#!/usr/bin/env python3
"""Local cohort coverage review. No employee score, ranking, network or inference."""
import argparse
import csv
import json
from datetime import date, timedelta


def review(rows, as_of, checkpoint_days):
    if checkpoint_days < 1:
        raise ValueError('checkpoint_days must be positive')
    seen = set()
    totals = dict(total=0, pending=0, mature=0, retained=0, early_exit=0,
                  retention_unknown=0, goal_observed=0, goal_met=0,
                  goal_not_met=0, manager_observed=0, both_observed=0)
    flags = []
    for row in rows:
        ident = row['episode_id'].strip()
        if not ident or ident in seen:
            raise ValueError('empty or duplicate episode_id')
        seen.add(ident)
        dates = {key: date.fromisoformat(row[key]) if row[key] else None
                 for key in ('start_on', 'known_employed_through', 'exit_on',
                             'goal_window_end', 'goal_reviewed_on',
                             'manager_window_end', 'manager_reviewed_on')}
        start = dates['start_on']
        if start is None or start > as_of:
            raise ValueError('every episode must have actually started by as_of')
        if any(value and (value < start or value > as_of) for value in dates.values()):
            raise ValueError('dates must fall between start and as_of')
        known, exited = dates['known_employed_through'], dates['exit_on']
        if known and exited:
            raise ValueError('use either continuous employment confirmation or known exit, not both')
        checkpoint = start + timedelta(days=checkpoint_days)
        mature = checkpoint <= as_of
        state = ('pending' if not mature else 'early_exit' if exited and exited <= checkpoint
                 else 'retained' if (exited and exited > checkpoint) or (known and known >= checkpoint)
                 else 'retention_unknown')
        goal = row['goal_result']
        if goal not in ('', 'met', 'not_met'):
            raise ValueError('goal_result must be empty, met or not_met')
        goal_present = bool(goal)
        if goal_present != bool(dates['goal_window_end']) or goal_present != bool(dates['goal_reviewed_on']):
            raise ValueError('goal result, window end and review date must be present together')
        manager = bool(dates['manager_reviewed_on'])
        if manager != bool(dates['manager_window_end']):
            raise ValueError('manager window and review date must be present together')
        for component, present in (('goal', goal_present), ('manager', manager)):
            if present and (state != 'retained' or dates[component + '_window_end'] != checkpoint
                            or dates[component + '_reviewed_on'] < checkpoint):
                raise ValueError('full-window observations require confirmed retention and the exact checkpoint window')
        totals['total'] += 1
        totals[state] += 1
        if mature:
            totals['mature'] += 1
        if state == 'retained':
            totals['goal_observed'] += goal_present
            totals['goal_met'] += goal == 'met'
            totals['goal_not_met'] += goal == 'not_met'
            totals['manager_observed'] += manager
            totals['both_observed'] += goal_present and manager
        flags.append(dict(episode_id=ident, checkpoint_on=checkpoint.isoformat(),
                          observation_state=state, goal_record_present=goal_present,
                          manager_record_present=manager))
    def fraction(n, d):
        return dict(numerator=n, denominator=d, percent=round(n / d * 100, 2) if d else None)
    mature, retained = totals['mature'], totals['retained']
    known_count = retained + totals['early_exit']
    return dict(as_of=as_of.isoformat(), checkpoint_days=checkpoint_days, counts=totals,
                retention_coverage=fraction(known_count, mature),
                retained_among_known=fraction(retained, known_count),
                retained_cohort_lower_bound=fraction(retained, mature),
                retained_cohort_upper_bound=fraction(retained + totals['retention_unknown'], mature),
                goal_coverage_among_confirmed_full_window=fraction(totals['goal_observed'], retained),
                manager_coverage_among_confirmed_full_window=fraction(totals['manager_observed'], retained),
                both_coverage_among_confirmed_full_window=fraction(totals['both_observed'], retained),
                combined_quality_score=None, episode_checks=flags)


if __name__ == '__main__':
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument('csv_path')
    parser.add_argument('--as-of', required=True, type=date.fromisoformat)
    parser.add_argument('--checkpoint-days', required=True, type=int)
    args = parser.parse_args()
    with open(args.csv_path, newline='', encoding='utf-8') as source:
        result = review(list(csv.DictReader(source)), args.as_of, args.checkpoint_days)
    print(json.dumps(result, indent=2))