"""Hermes CI Cost Tracker — 每次eval后记录token消耗"""
import json, os, csv, datetime

RESULTS_FILE = os.path.join(os.path.dirname(__file__), 'results.json')
COST_LOG = os.path.join(os.path.dirname(__file__), 'cost-history.csv')

# DeepSeek pricing (per 1M tokens)
INPUT_COST_PER_M = 0.07
OUTPUT_COST_PER_M = 0.28

with open(RESULTS_FILE) as f:
    data = json.load(f)

# results is a nested dict with 'results' key inside
results_inner = data.get('results', {}).get('results', [])
if not isinstance(results_inner, list):
    results_inner = data.get('results', {}).get('stats', {})

stats = data.get('results', {}).get('stats', {})
total_tests = stats.get('successes', 0) + stats.get('failures', 0) + stats.get('errors', 0)
passed = stats.get('successes', 0)
failed = stats.get('failures', 0)

# Token counts from stats.tokenUsage
token_usage = stats.get('tokenUsage', {})
total_input = token_usage.get('prompt', 0)
total_output = token_usage.get('completion', 0)
total_tokens = token_usage.get('total', total_input + total_output)
duration = int(stats.get('durationMs', 0) / 1000)
cost = (total_input * INPUT_COST_PER_M + total_output * OUTPUT_COST_PER_M) / 1_000_000

# Append to CSV
is_new = not os.path.exists(COST_LOG)
with open(COST_LOG, 'a', newline='') as f:
    w = csv.writer(f)
    if is_new:
        w.writerow(['timestamp', 'tests', 'passed', 'failed', 'tokens', 'cost_usd', 'duration_s'])
    w.writerow([
        datetime.datetime.utcnow().strftime('%Y-%m-%d %H:%M UTC'),
        total_tests, passed, failed, total_tokens, f'{cost:.4f}', duration
    ])

print(f'Tests: {total_tests} | Passed: {passed} | Failed: {failed}')
print(f'Tokens: {total_tokens} | Cost: ${cost:.4f} | Duration: {duration}s')
print(f'Log: {COST_LOG}')