?one line missing?
at each impulse separately when the 32-letter count has been finished.
In examining for a single impulse being wrong, we look at pairs
of letters which differ only in that impulse.
First examine the last impulse only; use the additional evidence
ot the pairs not used in that X, comparing the split on those pairs
with the corresponding splits on the average counts. (these counts are
available at Colossus). Particular note should be taken of splits in
the wrong direction or failures to achieve an expected large split.
These can often be explained. Suppose X3 is being exmIned
If there are a high number of 5's and many fewer 8's, we can expoect a
high number of 5's in Delta P to against Total Motor x's and these
will give rise to 9's in Delta D about twice as often as they will to /'s.
Consequently it is not a matter for surprise to flnd in such a count 9's
above /'s. In the same count there may be a similar, rather smaller
effect causing N to be as high as - or higher than - 3.
SimilarIy it the Delta D is rich in /, we will often find 8's
above 5's and, as a similar secondary effect, V may come above C and K
above J.
These exampIes apply to a scrutiny of the X3 setting.
similarly, if the X5 setting Is being investigated, in a message strong
in /'s it is not to be wondered at it if X comes up to F, or even Q to U.
A rule of thumb may be suggested but it is extremely crude :-
(a) If X3 is in question. Assume the following 16 lettes are expected
to be better than their X3 pairs - /,H,O,3,R,G,P,I,U,Q,5,J,F,X,Y,S.
Count in the in the supposed Delta D how many of these letters are better
than their pairs (not equal to). If the number is lower than 11, that is a fact
that requires explanation.
(In a sample run Bream, Gurnard, Stickleback, Jellyfish and Cod
of 50 messages completely set - (an unsatisfactory sample for statistics,
theoretically) - the numbers having from 8 to 16 of the letters going the