Option A: typographic editorial tiles + single treated hero image; balance brief topics

Frontend (the premium baseline):
- The hero is now the ONLY image slot. Soft feed images get an atmospheric
  gradient overlay; no over-reliance on inconsistent RSS image quality.
- Every secondary/lane card is a uniform typographic editorial tile: no
  thumbnails, equal visual weight, a faint topic wordmark watermark, a slim
  sage top accent, consistent source, reason text as the trust signal, visible
  Replace with quiet tuning actions. Fixes the jarring mixed-media row rhythm
  and removes muddy thumbnails entirely.

Backend (composition):
- _select_diverse now balances topics: no more than 2 of one topic while other
  topics have candidates (relaxing source then topic caps only to fill), so the
  daily five stop clustering medical/science items. Candidates now carry s.topic.

Tests updated for the topic-balance contract (79 total).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

goodnews/briefs.py

@@ -118,6 +118,8 @@ def _candidate_articles(
             s.reason_code,
             s.reason_text,
             s.model_name,
+            s.topic,
+            s.flavor,
             CASE WHEN date(COALESCE(a.published_at, a.discovered_at)) = date(?)
                  THEN 1 ELSE 0 END AS is_today
         FROM articles a
@@ -146,52 +148,42 @@ def _candidate_articles(
     ).fetchall()
 
 
-def _select_diverse(rows: list[sqlite3.Row], limit: int) -> list[sqlite3.Row]:
+def _select_diverse(rows: list[sqlite3.Row], limit: int, max_per_topic: int = 2) -> list[sqlite3.Row]:
     """Pick up to `limit` items from `rows` (already ranked best-first).
 
     Contract:
     1. Prefer higher-ranked items.
-    2. Source diversity: take at most one item per source while other sources
-       remain; only repeat a source once distinct sources are exhausted.
-    3. Category diversity: if the result ended up single-category and a different
-       category is available in the pool, swap in the highest-ranked off-category
-       candidate by evicting the lowest-ranked currently-selected item (so we
-       gain breadth without dropping a higher-ranked pick).
+    2. Source diversity: at most one item per source while other sources remain.
+    3. Topic balance: no more than `max_per_topic` of the same topic while other
+       topics still have candidates — so the five don't cluster (e.g. several
+       medical/science items) when community/culture/animals/environment exist.
+    4. Always fill to `limit` when enough candidates exist: the source and topic
+       caps are relaxed (in that order) only as needed to reach the count.
     """
     selected: list[sqlite3.Row] = []
+    selected_ids: set = set()
     seen_sources: set = set()
+    topic_count: dict = {}
 
-    # Pass 1: best-first, one per source.
-    for row in rows:
-        if len(selected) >= limit:
-            break
-        if row["source_name"] in seen_sources:
-            continue
-        selected.append(row)
-        seen_sources.add(row["source_name"])
-
-    # Pass 2: if short on distinct sources, backfill best-first regardless.
-    if len(selected) < limit:
-        selected_ids = {row["id"] for row in selected}
+    def consider(enforce_source: bool, enforce_topic: bool) -> None:
         for row in rows:
             if len(selected) >= limit:
-                break
+                return
             if row["id"] in selected_ids:
                 continue
+            if enforce_source and row["source_name"] in seen_sources:
+                continue
+            topic = row["topic"]
+            if enforce_topic and topic_count.get(topic, 0) >= max_per_topic:
+                continue
             selected.append(row)
             selected_ids.add(row["id"])
+            seen_sources.add(row["source_name"])
+            topic_count[topic] = topic_count.get(topic, 0) + 1
 
-    # Pass 3: ensure >= 2 categories when the pool allows it.
-    categories = {row["default_category"] for row in selected}
-    if len(categories) < 2:
-        selected_ids = {row["id"] for row in selected}
-        for row in rows:
-            if row["id"] in selected_ids:
-                continue
-            if row["default_category"] not in categories:
-                selected[-1] = row  # evict the lowest-ranked selected item
-                break
-
+    consider(enforce_source=True, enforce_topic=True)    # distinct source, topic-balanced
+    consider(enforce_source=True, enforce_topic=False)   # relax topic cap to fill
+    consider(enforce_source=False, enforce_topic=False)  # relax source too, last resort
     return selected