?possible line missing?
higher than the correct position itself. This means that these are
serious rival hypotheses - such as that X134 are correct and X2 is
a good slide of the correct position. This is a type of hypothesis worth
bearing in mind if, after four X's have been set with apparent
satisfaction, the last X is obstinate.
The hypothesis can be checked by doing short runs for each
separately on letters determined for the purpose from their strength in
the 16 - letter count.
Short Runs. On a short run the random scores are liable to be as high as
2 sigma and we are not accustomed to take much interest in an unsupported
score of under 2.5 sigma. In average circumstances such a score is liable
to make the corresponding setting about evens. Anything above 3 sigma is
worth taking pretty seriously : 4 sigma must on no account be ignored :
(it may be only a good slide or even a good anti-slide, but the setting
eventually selected must account for the occurrance of such a score).
The deciban chart also applies to short runs but is less
reliable : the expected bulges are extreemly variable. However, no
harm is likely to arise if it is used faithfully provided that the flnal
32-letter count is always looked at carefully. All exceptional features
or the count should be explained in terms of the factors mentioned in
the first part of the screed (allowing of course for reasonable random
deviation).
The final 32-letter count. There are two main reasons for a wrong de-X
being sent over with a confident comment.
(i) A feeble story being over-estimated at each stage. For instance a 3.6
sigma followed by a 3.5 sigma, that sets its last X with a 2.8 sigma. Such a
count is unlikely to be more than 3 decibans up in all, and should not be
described as "all certain" although there may be no one impulse that
strikes the eye as being doubtful. The comment "all certain" is defined
in this Section to mean "Each separate setting is better than 10:1 on".
(ii) Having one setting wrong. This can arise from a deceiving slide at