v1.2 strategy upgrade: small-capital fee-aware signals, two-model pipeline, auto deep-research escalation

.env.example

@@ -5,17 +5,21 @@ ALPACA_BASE_URL=https://paper-api.alpaca.markets
 
 # Runtime
 PAPER_MODE=true
+STARTING_CAPITAL_USD=100
 MAX_ORDER_USD=5
-MAX_DAILY_NOTIONAL=50
+MAX_DAILY_NOTIONAL=40
-MAX_OPEN_POSITIONS=6
+MAX_OPEN_POSITIONS=8
-MIN_CONFIDENCE=0.60
+MIN_CONFIDENCE=0.62
+FEE_PER_TRADE_USD=0.00
+SLIPPAGE_BPS=5
 TRADE_INTERVAL_HOURS=2
 CURATE_INTERVAL_MINUTES=30
 TIMEZONE=America/Los_Angeles
 
 # Ollama
 OLLAMA_URL=http://10.30.20.110:11434
-OLLAMA_MODEL=gemma3:latest
+OLLAMA_CURATOR_MODEL=gemma3:latest
+OLLAMA_DECISION_MODEL=agent-oss:latest
 
 # Data sources
 SEARX_URL=http://10.30.20.35:6969/search

bot.py

@@ -4,7 +4,16 @@ import json
 from sqlalchemy import func
 from config import settings
 from db import SessionLocal, BotDecision, TradeExecution, CuratedInsight
-from services import searx_news, ollama_decide, place_order, market_open, positions_snapshot, summarize_news_with_ollama
+from services import (
+    searx_news,
+    extra_research,
+    summarize_news_with_ollama,
+    strategy_signals,
+    llm_final_decision,
+    place_order,
+    market_open,
+    positions_snapshot,
+)
 
 scheduler = BackgroundScheduler(timezone=settings.timezone)
 
@@ -45,13 +54,21 @@ def run_cycle():
                 break
 
             news = searx_news(symbol)
-            decision = ollama_decide(symbol, news)
+            strat = strategy_signals(symbol, news)
+            decision = llm_final_decision(symbol, news, strat)
+
+            # Escalate to deeper research when model asks or confidence weak
+            if decision.get("needs_more_research") or decision.get("confidence", 0) < settings.min_confidence:
+                more = extra_research(symbol, decision.get("research_topics", []))
+                if more:
+                    news = news + more
+                    decision = llm_final_decision(symbol, news, strat)
 
             drow = BotDecision(
                 symbol=symbol,
                 action=decision["action"],
                 confidence=decision["confidence"],
-                reason=decision["reason"],
+                reason=f"{decision['reason']} | strat={strat['strategy']} score={strat['score']}",
                 market_context=json.dumps(news)[:60000],
                 order_usd=decision["order_usd"],
                 status="planned",
@@ -66,8 +83,14 @@ def run_cycle():
             )
 
             if should_trade:
-                notional = min(settings.max_order_usd, decision["order_usd"], settings.max_daily_notional - spent)
+                # Fee/slippage-aware cap for tiny bankroll
-                if notional <= 0:
+                effective_cost = settings.fee_per_trade_usd + (settings.slippage_bps / 10000.0) * decision["order_usd"]
+                notional = min(
+                    settings.max_order_usd,
+                    decision["order_usd"],
+                    settings.max_daily_notional - spent,
+                )
+                if notional <= effective_cost:
                     drow.status = "risk_blocked"
                     db.add(drow)
                     db.commit()
@@ -83,7 +106,7 @@ def run_cycle():
                     qty=float((res or {}).get("json", {}).get("qty", 0) or 0),
                     notional=notional,
                     alpaca_order_id=(res or {}).get("json", {}).get("id", ""),
-                    raw=json.dumps(res)[:60000],
+                    raw=json.dumps({"decision": decision, "strategy": strat, "broker": res})[:60000],
                 ))
                 db.commit()
                 if ok:

config.py

@@ -9,21 +9,32 @@ class Settings:
     alpaca_base = os.getenv("ALPACA_BASE_URL", "https://paper-api.alpaca.markets")
     paper_mode = os.getenv("PAPER_MODE", "true").lower() == "true"
 
+    # Capital/risk profile
+    starting_capital_usd = float(os.getenv("STARTING_CAPITAL_USD", "100"))
     max_order_usd = float(os.getenv("MAX_ORDER_USD", "5"))
-    max_daily_notional = float(os.getenv("MAX_DAILY_NOTIONAL", "50"))
+    max_daily_notional = float(os.getenv("MAX_DAILY_NOTIONAL", "40"))
-    max_open_positions = int(os.getenv("MAX_OPEN_POSITIONS", "6"))
+    max_open_positions = int(os.getenv("MAX_OPEN_POSITIONS", "8"))
     min_confidence = float(os.getenv("MIN_CONFIDENCE", "0.60"))
 
+    # Approx fee model for small-size optimization
+    fee_per_trade_usd = float(os.getenv("FEE_PER_TRADE_USD", "0.00"))
+    slippage_bps = float(os.getenv("SLIPPAGE_BPS", "5"))
+
+    # Scheduling
     trade_interval_hours = int(os.getenv("TRADE_INTERVAL_HOURS", "2"))
     curate_interval_minutes = int(os.getenv("CURATE_INTERVAL_MINUTES", "30"))
     timezone = os.getenv("TIMEZONE", "America/Los_Angeles")
 
+    # LLM stack (small for curation, larger for final decision)
     ollama_url = os.getenv("OLLAMA_URL", "http://10.30.20.110:11434")
-    ollama_model = os.getenv("OLLAMA_MODEL", "gemma3:latest")
+    ollama_curator_model = os.getenv("OLLAMA_CURATOR_MODEL", "gemma3:latest")
+    ollama_decision_model = os.getenv("OLLAMA_DECISION_MODEL", "agent-oss:latest")
 
+    # Data sources
     searx_url = os.getenv("SEARX_URL", "http://10.30.20.35:6969/search")
     scraper_api = os.getenv("SCRAPER_API_URL", "http://10.30.20.115:24125")
 
+    # App
     db_path = os.getenv("DB_PATH", "sqlite:///./bot.db")
     host = os.getenv("APP_HOST", "0.0.0.0")
     port = int(os.getenv("APP_PORT", "8089"))

services.py

@@ -4,7 +4,20 @@ import requests
 from config import settings
 
 
-def searx_news(symbol: str, limit: int = 10):
+def _ollama_generate(model: str, payload_obj: dict, timeout: int = 45):
+    payload = {
+        "model": model,
+        "stream": False,
+        "prompt": json.dumps(payload_obj),
+        "format": "json",
+    }
+    r = requests.post(f"{settings.ollama_url}/api/generate", json=payload, timeout=timeout)
+    r.raise_for_status()
+    resp = r.json().get("response", "{}")
+    return json.loads(resp)
+
+
+def searx_news(symbol: str, limit: int = 12):
     q = f"{symbol} stock news earnings guidance analyst macro risk"
     params = {"q": q, "format": "json", "language": "en"}
     try:
@@ -13,75 +26,129 @@ def searx_news(symbol: str, limit: int = 10):
         data = r.json()
         out = []
         for it in data.get("results", [])[:limit]:
-            out.append({"title": it.get("title", ""), "url": it.get("url", ""), "content": it.get("content", "")[:500]})
+            out.append({
+                "title": it.get("title", ""),
+                "url": it.get("url", ""),
+                "content": (it.get("content", "") or "")[:700],
+            })
+        return out
+    except Exception:
+        return []
+
+
+def extra_research(symbol: str, weak_points: list, limit: int = 6):
+    """Second-pass targeted research when confidence/coverage is weak."""
+    q = f"{symbol} {' '.join(weak_points[:3])} SEC filing guidance risks competition"
+    params = {"q": q, "format": "json", "language": "en"}
+    try:
+        r = requests.get(settings.searx_url, params=params, timeout=20)
+        r.raise_for_status()
+        data = r.json()
+        out = []
+        for it in data.get("results", [])[:limit]:
+            out.append({
+                "title": it.get("title", ""),
+                "url": it.get("url", ""),
+                "content": (it.get("content", "") or "")[:700],
+            })
         return out
     except Exception:
         return []
 
 
 def summarize_news_with_ollama(symbol: str, context_items: list):
-    payload = {
+    prompt = {
-        "model": settings.ollama_model,
-        "stream": False,
-        "prompt": json.dumps({
         "task": "Summarize market-moving info into a concise, neutral brief.",
         "symbol": symbol,
         "news": context_items,
-            "format": {"summary": "<=140 words", "bullish_points": ["..."], "bearish_points": ["..."]}
+        "format": {
-        }),
+            "summary": "<=140 words",
-        "format": "json",
+            "bullish_points": ["..."],
+            "bearish_points": ["..."],
+            "uncertainties": ["..."]
+        }
     }
     try:
-        r = requests.post(f"{settings.ollama_url}/api/generate", json=payload, timeout=45)
+        parsed = _ollama_generate(settings.ollama_curator_model, prompt)
-        r.raise_for_status()
-        resp = r.json().get("response", "{}")
-        parsed = json.loads(resp)
         return parsed.get("summary", "no-summary")
     except Exception:
         return "summary-unavailable"
 
 
-def ollama_decide(symbol: str, context_items: list):
+def strategy_signals(symbol: str, context_items: list):
+    """Proven-ish small-capital rules encoded as interpretable signals."""
+    text_blob = " ".join((x.get("title", "") + " " + x.get("content", "")) for x in context_items).lower()
+    bullish = sum(k in text_blob for k in ["beat", "raise guidance", "upgrade", "buyback", "record revenue"])
+    bearish = sum(k in text_blob for k in ["miss", "downgrade", "lawsuit", "probe", "cut guidance", "recall"])
+
+    # simple event momentum score
+    score = bullish - bearish
+    # conservative policy for tiny capital: only trade stronger score edges
+    if score >= 2:
+        action = "buy"
+    elif score <= -2:
+        action = "sell"
+    else:
+        action = "hold"
+
+    conf = min(0.85, 0.50 + abs(score) * 0.08)
+    return {
+        "strategy": "event-momentum-v1",
+        "score": score,
+        "action": action,
+        "confidence": conf,
+        "signals": {"bullish": bullish, "bearish": bearish},
+    }
+
+
+def llm_final_decision(symbol: str, context_items: list, strategy: dict):
     prompt = {
-        "task": "You are a conservative autonomous trading policy engine. Return strict JSON only.",
+        "task": "Final trading decision using all context and a conservative small-capital profile. Return strict JSON.",
+        "symbol": symbol,
         "constraints": {
             "actions": ["buy", "sell", "hold"],
             "max_order_usd": settings.max_order_usd,
-            "risk": "do not overtrade; prefer hold on weak signal",
+            "min_expected_edge_after_fees": "positive",
+            "fee_per_trade_usd": settings.fee_per_trade_usd,
+            "slippage_bps": settings.slippage_bps,
+            "avoid_overtrading": True,
         },
-        "symbol": symbol,
+        "strategy_prior": strategy,
         "news": context_items,
         "output_schema": {
             "action": "buy|sell|hold",
             "confidence": "0-1",
-            "reason": "short rationale",
             "order_usd": f"<= {settings.max_order_usd}",
+            "reason": "short rationale",
+            "needs_more_research": True,
+            "research_topics": ["..."]
         },
     }
-    payload = {
+
-        "model": settings.ollama_model,
-        "prompt": json.dumps(prompt),
-        "stream": False,
-        "format": "json",
-    }
     try:
-        r = requests.post(f"{settings.ollama_url}/api/generate", json=payload, timeout=45)
+        d = _ollama_generate(settings.ollama_decision_model, prompt, timeout=60)
-        r.raise_for_status()
-        resp = r.json().get("response", "{}")
-        d = json.loads(resp)
         action = str(d.get("action", "hold")).lower()
         if action not in {"buy", "sell", "hold"}:
             action = "hold"
         confidence = max(0.0, min(1.0, float(d.get("confidence", 0.5))))
         order_usd = min(float(d.get("order_usd", settings.max_order_usd)), settings.max_order_usd)
         reason = d.get("reason", "fallback")
-        return {"action": action, "confidence": confidence, "order_usd": order_usd, "reason": reason}
+        return {
+            "action": action,
+            "confidence": confidence,
+            "order_usd": order_usd,
+            "reason": reason,
+            "needs_more_research": bool(d.get("needs_more_research", False)),
+            "research_topics": d.get("research_topics", []) or [],
+        }
     except Exception:
         return {
-            "action": random.choice(["hold", "hold", "buy"]),
+            "action": strategy.get("action", "hold"),
-            "confidence": 0.3,
+            "confidence": min(strategy.get("confidence", 0.5), 0.55),
             "order_usd": min(1.0, settings.max_order_usd),
-            "reason": "fallback-mode",
+            "reason": "decision-fallback-strategy",
+            "needs_more_research": False,
+            "research_topics": [],
         }