# -*- coding: utf-8 -*-
"""Extract the windows-agent longitudinal record: 7 versions x 30 tasks.

Reads only. Writes a tidy CSV to the scratchpad for analysis.
"""
import csv
import io
import json
import os
import re

SRC = r'C:\xampp\htdocs\windows-agent\bench\results'
OUT = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'wa_tasks.csv')

runs = []
for fn in sorted(os.listdir(SRC)):
    if not fn.endswith('.json'):
        continue
    d = json.load(io.open(os.path.join(SRC, fn), encoding='utf-8'))
    m = re.match(r'(\d{4}-\d{2}-\d{2})_(\d{6})_(.+)\.json', fn)
    label = m.group(3) if m else fn
    vm = re.search(r'v(\d+)', label)
    runs.append(dict(file=fn, date=m.group(1) if m else '', label=label,
                     version=int(vm.group(1)) if vm else None,
                     model=d.get('model'), overall=d.get('overall'),
                     cost=d.get('estimated_cost_usd'), elapsed=d.get('elapsed_seconds'),
                     finished=d.get('finished_at'), results=d.get('results') or []))

runs.sort(key=lambda r: (r['version'] if r['version'] is not None else 999))

print('=' * 100)
print('RUN-LEVEL RECORD')
print('=' * 100)
print('  %-26s %-16s %6s %8s %8s %6s %6s' %
      ('label', 'model', 'score', 'cost $', 'minutes', 'tasks', 'errors'))
print('  ' + '-' * 96)
for r in runs:
    errs = sum(1 for t in r['results'] if (t.get('error') or '').strip())
    print('  %-26s %-16s %6.1f %8.4f %8.1f %6d %6d'
          % (r['label'], r['model'], r['overall'], r['cost'], r['elapsed'] / 60.0,
             len(r['results']), errs))

# tidy csv
cols = ['version', 'label', 'date', 'model', 'run_overall', 'run_cost_usd', 'run_minutes',
        'task_id', 'task_name', 'category', 'earned', 'possible', 'percent', 'penalty',
        'tool_calls', 'n_tools', 'tools', 'duration_s', 'input_tokens', 'output_tokens',
        'error', 'n_checks', 'checks_failed']
w = io.open(OUT, 'w', encoding='utf-8', newline='')
cw = csv.DictWriter(w, fieldnames=cols)
cw.writeheader()
for r in runs:
    for t in r['results']:
        checks = t.get('checks') or []
        cw.writerow({
            'version': r['version'], 'label': r['label'], 'date': r['date'],
            'model': r['model'], 'run_overall': r['overall'], 'run_cost_usd': r['cost'],
            'run_minutes': round(r['elapsed'] / 60.0, 2),
            'task_id': t.get('id'), 'task_name': t.get('name'), 'category': t.get('category'),
            'earned': t.get('earned'), 'possible': t.get('possible'),
            'percent': t.get('percent'), 'penalty': t.get('penalty'),
            'tool_calls': t.get('tool_calls'), 'n_tools': len(t.get('tools') or []),
            'tools': '|'.join(t.get('tools') or []),
            'duration_s': t.get('duration'),
            'input_tokens': t.get('input_tokens'), 'output_tokens': t.get('output_tokens'),
            'error': (t.get('error') or '').replace('\n', ' ')[:300],
            'n_checks': len(checks),
            'checks_failed': sum(1 for c in checks if not c.get('passed')),
        })
w.close()
print('\n  wrote %s (%d rows)' % (OUT, sum(len(r['results']) for r in runs)))

# categories
cats = {}
for t in runs[-1]['results']:
    cats[t['category']] = cats.get(t['category'], 0) + 1
print('  categories:', cats)