Article pages: structured "Why it belongs" editorial read

Per Codex — make /a/<id> feel like Upbeat Bytes has editorial judgment, not just
a summary wrapper. Trust-building, short, not an essay.

* article_summaries gains what_happened / why_matters / why_belongs (+ migration).
* summarize.explain_article: a separate, fallback-able LLM pass producing three
  short notes (parsed from a labelled WHAT/MATTERS/BELONGS format). generate_summary
  now stores them alongside the summary, and tops up older summaries on next view.
  get_explanation returns them only when all three are present.
* API: share_page + /api/summary expose the explanation.
* share.py: renders the three-part section (accent rule) when complete; otherwise
  the single "Why it's here" reason line is the calm fallback. The page polls and
  swaps in both the summary and the section as they cache.

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

goodnews/api.py

@@ -886,9 +886,12 @@ def create_app() -> FastAPI:
             if not row or row["duplicate_of"] is not None or not row["accepted"]:
                 return not_found
             summary = summarize.get_summary(conn, aid)
-        if not summary:
-            _kick_summary(aid, background_tasks)  # generate for next time; page polls
-        return HTMLResponse(share.render_share_page(dict(row), PUBLIC_BASE_URL, summary=summary))
+            explanation = summarize.get_explanation(conn, aid)
+        if not summary or not explanation:
+            _kick_summary(aid, background_tasks)  # generate/top-up for next time; page polls
+        return HTMLResponse(
+            share.render_share_page(dict(row), PUBLIC_BASE_URL, summary=summary, explanation=explanation)
+        )
 
     # --- Privacy-respecting first-party analytics -------------------------
 
@@ -1243,10 +1246,11 @@ def create_app() -> FastAPI:
     def article_summary(article_id: int, background_tasks: BackgroundTasks) -> dict:
         with get_conn() as conn:
             summary = summarize.get_summary(conn, article_id)
+            explanation = summarize.get_explanation(conn, article_id)
         if summary:
-            return {"status": "ready", "summary": summary}
+            return {"status": "ready", "summary": summary, "explanation": explanation}
         _kick_summary(article_id, background_tasks)
-        return {"status": "pending", "summary": None}
+        return {"status": "pending", "summary": None, "explanation": None}
 
     @app.get("/today", response_class=HTMLResponse)
     def today_digest() -> HTMLResponse:

goodnews/db.py

@@ -208,6 +208,9 @@ CREATE TABLE IF NOT EXISTS user_prefs (
 CREATE TABLE IF NOT EXISTS article_summaries (
     article_id INTEGER PRIMARY KEY REFERENCES articles(id) ON DELETE CASCADE,
     summary TEXT NOT NULL,
+    what_happened TEXT,
+    why_matters TEXT,
+    why_belongs TEXT,
     model TEXT,
     created_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP
 );
@@ -363,3 +366,9 @@ def _migrate(conn: sqlite3.Connection) -> None:
     rep_cols = {row["name"] for row in conn.execute("PRAGMA table_info(feedback_replies)")}
     if rep_cols and "message_html" not in rep_cols:
         conn.execute("ALTER TABLE feedback_replies ADD COLUMN message_html TEXT")
+
+    # article_summaries: structured "Why it belongs" fields added later.
+    sum_cols = {row["name"] for row in conn.execute("PRAGMA table_info(article_summaries)")}
+    for column in ("what_happened", "why_matters", "why_belongs"):
+        if sum_cols and column not in sum_cols:
+            conn.execute(f"ALTER TABLE article_summaries ADD COLUMN {column} TEXT")

goodnews/share.py

@@ -17,7 +17,8 @@ def _tag(name: str, content: str | None, attr: str = "property") -> str:
     return f'<meta {attr}="{escape(name)}" content="{escape(content)}">'
 
 
-def render_share_page(article: dict, base_url: str, summary: str | None = None) -> str:
+def render_share_page(article: dict, base_url: str, summary: str | None = None,
+                      explanation: dict | None = None) -> str:
     aid = article["id"]
     title = (article.get("title") or "Upbeat Bytes").strip()
     why = (article.get("reason_text") or article.get("description")
@@ -63,22 +64,54 @@ def render_share_page(article: dict, base_url: str, summary: str | None = None)
     groupings = f'<div class="chips">{chips}</div>' if chips else ""
     if summary:
         summary_block = f'<p class="summary" id="summary">{escape(summary)}</p>'
-        poll = ""
     else:
         pending = (
             f"✦ We’re putting together a quick summary — in the meantime, "
             f"read the full story at {escape(source)} below."
         )
         summary_block = f'<p class="summary pending" id="summary">{pending}</p>'
-        # Quietly poll; swap the real summary in the moment it's cached.
+
+    # The structured "Why it belongs" editorial section — only when the LLM gave a
+    # clean three-part read; otherwise the single reason line is the calm fallback.
+    def _why_row(lbl: str, text: str) -> str:
+        return f'<div class="why-row"><span class="lbl">{lbl}</span><p>{escape(text)}</p></div>'
+
+    if explanation:
+        why_block = (
+            '<div class="why" id="why">'
+            + _why_row("What happened", explanation["what_happened"])
+            + _why_row("Why it matters", explanation["why_matters"])
+            + _why_row("Why it belongs here", explanation["why_belongs"])
+            + '</div>'
+        )
+    else:
+        why_block = f'<div class="why" id="why"><div class="why-row"><span class="lbl">Why it’s here</span><p>{escape(why)}</p></div></div>'
+
+    # Quietly poll until BOTH the summary and the structured read are cached, then
+    # swap each in (older summaries get topped up with the section on first view).
+    poll = ""
+    if not summary or not explanation:
         poll = f"""<script>
 (function(){{
   var id={aid}, box=document.getElementById('summary'), n=0;
+  function row(l,t){{
+    var d=document.createElement('div'); d.className='why-row';
+    var s=document.createElement('span'); s.className='lbl'; s.textContent=l;
+    var p=document.createElement('p'); p.textContent=t;
+    d.appendChild(s); d.appendChild(p); return d;
+  }}
   function go(){{
     n++;
     fetch('/api/summary/'+id).then(function(r){{return r.json();}}).then(function(d){{
-      if(d&&d.status==='ready'&&d.summary){{ box.textContent=d.summary; box.className='summary'; }}
-      else if(n<12){{ setTimeout(go,2500); }}
+      var done=true;
+      if(d&&d.summary){{ box.textContent=d.summary; box.className='summary'; }} else {{ done=false; }}
+      if(d&&d.explanation){{
+        var w=document.getElementById('why');
+        if(w){{ w.innerHTML=''; w.appendChild(row('What happened',d.explanation.what_happened));
+          w.appendChild(row('Why it matters',d.explanation.why_matters));
+          w.appendChild(row('Why it belongs here',d.explanation.why_belongs)); }}
+      }} else {{ done=false; }}
+      if(!done && n<12){{ setTimeout(go,2500); }}
     }}).catch(function(){{ if(n<12) setTimeout(go,3000); }});
   }}
   go();
@@ -140,9 +173,13 @@ def render_share_page(article: dict, base_url: str, summary: str | None = None)
         font-size:1.7rem; line-height:1.2; margin:6px 0 14px; }}
   .summary {{ font-size:1.05rem; color:var(--ink); margin:0 0 18px; }}
   .summary.pending {{ color:var(--muted); font-style:italic; }}
-  .why {{ color:#3b4754; font-size:.92rem; margin:0 0 22px; }}
+  .why {{ color:#3b4754; font-size:.92rem; margin:0 0 22px;
+          border-left:2px solid var(--accent); padding-left:14px; }}
+  .why-row {{ margin:0 0 12px; }}
+  .why-row:last-child {{ margin-bottom:0; }}
+  .why-row p {{ margin:0; }}
   .why .lbl {{ display:block; text-transform:uppercase; letter-spacing:.08em; font-size:.7rem;
-               color:var(--muted); margin-bottom:3px; }}
+               color:var(--accent-deep); font-weight:600; margin-bottom:2px; }}
   .chips {{ display:flex; flex-wrap:wrap; gap:7px; margin:0 0 22px; }}
   .chip {{ background:#e0eef3; color:var(--accent-deep); border-radius:999px; padding:3px 11px;
            font-size:.7rem; text-transform:uppercase; letter-spacing:.06em; }}
@@ -165,7 +202,7 @@ def render_share_page(article: dict, base_url: str, summary: str | None = None)
         {source_html}
         <h1>{escape(title)}</h1>
         {summary_block}
-        <p class="why"><span class="lbl">Why it's here</span>{escape(why)}</p>
+        {why_block}
         {groupings}
         <div class="actions">
           <a class="primary" href="{escape(src_url)}" target="_blank" rel="noopener" data-src-click>Read the full story at {escape(source)}</a>

goodnews/summarize.py

@@ -76,6 +76,39 @@ def summarize_article(client: LocalModelClient, title: str, snippet: str, body_t
     return (client.chat_text(messages) or "").strip()[:1200]
 
 
+_EXPLAIN_SYSTEM = (
+    "You are the calm editor of a constructive-news site. For the given story write three "
+    "very short, plain-language notes — 1 to 2 factual sentences each, in your own words, no "
+    "markdown, no preamble, no quotes, no hype. Use EXACTLY this format and these labels, each "
+    "on its own line:\n"
+    "WHAT: <what actually happened — the plain gist>\n"
+    "MATTERS: <why it matters — the real-world, constructive significance>\n"
+    "BELONGS: <why it fits a calm, constructive news site — the human benefit, agency, or "
+    "grounded hope in it>"
+)
+
+
+def _parse_explain(text: str) -> dict:
+    def grab(label: str) -> str | None:
+        m = re.search(rf"\b{label}\s*:\s*(.+?)(?=\n\s*[A-Z]+\s*:|\Z)", text or "", re.IGNORECASE | re.DOTALL)
+        if not m:
+            return None
+        val = _WS.sub(" ", m.group(1)).strip().strip("-•* ").strip()
+        return val[:400] or None
+
+    return {"what_happened": grab("WHAT"), "why_matters": grab("MATTERS"), "why_belongs": grab("BELONGS")}
+
+
+def explain_article(client: LocalModelClient, title: str, snippet: str, body_text: str) -> dict:
+    """Three short editorial notes (what happened / why it matters / why it belongs)."""
+    material = (body_text or snippet or title or "")[:4000]
+    user = f"Title: {title}\n\nArticle text:\n{material}\n\nWrite the three notes."
+    text = client.chat_text(
+        [{"role": "system", "content": _EXPLAIN_SYSTEM}, {"role": "user", "content": user}]
+    ) or ""
+    return _parse_explain(text)
+
+
 def get_summary(conn: sqlite3.Connection, article_id: int) -> str | None:
     row = conn.execute(
         "SELECT summary FROM article_summaries WHERE article_id = ?", (article_id,)
@@ -83,25 +116,49 @@ def get_summary(conn: sqlite3.Connection, article_id: int) -> str | None:
     return row["summary"] if row else None
 
 
+def get_explanation(conn: sqlite3.Connection, article_id: int) -> dict | None:
+    """The structured 'Why it belongs' notes — only if all three are present (else
+    the page falls back to summary + reason_text)."""
+    row = conn.execute(
+        "SELECT what_happened, why_matters, why_belongs FROM article_summaries WHERE article_id = ?",
+        (article_id,),
+    ).fetchone()
+    if row and row["what_happened"] and row["why_matters"] and row["why_belongs"]:
+        return dict(row)
+    return None
+
+
 def generate_summary(conn: sqlite3.Connection, article_id: int, client: LocalModelClient | None = None) -> str | None:
-    """Generate + cache a summary for one article. Returns it, or None if skipped."""
-    if get_summary(conn, article_id):
-        return get_summary(conn, article_id)  # already cached
+    """Generate + cache a summary AND the structured explanation for one article.
+
+    Returns the summary, or None if skipped. Idempotent: a fully-cached article
+    (summary + explanation) is returned as-is; an older summary missing the
+    explanation is topped up on the next call (so existing pages gain the section).
+    """
+    existing = conn.execute(
+        "SELECT summary, why_belongs FROM article_summaries WHERE article_id = ?", (article_id,)
+    ).fetchone()
+    if existing and existing["summary"] and existing["why_belongs"]:
+        return existing["summary"]  # summary + a complete explanation already cached
     row = conn.execute(
         "SELECT a.title, a.description, a.canonical_url, a.duplicate_of, s.accepted "
         "FROM articles a LEFT JOIN article_scores s ON s.article_id = a.id WHERE a.id = ?",
         (article_id,),
     ).fetchone()
     if not row or row["duplicate_of"] is not None or not row["accepted"]:
-        return None
+        return existing["summary"] if existing else None
     client = client or LocalModelClient.from_env()
     body = _fetch_text(row["canonical_url"])
-    summary = summarize_article(client, row["title"], row["description"] or "", body)
+    summary = existing["summary"] if existing else summarize_article(
+        client, row["title"], row["description"] or "", body
+    )
     if not summary:
         return None
+    ex = explain_article(client, row["title"], row["description"] or "", body)
     conn.execute(
-        "INSERT OR REPLACE INTO article_summaries (article_id, summary, model) VALUES (?, ?, ?)",
-        (article_id, summary, client.model),
+        "INSERT OR REPLACE INTO article_summaries "
+        "(article_id, summary, what_happened, why_matters, why_belongs, model) VALUES (?, ?, ?, ?, ?, ?)",
+        (article_id, summary, ex["what_happened"], ex["why_matters"], ex["why_belongs"], client.model),
     )
     conn.commit()
     # Attention-triggered image enrichment: a summarized article is one a reader

tests/test_explain.py

@@ -0,0 +1,62 @@
+from goodnews import summarize, share
+from goodnews.db import connect, init_db
+
+
+def test_parse_explain_labels():
+    txt = ("WHAT: A town planted 500 trees.\n"
+           "MATTERS: It cools streets and cuts flooding.\n"
+           "BELONGS: Neighbors did it themselves — local agency and clear benefit.")
+    ex = summarize._parse_explain(txt)
+    assert ex["what_happened"].startswith("A town planted")
+    assert "cools streets" in ex["why_matters"]
+    assert "local agency" in ex["why_belongs"]
+
+
+def test_parse_explain_partial():
+    ex = summarize._parse_explain("WHAT: only this")
+    assert ex["what_happened"] == "only this" and ex["why_matters"] is None and ex["why_belongs"] is None
+
+
+def _seed_article(c):
+    c.execute("INSERT INTO sources (id,name,feed_url) VALUES (1,'S','http://s/f')")
+    c.execute("INSERT INTO articles (id,source_id,canonical_url,title,url_hash) VALUES (1,1,'http://a/1','Town plants trees','h')")
+    c.execute("INSERT INTO article_scores (article_id,accepted) VALUES (1,1)")
+    c.commit()
+
+
+def test_get_explanation_requires_all_three(tmp_path):
+    c = connect(str(tmp_path / "e.db")); init_db(c); _seed_article(c)
+    c.execute("INSERT INTO article_summaries (article_id,summary,what_happened) VALUES (1,'s','w')"); c.commit()
+    assert summarize.get_explanation(c, 1) is None      # partial → fall back
+    c.execute("UPDATE article_summaries SET why_matters='m', why_belongs='b' WHERE article_id=1"); c.commit()
+    assert summarize.get_explanation(c, 1) == {"what_happened": "w", "why_matters": "m", "why_belongs": "b"}
+
+
+def test_generate_summary_stores_explanation(tmp_path, monkeypatch):
+    c = connect(str(tmp_path / "g.db")); init_db(c); _seed_article(c)
+
+    class FakeClient:
+        model = "test"
+        def chat_text(self, messages):
+            if "three" in messages[0]["content"].lower():
+                return "WHAT: trees planted.\nMATTERS: cooler streets.\nBELONGS: local agency, clear benefit."
+            return "A town planted trees, cooling streets."
+
+    monkeypatch.setattr(summarize, "_fetch_text", lambda url: "body text")
+    s = summarize.generate_summary(c, 1, client=FakeClient())
+    assert s and "planted trees" in s
+    ex = summarize.get_explanation(c, 1)
+    assert "agency" in ex["why_belongs"] and "cooler streets" in ex["why_matters"]
+
+
+def test_share_renders_structured_or_fallback():
+    art = {"id": 1, "title": "T", "source_name": "Src", "source_id": 1,
+           "canonical_url": "http://x", "reason_text": "calm reason", "tags": ""}
+    full = share.render_share_page(art, "http://b", summary="sum",
+                                   explanation={"what_happened": "WH", "why_matters": "WM", "why_belongs": "WB"})
+    assert "What happened" in full and "Why it matters" in full and "Why it belongs here" in full
+    assert "WH" in full and "WB" in full
+    assert "fetch('/api/summary/'" not in full   # both cached → no poll needed
+    # no explanation → the single "Why it’s here" reason line is the calm fallback
+    fb = share.render_share_page(art, "http://b", summary="sum", explanation=None)
+    assert "Why it’s here" in fb and "calm reason" in fb

tests/test_summary.py

@@ -21,6 +21,8 @@ def test_generate_caches_summary(tmp_path, monkeypatch):
     c = _db(tmp_path / "t.sqlite3")
     monkeypatch.setattr(summarize, "_fetch_text", lambda url: "Full article body about voles.")
     monkeypatch.setattr(summarize, "summarize_article", lambda client, t, s, b: "Our own short summary.")
+    monkeypatch.setattr(summarize, "explain_article",
+                        lambda client, t, s, b: {"what_happened": "w", "why_matters": "m", "why_belongs": "b"})
     # avoid constructing a real LLM client
     monkeypatch.setattr(summarize.LocalModelClient, "from_env", classmethod(lambda cls: SimpleNamespace(model="m")))
     out = summarize.generate_summary(c, 1)
@@ -50,7 +52,7 @@ def test_summary_endpoint_pending_then_ready(tmp_path, monkeypatch):
     first = tc.get("/api/summary/1").json()
     assert first["status"] == "pending"
     # TestClient runs the background task; the next call sees the cache
-    assert tc.get("/api/summary/1").json() == {"status": "ready", "summary": "Cached summary."}
+    assert tc.get("/api/summary/1").json() == {"status": "ready", "summary": "Cached summary.", "explanation": None}
     # and the share page now embeds it (no pending poll)
     html = tc.get("/a/1").text
     assert "Cached summary." in html