Autoguiding is a second, small camera watching one star and nudging the mount a few times a second so that star — and therefore your frame — stays put. The report card is RMS: the root-mean-square of the corrections, in arcseconds, usually split into RA and Dec plus a combined total. It measures how much your stars are being smeared by tracking error, on average, while you image.
The number that matters isn't absolute — it's relative to your image scale. The working rule: total RMS under about half your image scale means guiding is not your limiting factor. At 2″/pixel (a typical short refractor + CMOS), 1.0″ RMS is fine. At 0.7″/pixel on a long focal length, the same 1.0″ RMS visibly bloats stars, and you need 0.35–0.5″ — which is genuinely hard and depends on the night's seeing as much as your mount.
The classic problems and their signatures
Chasing seeing: guide exposures that are too short (under ~1 s) make the mount chase atmospheric shimmer it can never catch, and RMS gets WORSE. Guide at 2–4 s so the star's position averages over the turbulence; let aggressiveness settle around defaults.
Dec backlash: corrections in declination reverse direction through a gear's dead zone; the signature is Dec overshooting and oscillating after each reversal, or lagging for seconds after a correction. Guiding assistants measure it; the fixes are mechanical adjustment, backlash compensation, or guiding with Dec corrections in one direction only.
Differential flexure: round stars in the guide camera, trailed stars in the main camera — the two optical systems are flexing relative to each other. The fix is stiffness (shorter/better guide-scope mounting) or an off-axis guider, which samples the guide star through the imaging telescope itself and makes the problem geometrically impossible.
Cable snags: a perfectly good RMS graph with occasional violent spikes at similar mount positions. Follow the cables, not the software settings.
- Calibrate near the intersection of the celestial equator and meridian; recalibrate after a meridian flip only if your software doesn't flip its own calibration.
- One bad night is weather. Judge changes on a week of data, not one session.