Alignment failed on half my subs. Now what?

Evaluate alignment leaves every frame in one of five states: Accepted, Aligned, Out of band, Won’t align, or Not yet screened. When a run comes back with a pile of the last three, work through the causes below in order. Most of the time the fit is telling you something specific, and looking at it in the Alignment Viewer is the fastest way to find out what.

Is the reference itself the problem?

Every frame is fitted against one reference, so a weak reference drags every fit down with it: low star count, poor focus, an odd crop, or a rotation far from the rest of your set. If most of a filter is failing while a handful of frames near the reference’s own framing succeed, suspect the reference before anything else.

Open the reference frame and judge it on its own merits, then pick a better one from the viewer or from Cull by metric. See what is the alignment reference and how do I change it.

A reference can also be a rejected frame. Its stars still drive every fit even though it never enters a stack, and the target page flags this directly with a re-anchor link. See if your reference gets rejected.

Is the star detection stale?

A fit works from each frame’s star catalog. If that catalog was built before you assigned or changed calibration, it describes the frame as it was captured rather than as it calibrates, and re-running the fit reuses the same stale catalog.

Use Force re-detect + evaluate under Advanced on the Alignment panel. It re-detects every frame on its currently calibrated pixels and re-fits from there. Reach for this right after assigning masters, or whenever you suspect a detection ran before calibration was in place. Accepted fits are kept.

If you’re not sure whether calibration or the matcher is at fault, Re-detect (no calibration) gives you a raw-pixel baseline to compare star counts against. A calibrated run that finds far fewer stars than the raw one points at the calibration, not the fit.

Is the rig tolerance too tight for a real field rotation?

A target shot across many nights accumulates field rotation, and a target near the pole accumulates it fast. If your rotation tolerance is narrower than what actually happened between sessions, otherwise good fits land as Out of band instead of Aligned.

Widen Rotation ± deg or Scale ± % on the Equipment page, under Alignment tolerances, for that rig. The shipped defaults are 5.0° of rotation and about 2% of scale. If your subs span a meridian flip, leave Allow 180° flip on so a flip fits normally instead of reading as a mismatch.

Screenshot: Equipment page, the alignment tolerances for a rig with a widened rotation window.

A tolerance change applies to fits made after it. Re-evaluate the filter to reclassify the existing frames under the new band.

Is it a genuinely hard field?

Some fields are harder to match regardless of tuning:

Very sparse fields. Too few stars leaves the matcher without enough points to lock a confident geometry.
Very dense fields. A crowded field, or a wide field of view, gives the matcher many plausible-looking correspondences to choose between, and it can settle on the wrong one.
Cross-filter frames. Narrowband filters produce a different, usually sparser, star field than broadband ones. A filter with markedly fewer detected stars than the rest of the target is more likely to struggle, since the reference framing was set from whichever filter won the original score, not necessarily this one.
Large rotations between frames. The bigger the angular gap between a frame and the reference, the more the star pattern has to be recognized under distortion, and the more likely the fit fails outright rather than landing out of band.

Re-running the same evaluation on these produces the same answer. Judge the frame in the viewer instead: where the star field genuinely doesn’t support a fit, move on.

What Out of band is actually telling you

Out of band describes the geometry, not the quality of the match. The matcher found a self-consistent rotation, scale and translation, and that geometry falls outside what this rig is configured to expect. The fit and the tolerance are two separate things, and either one can be the one that’s off.

If you know the rotation is real, for example a target you deliberately picked up again on a later night, the fit is correct and the tolerance is what needs adjusting. If the rotation looks wrong for what you actually shot, the fit itself is suspect even though it is internally consistent, and it’s worth opening in the viewer before trusting it.

When to accept manually, and when to reject

Open a flagged frame in the Alignment Viewer and flip between it and the reference to judge the framing directly.

Accept it when the geometry looks right to your eye, whether or not it falls inside the tolerance band. Accepting locks the fit: it’s excluded from any future automatic re-fit, and it holds until you reset it. An Out of band fit asks for confirmation before it locks, since you’re knowingly overriding the tolerance; an Aligned fit accepts straight away.

Reject the frame instead when the framing is actually wrong, when the field is too sparse or too crowded to trust, or when the frame shouldn’t be in the stack for reasons that have nothing to do with alignment, like trailing or cloud. A rejected frame drops out of the stack entirely and stops asking for attention.

Both actions are yours to take. The Needs attention panel exists so you look at each flagged frame before deciding.