296 TOWERS AND TANKS FOR WATER-WORKS. 



be asked for under extraordinary conditions, and should be granted 

 even more rarely than they are asked for, errors in the tests of 

 course excepted. 



" (14) Simple fairness requires that when it is desired that 

 material once fairly rejected should nevertheless be used, some 

 concession in price should be made. 



" ( T 5) Where commercial transactions are between honorable 

 people, there is no real necessity for marking rejected material 

 to prevent its being offered a second time. If it has failed once, 

 it will probably fail a second time, and if return freight is rigidly 

 collected on returned shipments the risk of loss is greater than 

 most shippers will care to incur. Moreover, it is so easy for the 

 consumer to put an inconspicuous private mark on rejected 

 material, that it is believed few will care to incur the probable 

 loss of business that will result from the detection of an effort 

 to dispose of a rejected shipment by offering it a second 

 time. 



" (16) All specifications in actual practical daily use need 

 revision from time to time, as new information is obtained, due 

 to progress in knowledge, changes in methods of manufacture, 

 and changes in the use of materials. A new specification that 

 is, one for a material which has hitherto been bought on the reputa- 

 tion of the makers and without any examination as to quality 

 will be fortunate if it does not require revision in from six to ten 

 months after it is first issued. 



" (17) In the enforcement of specifications, it is undoubtedly a, 

 breach of contract legitimately leading to a rejection if the specified 

 tests give results not wholly within the limits, and this is especially 

 true if the limits are reasonably wide. But it must be remembered 

 that no tests give the absolute truth, and where the results are 

 near, but just outside of the limit, the material may actually 

 be all right. It seems to us better, therefore, to allow a small 

 margin from the actual published limit, equal to the probable 

 limit of error in the method of testing employed, and allow for 



