Metrics

analyze.py

@@ -29,6 +29,7 @@ import math
 import sqlite3
 import logging
 import warnings
+from datetime import datetime
 from pathlib import Path
 
 # Silence transformer noise before importing
@@ -78,6 +79,37 @@ MIN_COVERAGE_SPREAD = 0.0   # disabled: signal_score naturally zeros out single-
 FINBERT_MIN_CONF    = 0.70  # drop neutral articles below this FinBERT confidence
 ALERT_THRESHOLD     = 0.35  # signal_score > this → alert
 
+# ---------------------------------------------------------------------------
+# Claude metrics
+# ---------------------------------------------------------------------------
+
+METRICS_FILE = Path(__file__).parent / "metrics.json"
+
+# Pricing: Claude Haiku 4.5 — https://www.anthropic.com/pricing
+_PRICE_INPUT_PER_TOKEN  = 0.80 / 1_000_000   # $0.80 per MTok
+_PRICE_OUTPUT_PER_TOKEN = 4.00 / 1_000_000   # $4.00 per MTok
+
+
+def calc_cost(input_tokens: int, output_tokens: int) -> float:
+    return round(input_tokens * _PRICE_INPUT_PER_TOKEN + output_tokens * _PRICE_OUTPUT_PER_TOKEN, 6)
+
+
+def update_metrics(input_tokens: int, output_tokens: int) -> None:
+    """Accumulate Claude token usage into metrics.json."""
+    cost = calc_cost(input_tokens, output_tokens)
+    data: dict = {}
+    if METRICS_FILE.exists():
+        try:
+            data = json.loads(METRICS_FILE.read_text())
+        except Exception:
+            pass
+    data["total_calls"]         = data.get("total_calls", 0) + 1
+    data["total_input_tokens"]  = data.get("total_input_tokens", 0) + input_tokens
+    data["total_output_tokens"] = data.get("total_output_tokens", 0) + output_tokens
+    data["total_cost_usd"]      = round(data.get("total_cost_usd", 0.0) + cost, 6)
+    data["last_updated"]        = datetime.now().isoformat(timespec="seconds")
+    METRICS_FILE.write_text(json.dumps(data, indent=2))
+
 
 # ---------------------------------------------------------------------------
 # Model loading
@@ -195,16 +227,16 @@ def coverage_spread_score(row) -> float:
 # Claude structured extraction
 # ---------------------------------------------------------------------------
 
-def claude_extract(title: str, text: str, tickers: list[str]) -> dict:
+def claude_extract(title: str, text: str, tickers: list[str]) -> tuple[dict, int, int]:
     """
     Use Claude Haiku to extract structured financial signal.
-    Returns {"confirmed_tickers", "magnitude", "timeframe", "reasoning"}.
+    Returns ({"confirmed_tickers", "magnitude", "timeframe", "reasoning"}, input_tokens, output_tokens).
     """
     import anthropic
 
     api_key = os.environ.get("ANTHROPIC_API_KEY")
     if not api_key:
-        return {"confirmed_tickers": tickers, "magnitude": 5, "timeframe": "days", "reasoning": "(no API key)"}
+        return {"confirmed_tickers": tickers, "magnitude": 5, "timeframe": "days", "reasoning": "(no API key)"}, 0, 0
 
     client = anthropic.Anthropic(api_key=api_key)
     ticker_ctx = "\n".join(
@@ -245,10 +277,10 @@ Fields:
         raw = msg.content[0].text.strip()
         raw = re.sub(r"^```(?:json)?\n?", "", raw)
         raw = re.sub(r"\n?```$", "", raw)
-        return json.loads(raw)
+        return json.loads(raw), msg.usage.input_tokens, msg.usage.output_tokens
     except Exception as e:
         print(f"  [warn] Claude failed: {e}")
-        return {"confirmed_tickers": tickers, "magnitude": 5, "timeframe": "days", "reasoning": str(e)[:120]}
+        return {"confirmed_tickers": tickers, "magnitude": 5, "timeframe": "days", "reasoning": str(e)[:120]}, 0, 0
 
 
 # ---------------------------------------------------------------------------
@@ -476,7 +508,9 @@ def analyze_articles(
         claude_data: dict = {}
         if use_claude and not dry_run and os.environ.get("ANTHROPIC_API_KEY"):
             print(f"  [claude] {slug[:50]}")
-            claude_data = claude_extract(title, full_text, list(matches.keys()))
+            claude_data, _in_tok, _out_tok = claude_extract(title, full_text, list(matches.keys()))
+            if _in_tok:
+                update_metrics(_in_tok, _out_tok)
 
         # ------------------------------------------------------------------
         # Phase 6 — yfinance momentum + scoring