General Report on Tunny
25D Page 163
runs not involving the chi-wheel to which the doubts belong ('running against doubts'); and may be worth while; e.g. if chi 5 is heavily doubted 4=/5=1=2 against the known characters of ΔΧ5, and 4=/1=2 against the doubted characters of ΔΧ5 are independent, and the latter is likely to be useful.
N.B. It should NOT be decibanned from the letter count against known characters.
In difficult wheel-breaking this device is used extensively. (R3 pp 13, 25.)
Doubting reduces the effective text: for example if one third of each of the four wheels is doubted, the remaining text is 
, i.e. less than one fifth of the whole.
When deciding how many characters to doubt, it is necessary to judge between the conflicting considerations of not losing too much text, and of not including too many wrong characters. 10 decibans is usually reasonable evidence for inclusion.
(b) Setting other messages (on Colossus).
That the evidence from a single message should suffice to make all wheels complete and certain is exceptional, but it will commonly make them sufficiently complete to set other messages, the addition of whose evidence, which is independent, will suffice. The addition of so much independent evidence is most effective; but rather prosaic, and apt to be unjustly neglected in favour of 'squeezing' a single message.
When there are more than a very few doubts, setting is complicated by 'variable R' e.g. if Χ3 is being set by means of 3x/1x2., Χ1, Χ2 are fixed in the cipher, whilst Χ3 is tried in all possible 29 positions. Of the places where 1x2., the only ones looked at are those where ΔΧ3 is known and this may vary considerably when Χ3 steps. Thus a large 3x1x2. may be due to a large 1x2., which is not relevant to setting ΔΧ3.
This is commonly circumvented by printing R, i.e. (1x2.) and the score, (3x1x2.), for all positions of Χ3, afterwards finding the sigma-age 
for promising scores.
A preferred modification which reduces unless printing is to run simultaneously, on two counters: 3x/1x2. with a high set total; 3./1x2. with a low set total, (with SIP if available). If the bulge of 3x/1x2. over 3./1x2. is significant, one score or the other must be printed. To
Back to General Report on Tunny. Contents.