Telegram Q&A bot: real retrieval + direction (was answering blind)

The bot only knew P&L + positions, so "what's on the watchlist?" and "why
isn't it trading?" both fell back to "positions: none" — useless. Now it
retrieves the actual desk state and gives the user direction.

- gather_context: today P&L + recent closed trades (cross-day, robust to the
  midnight boundary) + open positions + the live WATCHLIST (states + scores) +
  READY SIGNALS (entry/stop/vwap) + a why-blocked tally (risk_rule_triggered
  counts) + all-time.
- symbol_context: when a question names a ticker the bot knows, append a focused
  dive — its last trade + the setup criteria it passed/failed.
- Help menu (help / /start) lists exactly what it can answer; sent on startup so
  the user has direction. Sharper system prompt (concrete, quote real numbers,
  say what data exists when it can't answer).

Verified live: "what's on the watchlist?" -> "148 symbols: 97 blocked, 48 late,
3 ready (LAES, NOWL, PBK)"; "why isn't it trading?" -> coherent; "what happened
with VTAK?" -> "stop loss -$659, entry $1.14". Tests: help menu, retrieval
includes watchlist/blocks; 10/10 green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

telegram_bot.py

@@ -4,9 +4,10 @@ data, READ-ONLY, locked to the authorized chat, Ollama-backed.
 
 Runs as its OWN process (`python telegram_bot.py`) so it can NEVER impede the
 trading loop. It long-polls getUpdates; for each message from TELEGRAM_CHAT_ID it
-gathers a data snapshot from Postgres and asks the local Ollama model to answer.
-If Ollama is down/disabled it falls back to the raw snapshot. It never accepts
-trade commands — questions only (mirrors the view-only dashboard rule).
+RETRIEVES the relevant data from Postgres (a desk snapshot, plus a focused
+per-symbol dive when the question names a ticker) and asks the local Ollama model
+to answer. Falls back to the raw snapshot if Ollama is down. Never accepts trade
+commands. Sends a help menu so the user knows what it can answer (direction).
 
 Env: TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID (required), DATABASE_URL,
 OLLAMA_HOST / OLLAMA_MODEL / OLLAMA_TIMEOUT_SECONDS / OLLAMA_ENABLED.
@@ -17,78 +18,177 @@ from __future__ import annotations
 import json
 import logging
 import os
+import re
 import time
 import urllib.request
+from collections import Counter
 
 logger = logging.getLogger("telegram_bot")
 
 _API = "https://api.telegram.org"
-_ACTION_WORDS = ("buy", "sell", "flatten", "close all", "cancel", "short",
-                 "liquidate", "exit all", "place order", "submit")
+_ACTION_WORDS = ("buy ", "sell ", "flatten", "close all", "cancel ", "short ",
+                 "liquidate", "exit all", "place order", "submit order", "go long")
+_HELP_WORDS = ("help", "/help", "/start", "what can you", "what do you", "commands", "menu")
+_STOPWORDS = {"P&L", "PNL", "EOD", "VWAP", "ORB", "USD", "AI", "OK", "ET", "AM", "PM",
+              "WHY", "HOW", "THE", "AND", "ARE", "WAS", "DID", "ANY", "ALL", "NOT",
+              "YOU", "OUR", "WHAT", "WHEN", "BEST", "WORST", "TODAY", "NOW", "R"}
+
+HELP_TEXT = (
+    "\U0001F916 I'm your momentum desk assistant (read-only). Ask me in plain English about:\n"
+    "• today's P&L / trades — \"how did we do?\", \"best and worst trade?\"\n"
+    "• the book — \"what's open?\", \"how's PLUG doing?\"\n"
+    "• the watchlist — \"what's on the watchlist?\", \"anything ready?\"\n"
+    "• why we are / aren't trading — \"why no entries?\", \"what got blocked?\"\n"
+    "• a specific ticker — \"what happened with VTAK?\", \"why did we skip AMC?\"\n"
+    "• all-time — \"how are we doing overall?\"\n"
+    "I can't place or change trades — questions only."
+)
+
+
+def is_help(q: str) -> bool:
+    ql = q.strip().lower()
+    return ql in ("help", "/help", "/start", "menu") or any(w in ql for w in _HELP_WORDS)
+
+
+# --------------------------------------------------------------- data retrieval
+def _recent_closed(store, limit=12):
+    rows = []
+    for e in store.query_events(event_type="position_closed", limit=None):
+        p = json.loads(e["payload_json"])
+        rows.append((str(e["timestamp"]), p))
+    rows.sort(reverse=True)
+    return [p for _, p in rows[:limit]]
+
+
+def known_symbols(store) -> set:
+    syms = set()
+    try:
+        for w in store.query_events(event_type="symbol_state_changed", limit=None):
+            s = json.loads(w["payload_json"]).get("symbol")
+            if s:
+                syms.add(str(s).upper())
+    except Exception:  # noqa: BLE001
+        pass
+    return syms
 
 
-# ---------------------------------------------------------------- data context
 def gather_context(store) -> str:
-    """A compact, read-only snapshot of the current trading state for the LLM."""
     from storage.projections import (query_account_positions_snapshot,
-                                      query_session_pnl, query_alltime_score)
-    lines: list[str] = []
+                                     query_session_pnl, query_alltime_score,
+                                     query_watch_states_snapshot,
+                                     query_ready_signals_snapshot)
+    L: list[str] = []
     try:
         p = query_session_pnl(store)
-        lines.append(
-            f"TODAY P&L: total {p.get('total_pnl', 0):+.0f} "
-            f"(realized {p.get('realized_pnl', 0):+.0f}, "
-            f"unrealized {p.get('unrealized_pnl', 0):+.0f}); "
-            f"closed {p.get('closed_trades', 0)} "
-            f"W/L {p.get('wins', 0)}/{p.get('losses', 0)}.")
-        for t in (p.get("trades") or [])[:12]:
-            lines.append(f"  closed {t.get('symbol')}: {t.get('pnl', t.get('realized_pnl', 0)):+.0f} "
-                         f"({t.get('exit_reason', '')})")
+        L.append(f"TODAY (live session) P&L: total {p.get('total_pnl', 0):+.0f} "
+                 f"(realized {p.get('realized_pnl', 0):+.0f}, unrealized {p.get('unrealized_pnl', 0):+.0f}); "
+                 f"closed {p.get('closed_trades', 0)} W/L {p.get('wins', 0)}/{p.get('losses', 0)}.")
     except Exception as exc:  # noqa: BLE001
-        lines.append(f"(P&L unavailable: {exc})")
+        L.append(f"(today P&L unavailable: {exc})")
+    try:
+        rc = _recent_closed(store, 12)
+        if rc:
+            L.append("RECENT CLOSED TRADES (newest first):")
+            for t in rc:
+                L.append(f"  {t.get('symbol')}: {t.get('realized_pnl', 0):+.0f} "
+                         f"@entry {t.get('entry_price')} ({t.get('exit_reason', '')})")
+    except Exception:  # noqa: BLE001
+        pass
     try:
         snap = query_account_positions_snapshot(store)
         positions = snap[0]["positions"] if snap else []
         if positions:
-            lines.append(f"OPEN POSITIONS ({len(positions)}):")
+            L.append(f"OPEN POSITIONS ({len(positions)}):")
             for q in positions:
-                lines.append(
-                    f"  {q.get('symbol')}: qty {q.get('quantity')} @ "
-                    f"{q.get('avg_entry_price')} -> {q.get('current_price')} "
-                    f"(uPL {q.get('unrealized_pnl', 0):+.0f})")
+                L.append(f"  {q.get('symbol')}: qty {q.get('quantity')} @ {q.get('avg_entry_price')} "
+                         f"-> {q.get('current_price')} (uPL {q.get('unrealized_pnl', 0):+.0f})")
         else:
-            lines.append("OPEN POSITIONS: none (flat).")
-    except Exception as exc:  # noqa: BLE001
-        lines.append(f"(positions unavailable: {exc})")
-    try:
-        a = query_alltime_score(store)
-        lines.append(f"ALL-TIME: realized {a.get('total_realized', 0):+.0f} over "
-                     f"{a.get('trading_days', '?')} days, win% "
-                     f"{(a.get('win_rate') or 0)*100:.0f}, trades {a.get('trades', '?')}.")
+            L.append("OPEN POSITIONS: none (book is flat).")
     except Exception:  # noqa: BLE001
         pass
-    return "\n".join(lines)
+    try:
+        ws = query_watch_states_snapshot(store)
+        if ws:
+            by_state = Counter(str(w.get("state") or "unknown") for w in ws)
+            L.append(f"WATCHLIST: {len(ws)} symbols — " +
+                     ", ".join(f"{n} {st}" for st, n in by_state.most_common()))
+            hot = [w for w in ws if str(w.get("state")) in ("ready", "armed")]
+            for w in sorted(hot, key=lambda x: -(x.get("last_score") or 0))[:8]:
+                L.append(f"  {w.get('symbol')}: {w.get('state')} (score {w.get('last_score')})")
+    except Exception:  # noqa: BLE001
+        pass
+    try:
+        rs = query_ready_signals_snapshot(store)
+        if rs:
+            L.append(f"READY SIGNALS ({len(rs)}):")
+            for s in rs[:8]:
+                av = s.get("above_vwap")
+                L.append(f"  {s.get('symbol')}: entry {s.get('entry_price')} stop {s.get('stop_loss_price')} "
+                         f"vwap {s.get('vwap')} {'(above VWAP)' if av else '(BELOW VWAP)' if av is False else ''}")
+    except Exception:  # noqa: BLE001
+        pass
+    try:  # why we are/aren't entering — the recent block tally
+        blocks = Counter()
+        for e in store.query_events(event_type="risk_rule_triggered", limit=300):
+            blocks[json.loads(e["payload_json"]).get("rule_type")] += 1
+        if blocks:
+            L.append("RECENT ENTRY BLOCKS (why entries were skipped): " +
+                     ", ".join(f"{rt}×{n}" for rt, n in blocks.most_common(8)))
+    except Exception:  # noqa: BLE001
+        pass
+    try:
+        a = query_alltime_score(store)
+        L.append(f"ALL-TIME: realized {a.get('total_realized', 0):+.0f} over "
+                 f"{a.get('trading_days', '?')} days, win% {(a.get('win_rate') or 0)*100:.0f}, "
+                 f"trades {a.get('trades', '?')}.")
+    except Exception:  # noqa: BLE001
+        pass
+    return "\n".join(L)
+
+
+def symbol_context(store, sym: str) -> str:
+    from storage.projections import query_symbol_criteria
+    L = [f"\n=== DETAIL: {sym} ==="]
+    try:
+        for t in _recent_closed(store, 50):
+            if str(t.get("symbol")).upper() == sym:
+                L.append(f"last trade: {t.get('realized_pnl', 0):+.0f} @entry {t.get('entry_price')} "
+                         f"stop {t.get('stop_loss_price')} ({t.get('exit_reason', '')})")
+                break
+    except Exception:  # noqa: BLE001
+        pass
+    try:
+        c = query_symbol_criteria(store, sym)
+        crit = c.get("criteria") or []
+        if crit:
+            failed = [x.get("key") or x.get("label") for x in crit if not x.get("passed")]
+            passed = [x.get("key") or x.get("label") for x in crit if x.get("passed")]
+            L.append(f"setup score {c.get('score')} ({c.get('passed_count')} criteria passed); "
+                     f"passed: {', '.join(passed) or 'none'}; failed: {', '.join(failed) or 'none'}")
+    except Exception:  # noqa: BLE001
+        pass
+    return "\n".join(L) if len(L) > 1 else ""
 
 
 # ---------------------------------------------------------------------- ollama
 def ask_ollama(question: str, context: str, *, host: str, model: str,
                timeout: int) -> str | None:
-    """Free-form answer from the local model. None on any failure."""
     prompt = (
-        "You are a concise, READ-ONLY assistant for a momentum day-trading bot. "
-        "Answer the user's question using ONLY the data below. If the data does "
-        "not contain the answer, say so. Do not invent numbers. Keep it short.\n\n"
+        "You are the assistant for a small-cap momentum DAY-TRADING bot (opening-range "
+        "breakouts on $1-20 gappers). Answer the user's question using ONLY the DATA below, "
+        "which is the bot's live state. Be concrete and brief (a few sentences). Quote the "
+        "real numbers. If the data doesn't contain the answer, say so and tell the user what "
+        "you DO have (P&L, trades, the watchlist, ready signals, entry blocks, per-symbol "
+        "setup detail). Never invent tickers or numbers.\n\n"
         f"=== DATA ===\n{context}\n\n=== QUESTION ===\n{question}\n\n=== ANSWER ===\n")
     try:
         req = urllib.request.Request(
             f"{host}/api/generate",
             data=json.dumps({"model": model, "prompt": prompt, "stream": False,
-                             "options": {"temperature": 0.2, "num_predict": 400}}).encode(),
-            headers={"Content-Type": "application/json"},
-        )
+                             "options": {"temperature": 0.2, "num_predict": 500}}).encode(),
+            headers={"Content-Type": "application/json"})
         with urllib.request.urlopen(req, timeout=timeout) as resp:
-            body = json.loads(resp.read().decode())
-        text = (body.get("response") or "").strip()
+            text = (json.loads(resp.read().decode()).get("response") or "").strip()
         return text or None
     except Exception:  # noqa: BLE001
         logger.debug("ollama unavailable", exc_info=True)
@@ -97,18 +197,22 @@ def ask_ollama(question: str, context: str, *, host: str, model: str,
 
 def answer(store, question: str, *, host: str, model: str, timeout: int,
            ollama_enabled: bool = True) -> str:
-    """Build the read-only answer. Refuses action commands; falls back to the raw
-    snapshot when Ollama is down/disabled (never crashes)."""
-    if any(w in question.lower() for w in _ACTION_WORDS):
-        return ("I'm read-only — I can answer questions about the data but I "
-                "won't place or change trades. Ask me about P&L, positions, or "
-                "today's trades.")
+    if is_help(question):
+        return HELP_TEXT
+    if any(w in (" " + question.lower() + " ") for w in _ACTION_WORDS):
+        return ("I'm read-only — I answer questions about the data but won't place or "
+                "change trades. Try \"how did we do today?\" or send \"help\".")
     context = gather_context(store)
+    # focused per-symbol retrieval: tickers in the question that the bot actually knows
+    tokens = set(re.findall(r"\b[A-Z]{1,5}\b", question.upper())) - _STOPWORDS
+    known = known_symbols(store)
+    for sym in [t for t in tokens if t in known][:3]:
+        context += symbol_context(store, sym)
     if ollama_enabled:
         out = ask_ollama(question, context, host=host, model=model, timeout=timeout)
         if out:
             return out
-    return f"(LLM unavailable — raw snapshot)\n{context}"
+    return f"(LLM unavailable — here's the raw snapshot)\n{context}"
 
 
 # ------------------------------------------------------------------- transport
@@ -143,9 +247,8 @@ def main() -> None:
 
     from storage.event_store import EventStore
     store = EventStore("momentum")
-    logger.info("Q&A bot online (chat %s, ollama=%s %s) — ask me about the data",
-                allowed, ollama_enabled, model)
-    _send(token, allowed, "\U0001F916 Q&A bot online — ask me about P&L, positions, or today's trades.")
+    logger.info("Q&A bot online (chat %s, ollama=%s %s)", allowed, ollama_enabled, model)
+    _send(token, allowed, HELP_TEXT)
 
     offset = None
     while True: