1. Enclosed are Mr. Newman's comments on the Fish information
contained in the Monthly Information Letters #2 and Annexes B and F.
Although I don't feel competent to enter into the mathematical
phases of the discussion my feeling about the Jacobs method is
that it confuses the eventual goal of a rectangle with a high
overall score with the means to that goal which is matching
of rows and columns. It seems to me that the Jacobs test is
"Which pair of rows (or columns) if matched (or mismatched) will
contribute most heavily to the final score" rather than "Which
pair were the most likely match (or mismatch)". Since combining
of rows and columns results from the data produced by the test
it would seem that the second question is the proper one. It
may be demonstrable that the answers to the two questions are
identical but if this is so how can the accurate method be theoret-
ically fallacious? At any rate it yields probabilities which are
essential when several messages are being worked on. A possible
advantge of the Jacobs method is that it involves no assumption
as to the aprioi probability - but the same argument could be
used in favour of matching by cross-producting excesses. Mr.
Newman advises that some time ago they considered a method similar
to the IBM method but using only 6 rows at a time. However
their calculations showed that it wasn't sharp enough and this
is the occasion for the question in paragraph 4 of his notes.
In the IBM method why do you use blocks of 9 instead of 8? Pre-
sumably this sharpens the resuts but doesn't it increase the
dange of error if the rows which are used twice happen to be
abnormal? And, of course, it increases substantially the number
of tests required. The people here would like to know the time
factors involved in the IBM procedure.
2. I mentioned in Report #F-50 that the Jellyfish patterns
changed on 14 June. The wheels were broken on traffic of 18 June
(through a crib from Bream) but only messages of that day could
be set. This led to the suspicion that the wheel patterns were
changing daily but since Mu37 only had 20 dots it was by no means
certain. However 24 June has now been solved (also through a
Bream crib) and the patterns are ditterent. This time Mu37 had
28 dots and messages as short as 1100 letters of 24 June have
been set. No other day can be set on these patterns so it is
reasonably certain that they are now changing daily. As a result,
principal reliance is now being placed on attempted solution
through cribs although rectangle analysis on this link has not
been abandoned.