The mosaic measures the Moon before it lays the tiles

Written the day after two 43-panel runs each missed most of the half Moon. Every goto that night landed within two arcseconds. The tiles were in the wrong place because the grid was laid out around where the alignment thought the disc centre was. After an hour, a one-point model was several arcminutes out. That was more than the margin, and nothing checked it.

So the grid no longer trusts the alignment. It measures the Moon first, in the main camera, every run. None of this has been on the sky yet. It has only been run against a simulated Moon (astro-rig 1c01ab0 and 706f6d6).

The idea

My question was whether this could be dynamic rather than hard-coded maths: the Moon is in a different place and at a different distance every night. Measure a diameter on the best edges, measure another at 90 degrees to it, and that gives the crosshairs and a square to work in. Then lay the tiles out with the overlap.

What got built is that idea with more points. On a half Moon only one side has a real edge; the other side is the terminator, which is not a circle. So every point is on the sunlit limb, and three points on an arc are enough to fix a circle.

How a run starts now

1. Find the limb. From wherever the alignment says the centre is, walk out towards the bright limb until the camera sees a curved lit-to-black edge.
2. Learn the camera. Two small moves, and how the picture shifts with them, give the camera's rotation in the focuser and whether the image is mirrored. Nothing about how the camera sits is assumed.
3. Four limb points. The mount puts the limb exactly on the frame centre at four bearings around the bright limb: 18 and 55 degrees either side of the brightest point, well clear of the cusps. Each point is a mount position, so the limb is measured in the mount's own coordinates and not in pixels.
4. Fit the circle. That gives the true centre and the radius. The fit is refused if the points scatter by more than 0.4', or if the radius is more than 4% off the ephemeris.
5. Measure the plate scale. The moves and the edge shifts they produce give arcseconds a pixel. Focusing a Mak moves the primary mirror, so the scale changes a little from night to night. It has to be within 15% of the profile, or the run stops.
6. Lay the tiles square to the camera, not to alt/az, so neighbours overlap by exactly the chosen fraction on the sensor. The grid is sized from the measured diameter plus 4%. Every tile is checked against where the terminator is, and the dark ones are not shot.

Backlash was the hard part. There is about 1.5' of altitude slack on a reversal, and in the simulator a first version with that much slack put the centre 2' out and still passed its own check. Now every position is approached from the same side, overshooting by 2' and coming back on any axis that would otherwise reverse. The slack is then always taken up the same way and it cancels.

In rig-ui

The page draws the measured Moon: a dotted circle through the limb points, the sunlit side shaded, the four limb points marked, an arrow for "up", and every tile laid over it. Tiles that will be skipped are dashed, and each shot tile fills in with its thumbnail as it comes in. Rows and columns are no longer typed in; they come out of the measurement. The "fixed grid" box brings back the old rows-by-columns grid for a night when the measuring goes wrong.

The other two changes from 18/10

• A run off the Moon now stops itself. If three panels that should hold Moon come back black, and at least half of those checked so far are black, it keeps what it shot, goes back to the anchor and stops. On 18/10 that would have stopped the first run after about eleven panels instead of 43.
• Process can take several runs of one night and stitch them as one mosaic, which is how the half Moon was rescued by hand. It refuses runs from different nights and does not stack a run twice.

Does the camera's position or rotation matter?

• Rotation and mirroring: no, because they are measured at the start of every run. Just don't knock or reseat the camera mid-run.
• Position: only the main camera's view is used, so where the finder points doesn't matter once the Moon is somewhere in the Mak.
• Field rotation: the sky turns about 3 degrees in half an hour on an alt-az mount. The overlap and the stitcher's per-panel rotation absorb that.

To prove on the sky

• The limb walk on a real, ragged, mountainous limb, not a simulated one.
• Whether 2' of overshoot is really enough for the GTi's slack.
• The measured plate scale against the 0.193"/px the 127 profile assumes, which has never been measured.
• Exposure: about 40 ms, set on the bright limb with highlights near 240, as noted after 18/10.