Phase B1: multi-tag groupings model (backend)

Three-layer organization: primary topic (one per article, for ranking and
brief balance) + grouping tags (1-4 per article from a controlled vocabulary,
the organic "wandering" axis) + tonal flavor.

- taxonomy: add technology + learning topics; 4 calm tag families
  (Discovery & Wonder, People & Kindness, Solutions & Progress, Mind & Craft)
  defined in code, not the DB; ALLOWED_TAGS union + coerce_tags validation.
- db: article_tags(article_id, tag) join table + tag index.
- llm: tags added to the classifier json_schema (enum-constrained, maxItems 4)
  and system prompt; normalize_scores coerces tags; upsert_article_score
  replaces a row's tags atomically on every (re)classification.
- queries: feed gains a tag filter and exposes tags via group_concat; tag_counts.
- api: Article.tags, feed tag param, and /api/families with per-tag counts.
- tests: coerce/normalize/upsert/tag-filter/reclassify-replace/tag_counts +
  /api/families. 99 passing.

Corpus reclassify (re-tag + new primary topics) runs separately against the
local LLM. Frontend (B2) pairs with this; the live site is unchanged until then.

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

goodnews/llm.py

@@ -9,11 +9,15 @@ from collections.abc import Callable
 from dataclasses import dataclass
 
 from .taxonomy import (
+    ALLOWED_TAGS,
     FLAVORS,
+    MAX_TAGS,
     TOPICS,
     coerce_flavor,
+    coerce_tags,
     coerce_topic,
     flavors_prompt_block,
+    tags_prompt_block,
     topics_prompt_block,
 )
 
@@ -42,6 +46,7 @@ CLASSIFICATION_SCHEMA = {
         "accepted",
         "topic",
         "flavor",
+        "tags",
         "reason_code",
         "reason_text",
     ],
@@ -56,6 +61,7 @@ CLASSIFICATION_SCHEMA = {
         "accepted": {"type": "boolean"},
         "topic": {"type": "string", "enum": list(TOPICS)},
         "flavor": {"type": "string", "enum": list(FLAVORS)},
+        "tags": {"type": "array", "items": {"type": "string", "enum": list(ALLOWED_TAGS)}, "maxItems": MAX_TAGS},
         "reason_code": {"type": "string"},
         "reason_text": {"type": "string"},
     },
@@ -76,14 +82,20 @@ Judge emotional aftertaste, not simple positivity. Accept stories that leave a r
 
 Reject stories centered on fear, outrage, partisan conflict, crime, tragedy, disaster repetition, celebrity drama, market panic, or corporate PR without clear public benefit.
 
-Also assign one topic and one flavor, choosing the single best fit.
+Also assign one primary topic and one flavor (the single best fit), plus 1-4 grouping tags.
 
-Topic (what the story is about):
+Primary topic (what the story is mainly about):
 {topics}
 
 Flavor (why it belongs in a calm, uplifting digest):
 {flavors}
 
+Grouping tags — choose ONLY from this controlled vocabulary:
+{tags}
+
+Tag discipline: assign 1-4 tags; prefer fewer, stronger ones; never tag by weak
+association; pick tags a reader would reasonably use to find this story later.
+
 Return only JSON with this exact shape:
 {{
   "constructive_score": 0,
@@ -96,10 +108,11 @@ Return only JSON with this exact shape:
   "accepted": false,
   "topic": "one_of_the_allowed_topics",
   "flavor": "one_of_the_allowed_flavors",
+  "tags": ["one_to_four_allowed_tags"],
   "reason_code": "short_snake_case",
   "reason_text": "one concise sentence"
 }}
-""".format(topics=topics_prompt_block(), flavors=flavors_prompt_block())
+""".format(topics=topics_prompt_block(), flavors=flavors_prompt_block(), tags=tags_prompt_block())
 
 
 @dataclass
@@ -285,6 +298,7 @@ def normalize_scores(data: dict, model_name: str) -> dict:
         "accepted": 1 if bool(data.get("accepted")) else 0,
         "topic": coerce_topic(data.get("topic")),
         "flavor": coerce_flavor(data.get("flavor")),
+        "tags": coerce_tags(data.get("tags")),
         "reason_code": str(data.get("reason_code") or "model_no_reason")[:120],
         "reason_text": str(data.get("reason_text") or "")[:1000],
         "model_name": model_name,
@@ -333,6 +347,12 @@ def upsert_article_score(conn: sqlite3.Connection, article_id: int, scores: dict
             scores["model_name"],
         ),
     )
+    # Replace this article's grouping tags (controlled vocabulary, 0-4).
+    conn.execute("DELETE FROM article_tags WHERE article_id = ?", (article_id,))
+    for tag in scores.get("tags") or []:
+        conn.execute(
+            "INSERT OR IGNORE INTO article_tags (article_id, tag) VALUES (?, ?)", (article_id, tag)
+        )
 
 
 def _classification_candidates(