"""Check the PDF and figure_changes against the saved test results (every key number re-derived from the source files)."""
import subprocess, pickle, json
import pandas as pd

PDF = '/mnt/user-data/outputs/Binzagr_Data/06_Model_Review/Binzagr_Model_Review_Before_After.pdf'
W = '/home/claude/binzagr/boost/work'; D = '/mnt/user-data/outputs/Binzagr_Data/06_Model_Review/data'
txt = subprocess.run(['pdftotext', '-layout', PDF, '-'], capture_output=True, text=True).stdout
flat = ' '.join(txt.split())
fc = pd.read_csv(f'{W}/fc_summary.csv').set_index('level')
CT = pickle.load(open(f'{W}/churn_race.pkl', 'rb'))['table'].set_index('model'); ch = pickle.load(open(f'{W}/churn_race.pkl', 'rb'))
AT = pickle.load(open(f'{W}/adoption_race.pkl', 'rb'))['table'].set_index('model'); ad = pickle.load(open(f'{W}/adoption_race.pkl', 'rb'))
rf = pickle.load(open(f'{W}/reco_final_v2.pkl', 'rb')); RES = rf['res']; N = pickle.load(open(f'{W}/report_numbers.pkl', 'rb'))
fails = []
def need(s, why):
    ok = s in flat
    print(('PASS' if ok else 'FAIL'), f'{s!r:28s}', why)
    if not ok: fails.append(s)
p = lambda x: f'{100 * x:.1f}%'
need(p(fc.loc['sku', 'picked_at_site_bias']), 'SKU after, same bias')
need(p(fc.loc['sku', 'picked_wape']), 'SKU as fitted')
need(p(fc.loc['van', 'picked_at_site_bias']), 'van after, same bias')
need(p(fc.loc['principal', 'picked_at_site_bias']), 'principal after')
need(p(fc.loc['cluster', 'picked_at_site_bias']), 'cluster after')
need(p(fc.loc['sku', 'replica_site_style']), 'our copy of the per-SKU race')
need(p(fc.loc['sku', 'naive7']), 'our SKU repeat-last-week')
need(p(fc.loc['sku', 'wd4']), 'our SKU 4-week average')
need('74.1%', 'SKU blend at zero bias (fc_eval check)')
need(f'{CT.loc["Extra trees", "test_auc"]:.3f}', 'churn after AUC')
need(f'{CT.loc["LightGBM", "test_auc"]:.3f}', 'churn LightGBM AUC')
need(f'+{ch["ci"]["Extra trees"][0]:.3f} to +{ch["ci"]["Extra trees"][2]:.3f}', 'churn 90% interval')
need(f'{100 * CT.loc["Extra trees", "test_top20_capture"]:.0f}%', 'churn top-20 after')
need(p(RES['site_style']['test_p10']), 'NBS our copy of site scorer')
need(p(RES[rf['pick']]['test_p10']), 'NBS after (picked model)')
need(f'+{100 * RES[rf["pick"]]["gain_90pct_interval"][0]:.1f} to +{100 * RES[rf["pick"]]["gain_90pct_interval"][1]:.1f}', 'NBS 90% interval')
need(f'{AT.loc["LightGBM", "test_auc_within_mean"]:.2f} to {AT.loc[ad["pick"], "test_auc_within_mean"]:.2f}', 'adoption within-cluster before to after')
need('SAR 5,060,755', 'hold-out sales match')
need('4,097', 'NBS outlets judged')
need(f'SAR {(3454548 + N["lever"]["total_change"]) / 1e6:.3f}M', 'value total after')
need('Confidential - contains Binzagr data', 'footer')
for t_ in ['3.0× · Model picks 12%', '0.77 · Churn model 0.77', '38.6% vs 47.2% weekly error', 'SAR 927,163 past due (24%)', '−8% error', '4.0× the average rate']:
    need(t_, 'replacement table text')
F = pd.read_csv(f'{D}/before_after_figures.csv')
print('before_after rows', len(F), '| numbers unique', F.no.is_unique, '| types', F.type.value_counts().to_dict())
J = json.load(open(f'{D}/before_after_figures.json')); print('json rows', len(J['rows']))
for f_ in ['next_best_sku_picks_v2.csv', 'churn_scores_v2.csv']:
    d = pd.read_csv(f'{D}/{f_}'); print(f_, len(d), 'rows', d.columns.tolist())
pk = pd.read_csv(f'{D}/next_best_sku_picks_v2.csv', dtype={'outlet_id': str, 'sku': str})
print('picks: ranks 1..10 per outlet', bool(pk.groupby('outlet_id')['rank'].apply(lambda r: sorted(r) == list(range(1, len(r) + 1))).all()))
print('\nFAILED:', fails if fails else 'none')