Up to ten targets, one night: each gets its good/marginal/poor windows (darkness × altitude × Moon × clouds, as on the single-target page), then the night is handed out five minutes at a time to whichever target is best — with hysteresis, so you get a few solid blocks of at least 30 minutes, not a flicker. Targets that never win a block are listed with the reason.
Try: an autumn deep-sky night · ← previous night · next night →
Night of Friday, 04 September 2026 · astronomical night 21:26 → 05:13 (7.78 h) · Moon 43% (waning crescent) · weather: ERA5 history
Faded bands: each target's own windows (green good, amber marginal, pink poor). Dark blue: the schedule. Grey columns: twilight.
| local | target | hours | mean quality |
|---|---|---|---|
| 19:45–22:05 | — nothing worth pointing at — | 2.33 | |
| 22:05–05:50 | M33 | 7.75 | 0.72 |
| 05:50–06:55 | — nothing worth pointing at — | 1.08 |
| target | astronomy | overall | best window | highest | Moon sep. | scheduled h |
|---|---|---|---|---|---|---|
| M33 Triangulum Pinwheel | 86 | 85 | 22:45–05:15 (good) | 79.1° at 04:00 | 48.1° | 7.75 |
curl -sS "https://devops.majbase.com/rozhen-planner/plan-api?targets=M31,M42,M45,NGC%207000&date=2026-09-04"targets: names or RA Dec, separated by commas, semicolons or newlines, at most ten. The response carries targets[] (score, verdict, best_window, windows, transit, moon_separation), schedule[] (blocks with target, start/end in UTC and local, hours, mean_quality; target: null is idle time), unscheduled[] with a why, forecast_source, warnings[]. Missing targets → 400 {"error", "code": "targets_missing"}; an unreadable date → 400 {"error", "code": "invalid_date"}.