As I've discussed before, the traditional version in common use (as evidenced by the number of places you can download the paper version on the internet) has some serious problems and confounds, which are addressed in the one available for PEBL.
About a decade ago, Tim Salthouse and colleagues developed and tested  a version based variations introduced by  Oswald and Roth (1978) and also described by Vernon (1993). They called the "connections" test, which avoided some additional confounds found in the original trail-making test. And unlike the traditional trail-making test where you are timed on how long it takes to complete the test, this test gives you a time limit and sees how many targets you can complete.
One of the main benefits of the Connections test is that it minimizes motor movement. Salthouse's test was paper and pencil, but this could also be a problem for a computer-administered version, as older participants might be less familiar with using a computer mouse, or might have more difficulty with it even if familiar. They were concerned that the traditional trailmaking test had non-trivial motor movement times. To reduce or eliminate these, they created a matrix-based trailmaking test, where each next response was adjacent to the previous response (either vertically, horizontally, or diagonally on cell away). They then did the traditional comparison between single-set and switching trials to determine the cost of cognitive set switching. I believe that responses could be made using a keyboard arrow keys, which can avoid the problems with mouse movement entirely (although it introduces another psycho-motor modality that is probably harder to learn than the mouse).
To implement this in PEBL was fairly straightforward, as it was quite similar to the standard TMT. The main complication was to develop an algorithm to do the layout, so multiple tests could be given with test problems developed automatically. Salthouse originally had four forms--we'd like to have as many forms as you need. To add within-participant rigor, we will generate forms at the start of the test, but use each base form in both a switch and non-switch condition (rotated and flipped along the main diagonal).
The algorithm is quite simple, and works well for problems of small or large sizes. A legal path must start at the upper left and end at the lower right, and then hit every node along the way. So to start, we find a random path from start to finish made by just going right, down, or diagonal right/down. If we get to a horizontal or vertical edge, we just continue to move directly toward the end to complete the path.
Following this, we need to attach every node to the path. Any node that is currently adjacent to the path can usually be added by splicing it in, and there are often more than one options to splice. There are a few little problems to work out. One is that sometimes the path will cross itself. This may or may not be a problem, but this can be checked. Finally, sometimes you back yourself into a corner and there is no way to attach a node to the current path. In this case, we simply choose two neighbors of this 'problem' node, splice the current node in, and delete the path between those neighbors, sending them back to the 'unattached' points.
A screencast of this algorithm is below, along with a few trials of the task at work.
Data are recorded to three files. First, a .csv file called connections-path-X.csv records each click. Next, summary stats for each path are saved in connections-X.csv. The screenshot below shows some of these stats:
Each row is a trial, and it provides some summary statistics. These include the total number of nodes achieved, as well as the length of the path, and some other statistics that can help determine problem difficulty. Finally, a short 'report' file is saved that gives the most important information (time and accuracy), as seen below:
Mode: Single Switch Ratio/difference
Targets: 30 22.25 0.741667
RT: 20065.8 20063.5 0.999888
Clicks: 30 24.25 -5.75
OverClicks: 0 8 8
There are a few options available. You can set the the size of the problem, as well as the time limit, which is currently defaulted to 20 seconds. If you set time limit to -1, there will be no time limit, and each problem will continue until all nodes are clicked.
1. Salthouse, T. A., Toth, J., Daniels, K., Parks, C., Pak, R., Wolbrette, M., et al. (2000). Effects of aging on the efficiency of task switching in a variant of the Trail Making Test. Neuropsychology, 14, 102–111.
2. Oswald, W. D,, & Roth, E. (1978). Der Zahlen-Verbindungs Test (ZVT). Gottingen, Germany: Hogrefe.
3. Vernon, P. A. (1993). Der Zahlen-Verbindsungs Test and other trail-making correlates of general intelligence. Personality and Individual Differenc es, 14, 35-40