Clarify downstream-impact view in demo

hf_space/streamlit_app.py

@@ -55,10 +55,10 @@ CUSTOM_CSS = """
 .field {font-size: 0.88rem; line-height: 1.5; color: #374151; margin-top: 0.4rem;}
 .field b {color: #111827;}
 .grounding-block {margin-top: 0.75rem; display: grid; gap: 0.55rem;}
-.grounding-card {border-radius: 10px; padding: 0.65rem 0.75rem; border: 1px solid #bfdbfe; background: #eff6ff;}
+.grounding-card {border-radius: 10px; padding: 0.65rem 0.75rem; border: 1px solid #fed7aa; background: #fff7ed;}
 .grounding-card.additional {border-color: #fed7aa; background: #fff7ed;}
 .grounding-label {font-size: 0.7rem; font-weight: 900; text-transform: uppercase; letter-spacing: 0.05em; margin-bottom: 0.25rem;}
-.grounding-label.primary {color: #1d4ed8;}
+.grounding-label.primary {color: #c2410c;}
 .grounding-label.additional {color: #c2410c;}
 .grounding-title {font-size: 0.9rem; font-weight: 800; color: #111827; line-height: 1.35;}
 .grounding-meta {font-size: 0.78rem; color: #64748b; margin-top: 0.2rem;}
@@ -253,10 +253,9 @@ def _grounding_html(grounding: Optional[dict], label: str, kind: str) -> str:
     if isinstance(authors, list) and authors:
         meta.append(", ".join(str(author) for author in authors[:3]))
     meta_html = f"<div class='grounding-meta'>{_esc(' · '.join(meta))}</div>" if meta else ""
-    extra_class = " additional" if kind == "additional" else ""
     return (
-        f"<div class='grounding-card{extra_class}'>"
-        f"<div class='grounding-label {kind}'>{_esc(label)}</div>"
+        "<div class='grounding-card'>"
+        f"<div class='grounding-label additional'>{_esc(label)}</div>"
         f"<div class='grounding-title'>{_esc(title)}</div>"
         f"{meta_html}"
         "</div>"
@@ -278,8 +277,8 @@ def _collect_grounded_studies(discoveries: list[dict], ingredients: list[dict])
         if not isinstance(item, dict):
             continue
         copied = dict(item)
-        copied["_grounding_kind"] = "primary"
-        copied["_grounding_label"] = "Primary study"
+        copied["_grounding_kind"] = "grounding"
+        copied["_grounding_label"] = "Grounding"
         key = _study_key(copied)
         if key:
             seen.add(key)
@@ -298,8 +297,8 @@ def _collect_grounded_studies(discoveries: list[dict], ingredients: list[dict])
             if key and (key == canonical_key or key in seen):
                 continue
             copied = dict(ref)
-            copied["_grounding_kind"] = "additional"
-            copied["_grounding_label"] = f"Additional study for enabling contribution {idx}"
+            copied["_grounding_kind"] = "grounding"
+            copied["_grounding_label"] = f"Grounding for enabling contribution {idx}"
             copied.setdefault("role", annotation.get("role") or ", ".join(annotation.get("roles") or []))
             copied.setdefault("contribution", annotation.get("contribution"))
             copied.setdefault("rationale", annotation.get("rationale"))
@@ -316,7 +315,6 @@ def _render_reference_list(discoveries: list[dict], ingredients: Optional[list[d
         return
     for item in studies:
         title = item.get("ref_title") or item.get("title") or item.get("ref_id") or item.get("paper_id") or "Untitled reference"
-        is_additional = item.get("_grounding_kind") == "additional"
         meta = []
         if item.get("_grounding_label"):
             meta.append(str(item.get("_grounding_label")))
@@ -324,7 +322,7 @@ def _render_reference_list(discoveries: list[dict], ingredients: Optional[list[d
             meta.append(str(item.get("role")))
         if item.get("ref_year"):
             meta.append(str(item.get("ref_year")))
-        class_name = "cluster-card additional-study" if is_additional else "cluster-card"
+        class_name = "cluster-card additional-study"
         body = [f"<div class='{class_name}'><div class='cluster-title'>{_esc(title)}</div>"]
         if meta:
             body.append(f"<div class='cluster-meta'>{_esc(' · '.join(meta))}</div>")
@@ -345,6 +343,11 @@ def _render_claims_tab(payload: Optional[dict]):
         st.markdown("<div class='empty-card'>The run completed, but no target contributions were produced.</div>", unsafe_allow_html=True)
         return
 
+    st.markdown(
+        "<div class='soft-card'>This decomposition is done in hindsight: SciPaths uses the observed downstream citation clusters to identify reusable target contributions, then reconstructs the enabling contributions and groundings that made each target contribution possible.</div>",
+        unsafe_allow_html=True,
+    )
+
     for idx, claim in enumerate(claims, start=1):
         claim_id = claim.get("claim_id") or f"C{idx}"
         claim_text = claim.get("rewritten_claim") or claim.get("text") or "(missing target contribution text)"
@@ -385,7 +388,7 @@ def _render_claims_tab(payload: Optional[dict]):
                 grounding_parts = []
                 if canonical_grounding:
                     grounding_parts.append(
-                        _grounding_html(canonical_grounding, "Primary grounding", "primary")
+                        _grounding_html(canonical_grounding, "Grounding", "grounding")
                     )
                 for ref in extras:
                     if not isinstance(ref, dict):
@@ -396,13 +399,13 @@ def _render_claims_tab(payload: Optional[dict]):
                     ):
                         continue
                     grounding_parts.append(
-                        _grounding_html(ref, "Additional grounding", "additional")
+                        _grounding_html(ref, "Grounding", "grounding")
                     )
                 if not grounding_parts:
                     canonical_ref_id = ingredient.get("canonical_ref_id") or "__NONE__"
                     grounding_parts.append(
                         "<div class='grounding-card'>"
-                        "<div class='grounding-label primary'>Grounding</div>"
+                        "<div class='grounding-label additional'>Grounding</div>"
                         f"<div class='grounding-title'>{_esc(canonical_ref_id)}</div>"
                         "</div>"
                     )
@@ -428,7 +431,7 @@ def _render_claims_tab(payload: Optional[dict]):
                     unsafe_allow_html=True,
                 )
         with right:
-            st.markdown("<div class='section-label'>Grounded and additional studies</div>", unsafe_allow_html=True)
+            st.markdown("<div class='section-label'>Groundings</div>", unsafe_allow_html=True)
             _render_reference_list(discoveries, ingredients)
 
 
@@ -436,6 +439,10 @@ def _render_clusters_tab(discovery: Optional[dict], contributions: list[dict]):
     if not discovery:
         st.markdown("<div class='empty-card'>No refined cluster file is available yet.</div>", unsafe_allow_html=True)
         return
+    st.markdown(
+        "<div class='soft-card'>Citation clusters are forward-looking: they summarize how later papers use or extend the input paper, giving a downstream-impact view of what became reusable.</div>",
+        unsafe_allow_html=True,
+    )
     clusters = discovery.get("clusters") or []
     dropped = discovery.get("dropped_clusters") or []
     if not clusters:
@@ -796,7 +803,7 @@ def main():
 
     st.markdown("<div class='hero-title'>Forecasting Scientific Contribution Pathways</div>", unsafe_allow_html=True)
     st.markdown(
-        "<div class='hero-sub'>Run the SciPaths pipeline through refined downstream citation clusters, then derive target contributions from those clusters and decompose each target contribution into enabling contributions and grounded studies.</div>",
+        "<div class='hero-sub'>SciPaths first looks forward, grouping downstream papers by how they use or extend the input paper. It then performs the target contribution decomposition in hindsight: given those observed downstream-impact clusters, it identifies reusable target contributions and decomposes them into enabling contributions and groundings.</div>",
         unsafe_allow_html=True,
     )
 

hf_space/streamlit_config.py

@@ -1,7 +1,7 @@
 from runner import STEP_LABELS
 
 EXAMPLES = {
-    "Confidence-based MBR Decoding": "https://arxiv.org/abs/2311.14919",
+    "PASTA: Participant States in Narratives": "https://arxiv.org/abs/2208.00329",
     "AVerImaTeC": "https://arxiv.org/abs/2505.17978",
     "CSCD-NS (2022)": "https://arxiv.org/abs/2211.08788",
 }