I save every command and the resulting main output by location to a database with user session information. We also had a sign on screen that asked for basic information like school, location, age, and gender.

Then I created reports that allowed you to drill into any location and review commands. It's very useful to see that a particular point in the game breaks down with either guess the verb issues, or in the case of newbies, when they started to lose interest and type insane things.

I also had reports for a session, which is essentially a full transcript.

I think saving transcripts is an excellent first step, but building a detailed database of command, location, output, and any other game state information is extremely helpful for improving the final product.

I had been thinking I should write up a public summary of the discussions we'd been having before on the intfic forum - this would be good to standardise so that it's not Parchment specific.

The development version of Parchment that I'm slowly working on will make this easier. You'll be able to go $(document).bind('lineInput', handler) to listen in to all of the commands, without needing to hack the core source code. You could do that with another JS file, or even inline in your parchment.html.

The current proposals had only been about logging commands. Your idea to specifically log errors and library messages separately is an interesting one. We could perhaps do that by detecting Parchment based on interpreter ID... at least then nothing would happen for normal terps.

I'd taken a slightly different tack so as not to need AJAX or output modifications: I was hooking into player input internal to Parchment and sending only those commands, not the game's response, to cut down on data traffic and storage requirements at the other end.