The timestamp parse was written out six times, the distinct-production-hours count four times, and the per-day summary four times (today, tomorrow, Day+2, Day+3). They are now `_solar_timestamp_to_naive` (module level), a `production_hours` and `utilization` pair, and `day_part`, which covers all four days since the first-day %util and the peak can only fall on today or tomorrow. The two first-day hour counts keep their different rules (today: distinct hours at 0.1 W; tomorrow: data points at 1% of the panel), and the dead `now.strftime(...)` statement goes. 418 → 155 lines. The new goldens are unchanged, and old and new agree on 20,000 generated forecasts (timestamp formats, timezones, NaN and zero panels, None powers, threshold-boundary values) including every translate and logger call in order; two threshold mutations were caught.
agessaman
2dbd425dcb