Hero emotional-safety guardrail + calmer card polish

Hero guardrail (core to the promise, not cosmetic):
- New hero.py: the lead story is chosen with a stricter filter than the rest of
  the brief — very low cortisol/ragebait and no grief/medical/violence terms
  (cancer, glioblastoma, death, diagnosis, ...). Such constructive-but-charged
  stories stay among the five; they just never lead by default.
- /api/brief applies user avoid-terms FIRST, then lead_with_gentle, so personal
  boundaries always take precedence over the general guardrail.
- Verified live: the brief no longer leads with a glioblastoma story.

Card polish (per review):
- Secondary cards with no real image are now text-first (no row of empty media
  bands); hero still always shows media or a typographic fallback.
- Inline tuning actions are quiet until hover/focus on pointer devices, and stay
  visible (softer) on touch — less interface machinery.

Tests: hero safety + lead reordering (70 total).

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

goodnews/api.py

@@ -30,6 +30,7 @@ from pydantic import BaseModel
 from . import feeds, queries
 from .db import connect, init_db
 from .filters import filter_articles, prefs_from_json
+from .hero import lead_with_gentle
 from .llm import LocalModelClient
 from .moods import MOODS
 from .taxonomy import FLAVORS, TOPICS
@@ -280,7 +281,11 @@ def create_app() -> FastAPI:
         items = data["items"]
         if not fp.is_empty():
             # MVP: filter the stored brief DOWN; no refill from outside the brief.
+            # Runs before hero selection, so personal avoid-terms take precedence.
             items = filter_articles(items, fp, datetime.now(timezone.utc))
+        # Lead with an emotionally-safe story (constructive-but-charged stories
+        # stay in the five, just not as the first thing seen).
+        items = lead_with_gentle(items)
         return BriefResponse(
             brief_date=data["brief_date"],
             title=data["title"],

goodnews/hero.py

@@ -0,0 +1,53 @@
+"""Hero (lead-story) emotional-safety guardrail.
+
+The single most prominent thing on the page must feel safe to encounter before
+coffee. A story can be objectively hopeful — a cancer breakthrough, say — and
+still be personally painful to meet unprompted at the top of the page.
+
+So the lead is chosen with a STRICTER filter than the rest of the brief: it must
+be very calm and must not mention grief/medical/violence terms. Such stories are
+still welcome among the five — they just don't lead by default. Per-user
+avoid-terms are applied to the brief before this runs, so a reader's personal
+boundaries always take precedence.
+"""
+
+from __future__ import annotations
+
+from .filters import text_matches_avoid_terms
+
+# Terms too emotionally charged to lead with, even inside a constructive story.
+HERO_GUARD_TERMS = [
+    "cancer", "glioblastoma", "tumor", "tumour", "carcinoma", "chemotherapy", "chemo",
+    "death", "died", "dies", "dying", "dead", "fatal", "terminal", "deadly",
+    "diagnosis", "diagnosed", "dementia", "alzheimer", "parkinson",
+    "suicide", "murder", "killed", "killing", "war", "assault", "abuse",
+    "overdose", "hospice", "palliative", "disease",
+]
+
+HERO_MAX_CORTISOL = 1
+HERO_MAX_RAGEBAIT = 1
+
+
+def safe_to_lead(article: dict) -> bool:
+    """True if an article is calm enough to be the page's lead."""
+    if (article.get("cortisol_score") or 0) > HERO_MAX_CORTISOL:
+        return False
+    if (article.get("ragebait_score") or 0) > HERO_MAX_RAGEBAIT:
+        return False
+    blob = f"{article.get('title') or ''} {article.get('description') or ''}"
+    return not text_matches_avoid_terms(blob, HERO_GUARD_TERMS)
+
+
+def lead_with_gentle(items: list[dict]) -> list[dict]:
+    """Reorder so the highest-ranked lead-safe item is first (the hero).
+
+    The remaining items keep their ranked order. If nothing qualifies, the list
+    is returned unchanged — the brief still needs a lead, and we never drop a
+    constructive story, we only decline to make a charged one the first thing seen.
+    """
+    for i, article in enumerate(items):
+        if safe_to_lead(article):
+            if i > 0:
+                return [items[i], *items[:i], *items[i + 1:]]
+            return items
+    return items