--- a/RevisionLog.txt
+++ b/RevisionLog.txt
@@ -299,4 +299,25 @@
 BTW ransac does prune some bad points, but not enough, it
 needs attention too.
 
-BTW +27 July  Let's get on with pairwise matching...
+
+Set up an image pairs list first, use layout if available, else
+list all pairs. Sort the list on 1st member to minimize the no.
+of k-d tree builds.  Match thru k-d tree giving a CP list for
+each pair.
+
+Do a 'rotational alignment' on each pair to cull out wild CPs.  
+This uses the kp coords mapped to sphere.  Rotate one image around 
+Z to minimize the variance of the difference vectors across each CP 
+-- 3d variance is best because all good CPs will be parallel after 
+alignment.  But the rotation is 2d of course.  Use L-M unless an
+analytic solution presents itself (surely one exists).  Throw out
+obvious outlier CPs and repeat, until there are no obvious outliers.
+
+Using the pair alignment just computed, one could then refine kp
+positions by remapping a patch around each CP from original images
+to panospace and doing correlation (shifts only).  Woth a try.
+
+Finally, review the ransac procedure to see if it could be made to
+use pairwise and/or global alignments effectively.
+