By-section 'All edit types' cell click: one combined summary of all edits, not per-label

api.py

@@ -558,6 +558,12 @@ async def download_demo_csv(article: str) -> FileResponse:
 class SummarizePayload(BaseModel):
     article: str
     label: int
+    # Combined multi-label summary: when set (length > 1), rows are matched if
+    # they carry ANY of these labels, and every covered label's explanation
+    # snippet is summarized together into a single write-up. Sent by the
+    # by-section heatmap's "All scientific edit types" cell clicks. `label`
+    # still holds labels[0] for backward-compatible single-label callers.
+    labels: Optional[List[int]] = None
     period: str  # one of "Y" | "2Q" | "Q" — matches the frontend Period type
     # e.g. "2018", "2018Q3", "2018H1". Omit / send null to summarize across
     # every period (the "Summarize all periods" affordance in Over-Time).
@@ -784,6 +790,16 @@ async def summarize_edits(payload: SummarizePayload) -> Dict[str, Any]:
     section-group) cell using a fast OpenAI model. Used by interactive chart
     drill-downs.
     """
+    # Normalize the label set. A combined request (by-section "All edit types"
+    # cell) carries multiple labels; a single-label request carries just one.
+    # `combined_labels` is the deduped, sorted set we match/summarize over;
+    # `is_combined` gates the multi-label prompt and response labelling.
+    if payload.labels:
+        combined_labels = sorted({int(l) for l in payload.labels})
+    else:
+        combined_labels = [int(payload.label)]
+    is_combined = len(combined_labels) > 1
+
     # Client-supplied path: the dataset is in the user's browser, not cached
     # on the server. The frontend has already filtered rows by (label, period,
     # sections) and sent the relevant explanation snippets. Skip the server
@@ -824,11 +840,19 @@ async def summarize_edits(payload: SummarizePayload) -> Dict[str, Any]:
     if client_snippets is not None:
         snippets = client_snippets
         # We don't know the unmatched count on the client path; use the count
-        # the client reported.
+        # the client reported. On this path each reported snippet is one
+        # non-empty explanation, so the "with explanation text" count is just
+        # the snippet count.
         match_count_for_response = matched_count or len(snippets)
+        rows_with_expl = len(snippets)
     else:
-        # Filter to rows that contain the requested label.
-        mask = df["Labels"].apply(lambda v: payload.label in _parse_labels_field(v))
+        # Filter to rows that contain any of the requested label(s). For a
+        # single-label request this is just `payload.label`; for a combined
+        # request it's any of the covered labels.
+        label_set = set(combined_labels)
+        mask = df["Labels"].apply(
+            lambda v: bool(label_set.intersection(_parse_labels_field(v)))
+        )
 
         # Never summarize hidden structural sections, even if the union-find
         # grouped_idx lumped one in with a visible section the user selected.
@@ -863,12 +887,21 @@ async def summarize_edits(payload: SummarizePayload) -> Dict[str, Any]:
                 ),
             }
 
-        # Pull the relevant Explanation snippet for each matched row.
+        # Pull the relevant Explanation snippet(s) for each matched row. For a
+        # combined request, gather every covered label's snippet the row
+        # carries so the summary spans all edit types in the cell.
         snippets = []
+        rows_with_expl = 0
         for _, row in matched.iterrows():
-            text = _explanation_for_label(row.get("Explanation"), payload.label)
-            if text:
-                snippets.append(text)
+            raw_expl = row.get("Explanation")
+            row_had_snippet = False
+            for lbl in combined_labels:
+                text = _explanation_for_label(raw_expl, lbl)
+                if text:
+                    snippets.append(text)
+                    row_had_snippet = True
+            if row_had_snippet:
+                rows_with_expl += 1
 
         if not snippets:
             return {
@@ -913,46 +946,100 @@ async def summarize_edits(payload: SummarizePayload) -> Dict[str, Any]:
         if payload.period_key
         else "across the article's full revision history"
     )
-    user_prompt = (
-        f"You are summarizing Wikipedia revision edits for the article "
-        f"'{payload.article}'.\n\n"
-        f"Below are explanations from {len(snippets_for_prompt)} edits "
-        f"classified as '{label_name}' (label {payload.label}) {period_note}, "
-        f"{section_note}.\n\n"
-        f"Definition of '{label_name}':\n{label_definition}\n\n"
-        f"Write a tight, scannable summary as 3–6 short bullets that "
-        f"synthesize the patterns. Each bullet must:\n"
-        f"- Start with '- '.\n"
-        f"- Be one sentence, ideally under 20 words.\n"
-        f"- Cover a distinct theme or grouping — do NOT restate the same "
-        f"point twice and do NOT enumerate every edit.\n"
-        f"- Lead with the substance, not filler (avoid openers like "
-        f"'This edit…' or 'The article…').\n\n"
-        f"Scope rules — read carefully:\n"
-        f"- Only discuss evidence that directly satisfies the definition of "
-        f"'{label_name}' given above. Do NOT mention edits or details that "
-        f"appeared in the explanations but fall outside this definition "
-        f"(e.g. for an 'Academic References Added' summary, do not mention "
-        f"named-reference tags, formatting cleanup, or non-academic links — "
-        f"the definition is specifically about new academic papers with "
-        f"identifiers like DOI/PMID/PMC/arXiv/ISSN).\n"
-        f"- If something feels noteworthy but doesn't match the definition, "
-        f"omit it. It will be covered by a different label's summary.\n\n"
-        f"Formatting requirements:\n"
-        f"- Use Markdown **bold** (double asterisks) only around concrete "
-        f"evidence that directly satisfies the definition above — e.g. the "
-        f"specific technical terms, references, wikilinks, numeric values, "
-        f"or researcher names mentioned in the explanations. Bold only the "
-        f"evidence phrase itself, not the surrounding prose. Do not bold "
-        f"anything that isn't itself an instance of the label.\n"
-        f"- After the bullets, add a separate final line beginning with "
-        f"'Bottom line: ' that gives a one-sentence takeaway about what "
-        f"these edits collectively did to the article during this period.\n"
-        f"- Do not output any prose outside the bullets and the Bottom line "
-        f"line. No headings, no intro sentence, no closing remark.\n\n"
-        f"Explanations:\n"
-        + "\n\n".join(f"- {s}" for s in snippets_for_prompt)
-    )
+
+    if is_combined:
+        # Combined summary: one write-up spanning every scientific edit type
+        # present in the cell. The scope is the union of the covered labels'
+        # definitions rather than a single one, so the model synthesizes
+        # across edit types instead of being told to omit "other labels".
+        covered_names = [
+            label_name_map.get(l, f"Label {l}") for l in combined_labels
+        ]
+        definitions_block = "\n\n".join(
+            f"'{label_name_map.get(l, f'Label {l}')}' (label {l}):\n"
+            f"{_label_definition(l, payload.article)}"
+            for l in combined_labels
+        )
+        response_label_name = "All scientific edit types"
+        user_prompt = (
+            f"You are summarizing Wikipedia revision edits for the article "
+            f"'{payload.article}'.\n\n"
+            f"Below are explanations from {len(snippets_for_prompt)} edit "
+            f"observations {period_note}, {section_note}. They span every "
+            f"scientific edit type present in this selection: "
+            f"{', '.join(covered_names)}.\n\n"
+            f"Definitions of the covered edit types:\n{definitions_block}\n\n"
+            f"Write a tight, scannable summary as 3–6 short bullets that "
+            f"synthesize what changed across ALL of these edit types together. "
+            f"Each bullet must:\n"
+            f"- Start with '- '.\n"
+            f"- Be one sentence, ideally under 20 words.\n"
+            f"- Cover a distinct theme or grouping — do NOT restate the same "
+            f"point twice and do NOT enumerate every edit.\n"
+            f"- Lead with the substance, not filler (avoid openers like "
+            f"'This edit…' or 'The article…').\n\n"
+            f"Scope rules — read carefully:\n"
+            f"- Summarize the edits as a whole. Group related changes across "
+            f"edit types; you do not need one bullet per edit type.\n"
+            f"- Only discuss evidence that satisfies at least one of the "
+            f"definitions above. Ignore anything in the explanations that "
+            f"falls outside all of them.\n\n"
+            f"Formatting requirements:\n"
+            f"- Use Markdown **bold** (double asterisks) only around concrete "
+            f"evidence — e.g. the specific technical terms, references, "
+            f"wikilinks, numeric values, or researcher names mentioned in the "
+            f"explanations. Bold only the evidence phrase itself, not the "
+            f"surrounding prose.\n"
+            f"- After the bullets, add a separate final line beginning with "
+            f"'Bottom line: ' that gives a one-sentence takeaway about what "
+            f"these edits collectively did to the article during this period.\n"
+            f"- Do not output any prose outside the bullets and the Bottom "
+            f"line line. No headings, no intro sentence, no closing remark.\n\n"
+            f"Explanations:\n"
+            + "\n\n".join(f"- {s}" for s in snippets_for_prompt)
+        )
+    else:
+        response_label_name = label_name
+        user_prompt = (
+            f"You are summarizing Wikipedia revision edits for the article "
+            f"'{payload.article}'.\n\n"
+            f"Below are explanations from {len(snippets_for_prompt)} edits "
+            f"classified as '{label_name}' (label {payload.label}) {period_note}, "
+            f"{section_note}.\n\n"
+            f"Definition of '{label_name}':\n{label_definition}\n\n"
+            f"Write a tight, scannable summary as 3–6 short bullets that "
+            f"synthesize the patterns. Each bullet must:\n"
+            f"- Start with '- '.\n"
+            f"- Be one sentence, ideally under 20 words.\n"
+            f"- Cover a distinct theme or grouping — do NOT restate the same "
+            f"point twice and do NOT enumerate every edit.\n"
+            f"- Lead with the substance, not filler (avoid openers like "
+            f"'This edit…' or 'The article…').\n\n"
+            f"Scope rules — read carefully:\n"
+            f"- Only discuss evidence that directly satisfies the definition of "
+            f"'{label_name}' given above. Do NOT mention edits or details that "
+            f"appeared in the explanations but fall outside this definition "
+            f"(e.g. for an 'Academic References Added' summary, do not mention "
+            f"named-reference tags, formatting cleanup, or non-academic links — "
+            f"the definition is specifically about new academic papers with "
+            f"identifiers like DOI/PMID/PMC/arXiv/ISSN).\n"
+            f"- If something feels noteworthy but doesn't match the definition, "
+            f"omit it. It will be covered by a different label's summary.\n\n"
+            f"Formatting requirements:\n"
+            f"- Use Markdown **bold** (double asterisks) only around concrete "
+            f"evidence that directly satisfies the definition above — e.g. the "
+            f"specific technical terms, references, wikilinks, numeric values, "
+            f"or researcher names mentioned in the explanations. Bold only the "
+            f"evidence phrase itself, not the surrounding prose. Do not bold "
+            f"anything that isn't itself an instance of the label.\n"
+            f"- After the bullets, add a separate final line beginning with "
+            f"'Bottom line: ' that gives a one-sentence takeaway about what "
+            f"these edits collectively did to the article during this period.\n"
+            f"- Do not output any prose outside the bullets and the Bottom line "
+            f"line. No headings, no intro sentence, no closing remark.\n\n"
+            f"Explanations:\n"
+            + "\n\n".join(f"- {s}" for s in snippets_for_prompt)
+        )
 
     try:
         import openai
@@ -978,10 +1065,14 @@ async def summarize_edits(payload: SummarizePayload) -> Dict[str, Any]:
     return {
         "summary": summary_text.strip(),
         "edit_count": match_count_for_response,
-        "explanation_count": len(snippets),
+        # Rows that contributed at least one explanation snippet. In combined
+        # mode a single row may contribute several snippets, so this stays a
+        # per-row count rather than the raw snippet total.
+        "explanation_count": rows_with_expl,
         "truncated": truncated,
         "label": payload.label,
-        "label_name": label_name,
+        "labels": combined_labels if is_combined else None,
+        "label_name": response_label_name,
         "period_key": payload.period_key,
     }
 

frontend_build/asset-manifest.json

@@ -1,7 +1,7 @@
 {
   "files": {
     "main.css": "/static/css/main.ccff2033.css",
-    "main.js": "/static/js/main.caab1ec2.js",
+    "main.js": "/static/js/main.8c4ac153.js",
     "static/js/206.b8423ec3.chunk.js": "/static/js/206.b8423ec3.chunk.js",
     "static/media/KaTeX_AMS-Regular.ttf": "/static/media/KaTeX_AMS-Regular.853be92419a6c3766b9a.ttf",
     "static/media/KaTeX_Main-Regular.ttf": "/static/media/KaTeX_Main-Regular.9eba1d77abcf2aa6e94e.ttf",
@@ -67,6 +67,6 @@
   },
   "entrypoints": [
     "static/css/main.ccff2033.css",
-    "static/js/main.caab1ec2.js"
+    "static/js/main.8c4ac153.js"
   ]
 }

frontend_build/index.html

@@ -1 +1 @@
-<!doctype html><html lang="en"><head><meta charset="utf-8"/><link rel="icon" href="/favicon.ico"/><meta name="viewport" content="width=device-width,initial-scale=1"/><meta name="theme-color" content="#0e7490"/><meta name="description" content="Wikistar — understand how scientific revisions evolve over time for a specific Wikipedia page."/><link rel="apple-touch-icon" href="/wikistar-192.png"/><link rel="manifest" href="/manifest.json"/><title>Wikistar — Wikipedia Edit History Explorer</title><script defer="defer" src="/static/js/main.caab1ec2.js"></script><link href="/static/css/main.ccff2033.css" rel="stylesheet"></head><body><noscript>You need to enable JavaScript to run this app.</noscript><div id="root"></div></body></html>
+<!doctype html><html lang="en"><head><meta charset="utf-8"/><link rel="icon" href="/favicon.ico"/><meta name="viewport" content="width=device-width,initial-scale=1"/><meta name="theme-color" content="#0e7490"/><meta name="description" content="Wikistar — understand how scientific revisions evolve over time for a specific Wikipedia page."/><link rel="apple-touch-icon" href="/wikistar-192.png"/><link rel="manifest" href="/manifest.json"/><title>Wikistar — Wikipedia Edit History Explorer</title><script defer="defer" src="/static/js/main.8c4ac153.js"></script><link href="/static/css/main.ccff2033.css" rel="stylesheet"></head><body><noscript>You need to enable JavaScript to run this app.</noscript><div id="root"></div></body></html>