Ususally the most important information is the first and last step

E.g.

The TopHat run should be called

TopHat on SOLiD 24A

The alignment stats should be

SAM/BAM Summary Metrics of Solid 24A

With the rest of the tools in the chain identified in the "more information" box.

This would also give graph generating tools a fighting chance to present something useful in any graphs generated.

E.g.

GC Bias Plot of Solid 24A could have a title of Solid 24A instead of dataset_234.dat

What do you think of this first-last model?

This model breaks down during experimentation. E.g. let's say three different methods for trimming a FastQ dataset are tried before mapping with Bowtie. Currently, the Bowtie runs are named differently b/c each trimmed dataset is a unique input. Using first-last model, all datasets are named the same and it is not possible to differentiate b/t them without looking at the inputs, which requires clicking on the rerun/info button and finding the input(s). The current approach used by Galaxy lists the inputs in the dataset title to avoid these issues.

Datasets with the same name becomes more problematic as more steps are added b/t first and last because, while they have the same name, the steps taken to produce them may be very different.

The first-last model could be nice for workflows, though, perhaps as an extension of the "rename dataset" actions or a kind of global "rename dataset" action.