[s]) return MODEL_META[s].display;
   const i = s.indexOf("/");
   return i >= 0 ? s.slice(i + 1) : s;
 }
 
+function modelProvider(s: string): string {
+  if (MODEL_META[s]) return MODEL_META[s].provider;
+  const org = s.includes("/") ? s.slice(0, s.indexOf("/")).toLowerCase() : "";
+  if (org.includes("openai")) return "openai";
+  if (org.includes("qwen")) return "qwen";
+  return "";
+}
+
 // ---------- main ------------------------------------------------------------
 
 function main() {
@@ -548,6 +629,7 @@ function main() {
   const manifest = readManifest();
   const datasets = readDatasetRegistry();
   const retrieverReg = readRetrieverRegistry();
+  const methodCites = readMethodCitations();
   const runDetails = readRunDetails(retrieverReg);
 
   // Counts.
@@ -591,19 +673,29 @@ function main() {
   const datasetList: DatasetEntry[] = Object.values(datasets).map((d) => ({
     ...d,
     run_count: datasetCounts.get(d.id) ?? 0,
+    ...(DATASET_HF_URLS[d.id] ? { hf_url: DATASET_HF_URLS[d.id] } : {}),
   }));
   datasetList.sort((a, b) => a.id.localeCompare(b.id));
   writeJSON(path.join(OUT_DIR, "datasets.json"), datasetList);
 
   const methodList = Array.from(methodCounts.entries())
-    .map(([id, run_count]) => ({ id, run_count, display: methodDisplay.get(id) ?? id }))
+    .map(([id, run_count]) => {
+      const cite = citationFor(id, methodCites);
+      return {
+        id,
+        run_count,
+        display: methodDisplay.get(id) ?? id,
+        ...(cite ? { paper: cite.paper, paper_url: cite.paper_url } : {}),
+      };
+    })
     .sort((a, b) => a.id.localeCompare(b.id));
   writeJSON(path.join(OUT_DIR, "methods.json"), methodList);
 
   const modelList = Array.from(modelCounts.entries())
     .map(([id, run_count]) => ({
       id,
       display: displayModel(id),
+      provider: modelProvider(id),
       run_count,
       slug: encodePathSegment(id),
     }))

diff --git a/reproducibility/site/src/lib/providerIcon.ts b/reproducibility/site/src/lib/providerIcon.ts
@@ -0,0 +1,26 @@
+// Inline monochrome provider logos (currentColor), shared by Astro pages and the
+// home-page client script. Returns an empty string for unknown providers so
+// callers can render name-only without a placeholder.
+
+const ICONS: Record<string, string> = {
+  // OpenAI mark
+  openai:
+    '<path d="M22.2819 9.8211a5.9847 5.9847 0 0 0-.5157-4.9108 6.0462 6.0462 0 0 0-6.5098-2.9A6.0651 6.0651 0 0 0 4.9807 4.1818a5.9847 5.9847 0 0 0-3.9977 2.9 6.0462 6.0462 0 0 0 .7427 7.0966 5.98 5.98 0 0 0 .511 4.9107 6.051 6.051 0 0 0 6.5146 2.9001A5.9847 5.9847 0 0 0 13.2599 24a6.0557 6.0557 0 0 0 5.7718-4.2058 5.9894 5.9894 0 0 0 3.9977-2.9001 6.0557 6.0557 0 0 0-.7475-7.0729zm-9.022 12.6081a4.4755 4.4755 0 0 1-2.8764-1.0408l.1419-.0804 4.7783-2.7582a.7948.7948 0 0 0 .3927-.6813v-6.7369l2.0201 1.1686a.071.071 0 0 1 .038.052v5.5826a4.504 4.504 0 0 1-4.4945 4.4944zm-9.6607-4.1254a4.4708 4.4708 0 0 1-.5346-3.0137l.142.0852 4.783 2.7582a.7712.7712 0 0 0 .7806 0l5.8428-3.3685v2.3324a.0804.0804 0 0 1-.0332.0615L9.74 19.9502a4.4992 4.4992 0 0 1-6.1408-1.6464zM2.3408 7.8956a4.485 4.485 0 0 1 2.3655-1.9728V11.6a.7664.7664 0 0 0 .3879.6765l5.8144 3.3543-2.0201 1.1685a.0757.0757 0 0 1-.071 0l-4.8303-2.7865A4.504 4.504 0 0 1 2.3408 7.872zm16.5963 3.8558L13.1038 8.364 15.1192 7.2a.0757.0757 0 0 1 .071 0l4.8303 2.7913a4.4944 4.4944 0 0 1-.6765 8.1042v-5.6772a.79.79 0 0 0-.407-.667zm2.0107-3.0231l-.142-.0852-4.7735-2.7818a.7759.7759 0 0 0-.7854 0L9.409 9.2297V6.8974a.0662.0662 0 0 1 .0284-.0615l4.8303-2.7866a4.4992 4.4992 0 0 1 6.6802 4.66zM8.3065 12.863l-2.02-1.1638a.0804.0804 0 0 1-.038-.0567V6.0742a4.4992 4.4992 0 0 1 7.3757-3.4537l-.142.0805L8.704 5.459a.7948.7948 0 0 0-.3927.6813zm1.0976-2.3654l2.602-1.4998 2.6069 1.4998v2.9994l-2.5974 1.4997-2.6067-1.4997Z"/>',
+  // Qwen mark (Simple Icons)
+  qwen:
+    '<path d="M23.919 14.545 20.817 9.17l1.47-2.544a.56.56 0 0 0 0-.566l-1.633-2.83a.57.57 0 0 0-.49-.283h-6.207L12.487.402a.57.57 0 0 0-.49-.284H8.732a.56.56 0 0 0-.49.284L5.139 5.775h-2.94a.56.56 0 0 0-.49.284L.077 8.887a.56.56 0 0 0 0 .567L3.18 14.83l-1.47 2.545a.56.56 0 0 0 0 .566l1.634 2.83a.57.57 0 0 0 .49.283h6.205l1.47 2.545a.57.57 0 0 0 .49.284h3.266a.57.57 0 0 0 .49-.284l3.104-5.375h2.94a.57.57 0 0 0 .49-.283l1.634-2.828a.55.55 0 0 0-.004-.568M8.733.686l1.634 2.828-1.634 2.828H21.8L20.164 9.17H7.425L5.63 6.06Zm1.306 19.801-6.205-.002 1.634-2.83h3.265L2.201 6.344h3.267q3.182 5.517 6.367 11.032zm10.124-5.66L18.53 12l-6.532 11.315-1.634-2.83c2.129-3.673 4.25-7.351 6.373-11.028h3.592l3.102 5.374z"/>',
+};
+
+/** Inline SVG markup for a provider logo, sized to 1em and inheriting color. */
+export function providerIconSvg(provider: string | undefined | null): string {
+  const body = ICONS[(provider ?? "").toLowerCase()];
+  if (!body) return "";
+  return `<svg class="lb-model-icon" viewBox="0 0 24 24" fill="currentColor" aria-hidden="true" focusable="false">${body}</svg>`;
+}
+
+/** provider → inline SVG map, for passing into client scripts via define:vars. */
+export function providerIconMap(): Record<string, string> {
+  const out: Record<string, string> = {};
+  for (const k of Object.keys(ICONS)) out[k] = providerIconSvg(k);
+  return out;
+}
diff --git a/reproducibility/site/src/pages/datasets/[id].astro b/reproducibility/site/src/pages/datasets/[id].astro
@@ -99,6 +99,16 @@ const colCount = 4 + metricCols.length;
           <a href="/datasets/" class="text-xs text-qg-fg-muted hover:text-qg-fg">← Datasets</a>
           <h2 class="text-xl font-semibold text-qg-fg md:text-2xl">{title}</h2>
           <span class="qg-mono text-xs text-qg-fg-muted">{id}</span>
+          {datasetMeta?.hf_url && (
+            <a
+              class="text-xs font-medium text-qg-accent hover:underline"
+              href={datasetMeta.hf_url}
+              target="_blank"
+              rel="noopener noreferrer"
+            >
+              HuggingFace ↗
+            </a>
+          )}
           <div class="h-px flex-1 bg-qg-border"></div>
           <span class="text-[11px] uppercase tracking-wider text-qg-fg-muted">
             All results produced by

diff --git a/reproducibility/site/src/pages/datasets/index.astro b/reproducibility/site/src/pages/datasets/index.astro
@@ -23,6 +23,16 @@ import datasets from "../../data/datasets.json";
             </div>
             <div class="mt-1 qg-mono text-xs text-qg-fg-muted">{d.id}</div>
           </a>
+          {d.hf_url && (
+            <a
+              class="mt-2 inline-block text-xs font-medium text-qg-accent hover:underline"
+              href={d.hf_url}
+              target="_blank"
+              rel="noopener noreferrer"
+            >
+              HuggingFace ↗
+            </a>
+          )}
         </li>
       ))
     }

diff --git a/reproducibility/site/src/pages/index.astro b/reproducibility/site/src/pages/index.astro
@@ -6,7 +6,19 @@ import runs from "../data/runs.json";
 import retrievers from "../data/retrievers.json";
 import models from "../data/models.json";
 import methods from "../data/methods.json";
+import datasets from "../data/datasets.json";
 import { buildReproduceCmds, retrieveHint, evaluateHint, type RunLike } from "../lib/reproduce";
+import { providerIconMap } from "../lib/providerIcon";
+
+// id → HuggingFace dataset URL, for the link beside the dataset selector.
+const datasetHfMap: Record<string, string> = Object.fromEntries(
+  (datasets as any[]).filter((d) => d.hf_url).map((d) => [d.id, d.hf_url]),
+);
+// id → polished display + provider, for icons in the table's LLM column.
+const modelMetaMap: Record<string, { display: string; provider: string }> = Object.fromEntries(
+  (models as any[]).map((m) => [m.id, { display: m.display ?? m.id, provider: m.provider ?? "" }]),
+);
+const PROVIDER_ICONS = providerIconMap();
 
 const populated = overview.run_count > 0;
 
@@ -182,6 +194,7 @@ const jsonLd = JSON.stringify({
           <span class="lb-selector-label">Dataset</span>
           {/* populated entirely by JS on init */}
           <select id="lb-ds-select" class="lb-ds-select"></select>
+          <a id="lb-ds-hf" class="lb-hf-link lb-hidden" target="_blank" rel="noopener noreferrer">HuggingFace ↗</a>
         </div>
         <div class="lb-divider"></div>
         <div class="lb-ctrl-grp">
@@ -208,12 +221,25 @@ const jsonLd = JSON.stringify({
         <div class="lb-divider"></div>
         <div class="lb-ctrl-grp">
           <span class="lb-filter-label">Method</span>
-          <select id="lb-method-select" class="lb-ds-select">
-            <option value="">All methods</option>
-            {(methods as any[]).map((m) => (
-              <option value={m.id}>{m.display ?? m.display_name ?? m.id}</option>
-            ))}
-          </select>
+          {/* Custom dropdown: native <select> can't host the per-row paper link */}
+          <div class="lb-dropdown" id="lb-method-dd">
+            <button type="button" id="lb-method-toggle" class="lb-ds-select lb-dd-toggle" aria-haspopup="listbox" aria-expanded="false">
+              <span id="lb-method-label" class="lb-dd-toggle-label">All methods</span>
+            </button>
+            <div id="lb-method-menu" class="lb-dd-menu lb-hidden" role="listbox">
+              <div class="lb-dd-opt" role="option" data-value="">
+                <span class="lb-dd-opt-name">All methods</span>
+              </div>
+              {(methods as any[]).map((m) => (
+                <div class="lb-dd-opt" role="option" data-value={m.id}>
+                  <span class="lb-dd-opt-name">{m.display ?? m.display_name ?? m.id}</span>
+                  {m.paper_url && (
+                    <a class="lb-dd-paper" href={m.paper_url} target="_blank" rel="noopener noreferrer" data-paper>paper ↗</a>
+                  )}
+                </div>
+              ))}
+            </div>
+          </div>
         </div>
         <div class="lb-divider"></div>
         <div class="lb-search-wrap">
@@ -360,6 +386,9 @@ const jsonLd = JSON.stringify({
   METRIC_LABEL,
   methodsData:      methods,
   modelsData:       models,
+  datasetHfMap,
+  modelMetaMap,
+  PROVIDER_ICONS,
 }}>
   // ------ viewport-width tracker for expand panel (sticky positioning) -------
   const scrollEl = document.getElementById("lb-scroll");
@@ -422,6 +451,7 @@ const jsonLd = JSON.stringify({
     const ids = datasets.map((d) => d.id);
     state.dataset = ids.includes(state.dataset) ? state.dataset : (ids[0] ?? "");
     dsSelect.value = state.dataset;
+    updateDatasetHf(state.dataset);
   }
 
   document.querySelectorAll("#lb-bench-pills button").forEach((btn) => {
@@ -434,8 +464,23 @@ const jsonLd = JSON.stringify({
     });
   });
 
+  // HuggingFace link beside the dataset selector — updates with selection.
+  const dsHfLink = document.getElementById("lb-ds-hf");
+  function updateDatasetHf(id) {
+    if (!dsHfLink) return;
+    const url = (datasetHfMap ?? {})[id];
+    if (url) {
+      dsHfLink.href = url;
+      dsHfLink.classList.remove("lb-hidden");
+    } else {
+      dsHfLink.removeAttribute("href");
+      dsHfLink.classList.add("lb-hidden");
+    }
+  }
+
   dsSelect?.addEventListener("change", (e) => {
     state.dataset = e.target.value;
+    updateDatasetHf(state.dataset);
     renderTable();
   });
 
@@ -457,9 +502,41 @@ const jsonLd = JSON.stringify({
     filterState.model = e.target.value ?? "";
     renderTable();
   });
-  document.getElementById("lb-method-select")?.addEventListener("change", (e) => {
-    filterState.method = e.target.value ?? "";
-    renderTable();
+  // ------ method custom dropdown (per-row paper links) ------------------------
+  const methodDd     = document.getElementById("lb-method-dd");
+  const methodToggle = document.getElementById("lb-method-toggle");
+  const methodMenu   = document.getElementById("lb-method-menu");
+  const methodLabel  = document.getElementById("lb-method-label");
+
+  function closeMethodMenu() {
+    methodMenu?.classList.add("lb-hidden");
+    methodToggle?.setAttribute("aria-expanded", "false");
+  }
+  function openMethodMenu() {
+    methodMenu?.classList.remove("lb-hidden");
+    methodToggle?.setAttribute("aria-expanded", "true");
+  }
+  methodToggle?.addEventListener("click", (e) => {
+    e.stopPropagation();
+    if (methodMenu?.classList.contains("lb-hidden")) openMethodMenu();
+    else closeMethodMenu();
+  });
+  methodMenu?.querySelectorAll(".lb-dd-opt").forEach((opt) => {
+    opt.addEventListener("click", (e) => {
+      // Clicks on the paper link open the paper without changing selection.
+      if (e.target.closest("[data-paper]")) { e.stopPropagation(); return; }
+      filterState.method = opt.dataset.value ?? "";
+      if (methodLabel) {
+        methodLabel.textContent = opt.querySelector(".lb-dd-opt-name")?.textContent ?? "All methods";
+      }
+      methodMenu.querySelectorAll(".lb-dd-opt").forEach((o) => o.classList.remove("selected"));
+      opt.classList.add("selected");
+      closeMethodMenu();
+      renderTable();
+    });
+  });
+  document.addEventListener("click", (e) => {
+    if (methodDd && !methodDd.contains(e.target)) closeMethodMenu();
   });
 
   // ------ search --------------------------------------------------------------
@@ -606,6 +683,7 @@ const jsonLd = JSON.stringify({
 
       const mTagIdx  = methodColorMap[row.method_id] ?? 0;
       const mdTagIdx = modelColorMap[row.model] ?? 0;
+      const modelIcon = PROVIDER_ICONS[(modelMetaMap[row.model] ?? {}).provider] ?? "";
       const retrieverCell = showRetrieverCol
         ? `<td class="lb-col-retriever">${escHtml(row.retriever_display ?? row.retriever_id)}</td>`
         : "";
@@ -614,7 +692,7 @@ const jsonLd = JSON.stringify({
 <tr class="data" data-row-key="${escHtml(rowKey)}" data-method="${escHtml(row.method_id)}" data-model="${escHtml(row.model)}" data-retriever="${escHtml(row.retriever_id)}">
   <td class="lb-col-rank">${rankBadge(ranks[i])}</td>
   <td class="lb-col-method"><span class="lb-badge lb-mtag-${mTagIdx}">${escHtml(row.method_display ?? row.method_id)}</span></td>
-  <td class="lb-col-model"><span class="lb-badge lb-mono lb-mdtag-${mdTagIdx}">${escHtml(row.model_display ?? row.model)}</span></td>
+  <td class="lb-col-model"><span class="lb-badge lb-mono lb-mdtag-${mdTagIdx}">${modelIcon}${escHtml(row.model_display ?? row.model)}</span></td>
   ${retrieverCell}
   <td class="lb-col-metric${pBest ? " best" : ""}" data-primary-value="${pv ?? ""}">
     <span class="metric-val">${pvStr}</span>