Strip leaked peak-hour info from observation, force tool discovery

- Remove audience_active_hours and competitor_recent_post_hours from
format_obs: model was reading them and skipping query_* entirely
- Rewrite TOOL POLICY: tools are free, encourage aggressive discovery,
multiple segment queries on day 1 instead of one
- Empty Tool results now reads "(none — call query_* tools to discover)"
to nudge the model on its first turn

Made-with: Cursor

training/train_grpo.ipynb

@@ -441,18 +441,21 @@
         "- topic:        free-form string\n",
         "- empty scheduled_actions = full day rest\"\"\")\n",
         "\n",
-    "SYSTEM_PROMPT = _SYSTEM_BASE + textwrap.dedent(\"\"\"\n",
-    "\n",
-    "TOOL POLICY (MANDATORY — empty tool_calls on day 1 = wasted day):\n",
-    "- days_elapsed == 0  -> call AT LEAST these in tool_calls:\n",
-    "    {\"name\": \"query_trends\",        \"arguments\": {\"niche\": \"<one of TOPIC_CATEGORIES keys>\"}}\n",
-    "    {\"name\": \"query_audience\",      \"arguments\": {\"segment_id\": \"young_professionals\"}}\n",
-    "    {\"name\": \"query_creator_pool\",  \"arguments\": {}}\n",
-    "    {\"name\": \"query_competitor\",    \"arguments\": {\"competitor_id\": \"niche_expert\", \"window_days\": 7}}\n",
-    "- days_elapsed >= 1  -> before scheduling, call:\n",
-    "    {\"name\": \"predict_engagement\",  \"arguments\": {\"scheduled_actions\": [...]}}\n",
-    "  and at least one query_* tool whose result you don't already have in Tool results.\n",
-    "- audience_active_hours in the observation is a coarse hint; query_audience returns ranked topic affinities you cannot get otherwise.\"\"\")\n",
+        "SYSTEM_PROMPT = _SYSTEM_BASE + textwrap.dedent(\"\"\"\n",
+        "\n",
+        "TOOL POLICY (tool_calls cost nothing — call them aggressively):\n",
+        "- The observation tells you ONLY your account stats. Audience peak hours, segment\n",
+        "  affinities, trending topics/tags and competitor schedules are NOT given. You must\n",
+        "  discover them via tool_calls and read them from `Tool results` next turn.\n",
+        "- days_elapsed == 0  -> call EVERY discovery tool you might need, e.g.:\n",
+        "    {\"name\": \"query_trends\",        \"arguments\": {\"niche\": \"<TOPIC_CATEGORIES key>\"}}\n",
+        "    {\"name\": \"query_audience\",      \"arguments\": {\"segment_id\": \"young_professionals\"}}\n",
+        "    {\"name\": \"query_audience\",      \"arguments\": {\"segment_id\": \"students\"}}\n",
+        "    {\"name\": \"query_creator_pool\",  \"arguments\": {}}\n",
+        "    {\"name\": \"query_competitor\",    \"arguments\": {\"competitor_id\": \"niche_expert\", \"window_days\": 7}}\n",
+        "- days_elapsed >= 1  -> before scheduling posts, call:\n",
+        "    {\"name\": \"predict_engagement\",  \"arguments\": {\"scheduled_actions\": [...]}}\n",
+        "  and any query_* whose result is missing from `Tool results`.\"\"\")\n",
         "SYSTEM_PROMPT_EVAL = SYSTEM_PROMPT\n",
         "SYSTEM_PROMPT_TRAIN = SYSTEM_PROMPT\n",
         "\n",
@@ -466,23 +469,18 @@
         "        signals_str = (f\"Signals: watch={signals.watch_time:.3f} \"\n",
         "                       f\"sends={signals.sends_per_reach:.3f} \"\n",
         "                       f\"saves={signals.saves:.3f}\\n\")\n",
-        "    meta = getattr(obs, \"metadata\", None) or {}\n",
-        "    aud = meta.get(\"audience_active_hours\") or []\n",
-        "    comp = meta.get(\"competitor_recent_post_hours\") or []\n",
-        "    tool_str = \"\"\n",
-        "    for tr in getattr(obs, \"tool_results\", []):\n",
-        "        if tr.success:\n",
-        "            tool_str += f\"  {tr.name}: {json.dumps(tr.data)}\\n\"\n",
-        "    if not tool_str:\n",
-        "        tool_str = \"  (none)\\n\"\n",
-        "    return (f\"Day: {day_name} | days_elapsed={obs.days_elapsed}\\n\"\n",
-        "            f\"Energy: {obs.creator_energy:.2f} | Followers: {obs.follower_count}\\n\"\n",
-        "            f\"Engagement: {obs.engagement_rate:.3f} | Queue: {obs.content_queue_size}\\n\"\n",
-        "            f\"{signals_str}\"\n",
-        "            f\"audience_active_hours: {aud}\\n\"\n",
-        "            f\"competitor_recent_post_hours: {comp}\\n\"\n",
-        "            f\"Tool results:\\n{tool_str}\"\n",
-        "            f\"Plan today's actions (JSON only):\")\n",
+    "    tool_str = \"\"\n",
+    "    for tr in getattr(obs, \"tool_results\", []):\n",
+    "        if tr.success:\n",
+    "            tool_str += f\"  {tr.name}: {json.dumps(tr.data)}\\n\"\n",
+    "    if not tool_str:\n",
+    "        tool_str = \"  (none — call query_* tools to discover)\\n\"\n",
+    "    return (f\"Day: {day_name} | days_elapsed={obs.days_elapsed}\\n\"\n",
+    "            f\"Energy: {obs.creator_energy:.2f} | Followers: {obs.follower_count}\\n\"\n",
+    "            f\"Engagement: {obs.engagement_rate:.3f} | Queue: {obs.content_queue_size}\\n\"\n",
+    "            f\"{signals_str}\"\n",
+    "            f\"Tool results:\\n{tool_str}\"\n",
+    "            f\"Plan today's actions (JSON only):\")\n",
         "\n",
         "\n",
         "def is_well_formed_response(text):\n",