feat(narrative): orchestrate round translation with state updates

translate_round() ties together:
  - reading the round's actions from actions.jsonl
  - generating prose via the LLM
  - updating per-character emotional state
  - persisting the beat, characters, and file offset

Full test suite now at 19/19 passing across action mapper, character
engine, story store, and translator.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

backend/app/services/narrative/narrative_translator.py

@@ -183,3 +183,57 @@ def generate_prose(actions: list, characters: list, tone: str, previous_beats: l
         previous=prev_prose or "(this is the first scene)",
     )
     return call_llm(prompt)
+
+
+# ---------------------------------------------------------------------------
+# Round orchestration — the entry point most callers use
+# ---------------------------------------------------------------------------
+
+def translate_round(sim_dir: str, platform: str, target_round: int, tone: str = "neutral") -> dict:
+    """Translate a single simulation round into a story beat end-to-end.
+
+    Steps:
+      1. Read the round's actions from `{sim_dir}/{platform}/actions.jsonl`
+         (starting from the saved file offset).
+      2. Generate prose via the LLM using current character state.
+      3. Apply emotional-state deltas for every action.
+      4. Persist the new beat, updated character state, and new file offset.
+
+    Returns the newly created story beat dict.
+    """
+    # Lazy imports keep module import lightweight for unit tests of helpers
+    from app.services.narrative.story_store import StoryStore
+    from app.services.narrative.character_engine import apply_action_emotional_delta
+
+    store = StoryStore(sim_dir)
+    actions_path = os.path.join(sim_dir, platform, "actions.jsonl")
+    start_offset = store.get_file_offset(platform)
+
+    actions, new_offset = read_actions_for_round(actions_path, start_offset, target_round)
+
+    characters = store.load_characters()
+    previous_beats = store.get_all_beats()
+
+    prose = generate_prose(actions, characters, tone, previous_beats)
+
+    # De-duplicated list of character names that acted this round
+    involved = sorted({a.get("agent_name") for a in actions if a.get("agent_name")})
+
+    # Mutate character emotional state in-place, then persist
+    char_by_name = {c["name"]: c for c in characters}
+    for action in actions:
+        char = char_by_name.get(action.get("agent_name"))
+        if char:
+            apply_action_emotional_delta(char, action.get("action_type", ""))
+    store.save_characters(list(char_by_name.values()))
+
+    beat = {
+        "round": target_round,
+        "prose": prose,
+        "characters": involved,
+        "action_count": len(actions),
+        "platform": platform,
+    }
+    store.append_beat(beat)
+    store.set_file_offset(platform, new_offset)
+    return beat

backend/tests/test_narrative_translator.py

@@ -79,3 +79,42 @@ def test_generate_prose_calls_llm_with_context():
 def test_generate_prose_empty_actions_returns_placeholder():
     result = generate_prose([], [], tone="any", previous_beats=[])
     assert "quiet" in result.lower() or "pause" in result.lower()
+
+
+# ---- Round orchestration tests ----
+from app.services.narrative.narrative_translator import translate_round
+from app.services.narrative.story_store import StoryStore
+
+
+def test_translate_round_produces_beat(tmp_path):
+    sim_dir = str(tmp_path / "sim_test")
+    os.makedirs(sim_dir)
+    platform_dir = os.path.join(sim_dir, "twitter")
+    os.makedirs(platform_dir)
+    actions_path = os.path.join(platform_dir, "actions.jsonl")
+
+    lines = [
+        {"round": 1, "agent_id": 1, "agent_name": "Alice", "action_type": "CREATE_POST",
+         "action_args": {"content": "Hi"}, "success": True, "timestamp": "t"},
+        {"event_type": "round_end", "round": 1, "timestamp": "t"},
+    ]
+    with open(actions_path, "w") as f:
+        for line in lines:
+            f.write(json.dumps(line) + "\n")
+
+    store = StoryStore(sim_dir)
+    store.save_characters([{
+        "id": "1", "name": "Alice",
+        "emotional_state": {"current": {k: 0.0 for k in ["anger","fear","joy","sadness","trust","surprise"]}},
+    }])
+
+    with patch("app.services.narrative.narrative_translator.call_llm") as mock_llm:
+        mock_llm.return_value = "Alice spoke into the void."
+        beat = translate_round(sim_dir, platform="twitter", target_round=1, tone="neutral")
+
+    assert beat["round"] == 1
+    assert beat["prose"] == "Alice spoke into the void."
+    assert "Alice" in beat.get("characters", [])
+
+    stored_beats = store.get_all_beats()
+    assert len(stored_beats) == 1