The master schedule maps subjects to systems. Each line of the master schedule consists of a PIN followed by the nine site numbers that the user was to be connected to, in the order in which the user was to be connected to them. So, a line reading 1234 04 08 03 07 02 06 01 05 09 means the user with PIN 1234 should call system 04 (Colorado) first, system 08 (Lucent) second, system 03 (CMU) third, and so forth. This example line can be construed as skipping forward by fours, modulo 9. Here are more details on the way in which the 72 original subjects were used/scheduled. The master schedule came about in the first place after some sites decided they could not do the handshaking protocol. An initial version of the master schedule was intended to somewhat mimic important characteristics of the load-balancing scheduling algorithm and data structures that NIST presented at the Charleston, SC meeting. This initial version favored even representation of all possible ordered pairs of systems, and was created using a random-number generator. At Marilyn Walker's request, the master schedule was redone/replaced a few days before data collection began (changing from the schedule favoring even representation of all possible ordered pairs of systems to a schedule that had the same number of each of the nine scenarios performed by each of the nine systems). This change was considered to be a more desireable experimental design for the planned statistical analysis. The resulting replacement master schedule (the one actually used) has eight 9x9 Latin Squares. They come in pairs, so that the second square is basically the first backwards, and so forth. An alternate way of looking at the pairing of squares is that (wrapping around modulo the 9 positions) the first square goes forward by ones, the second square forward by eights, and so forth. As a result, all but 6 of the 72 possible ordered pairs of systems occur at least 8 times in the schedule. The missing ordered pairs are 07 followed by 01 (a subject calling MITRE and then AT&T next) 08 followed by 02 (Lucent then BBN) 09 followed by 03 (SRI then CMU) 01 followed by 07 (AT&T then MITRE) 02 followed by 08 (BBN then Lucent) 03 followed by 09 (CMU then SRI) These pairs are missing because one square goes forward by threes and one square goes backward by threes, and 3 evenly divides 9, so that there's no wrap-around modulo 9. Of course, not all scheduled calls were eventually made, since some users bailed out and some calls struck out. In the following table, additional PINs on a line are replacement subjects. In other words, all PINs on a line had the same schedule and are interchangeable for analysis. Most replacements were for subjects who had made few or no calls at the point when they were replaced (some later made all 9 calls). In addition, three subjects were replaced due to characteristics of their speech or accent, and these are noted below. First Latin Square 1401 1491 2209 3017 3825 4633 5441 2198 6249 2502 7057 7865 3310 (Thus, original subjects 1401, 2209, 3017, ..., 7057, 7865 make up the first Latin Square, and four of those subjects were replaced because they had not made all their calls) Second Latin Square 1805 2613 3421 4229 1996 5037 5845 6653 7461 8269 Third Latin Square 1603 1087 2411 3219 1390 4027 1895 4835 5643 2805 6451 7259 8067 Fourth Latin Square 2007 2815 1188 3623 1794 4431 5239 6047 2299 6855 3108 7663 8471 1592 (subject 8471 said to have foreign accent) Fifth Latin Square 1906 2714 3522 4330 3815 (subject 4330 said to have foreign accent) 5138 5946 6754 3007 7562 3714 (subject 7562 has impaired speech -- not always evident) 8370 3613 Sixth Latin Square 2108 2916 1289 3724 4532 5340 6148 6956 7764 8572 Seventh Latin Square 3118 1693 3926 4734 5542 2906 6350 2603 7158 7966 3411 1502 2310 1300 Eighth Latin Square 5744 2300 6552 2704 7360 8168 3512 1704 2512 3320 4128 4936 2097