All your ways of attacking the mention IDS scheme from the parent can work, given the right circumstances. Basically what it comes down to is that if the attacker gained access successfully and is in control of the targeted machine it can always send out messages, as if it was not infected. At this point it is too late trying to detect anything. With a challenge-response scheme you might be able to ask for the exact contents of /usr/bin/md5 but you will never be able to tell whether there is also a /usr/bin/md5_modified on the server which is hidden and only used at some point.

Two alternatives might be possible to still have a relatively easy to manage and save IDS: You can boot from a r/o media every night and use this system to do the file integrity check (hard to automate a single boot from cd) or you could send a large number of log statements about logins, running processes, firewall logs to a different machine where any unusual events trigger an notice to check the system by hand.

It is definitely possible to use some narrow bandpass filters. In the infrared region there are various filters for available that have a wavelength window of 10 nm at 1000 nm. These filters are not available at Walmart, but they are not too costly either. Depending on size, quality, wavelength and other parameters you should be able to buy some for $50 (Thorlabs).

To actually hack the Kinect you have to test, whether there are other infrared filters used and if the camera is sensitive enough at different wavelengths. I don't think the properties of the reflecting materials should be of any concern. The reflection of materials in a household room should not change for a small frequency difference in the infrared region.

Using a time-multiplex approach with shutters or just software which switches the cameras on and off might work well in theory but should be rather impractical to do without significant changes to the Kinect hardware.

In short:

Desktops, not laptops. More reliable, cheaper, faster. Only get laptops if there is not enough desk space in the office. Avoid thin clients.

Brand: Do not build them yourselves. Get something from a big supplier. Get the business line, Dell or Lenovo.

Authentication and user managment: Do not bother below 30 persons. Above: Active Directory, LDAP, ...

Internal media server: If your users are competent a simple file server will be enough. Dell or Lenovo. Lots of SATA disks. More disks for onsite backup. Get offsite backup, too.

Web server: Do not inhouse your webserver. It provides you with no real advantages. You cannot compete on connection, reliability, UPS, etc with a big hoster. Get a development web server for testing inhouse

OS and software: Stick with what your users are competent with. Ubuntu is really nice, get it with enough Linux experience. Mac if you have enough money. Windows 7 is good and works.

General: Avoid "big" solutions. Do not blow money on anything were a salesmen visits your NGO. Learn from other NGOs.

In the natural sciences Wikipedia is an important tool in research. In independent reviews the accuracy was on an equal level as other encyclopedias (Britannica), see for example: http://upload.wikimedia.org/wikipedia/commons/b/b5/Wpausstellung-18.pdf (german language).
It provides a free source with fulltext search. In many cases the original research is cited, so that you can look for more detailed information.

Just imagine trying to get quick information about something without. I am currently working on Quantum criticality. A quick google search provides you with tons of information, the wikipedia entry is a accurate one-page document which cites the most important theoretical papers from the past few years.

There are quite a few points to consider how the game should adapt. Besides the simple fact that the main focus should always be on the fun aspect of the game I have two examples on my mind of good and bad adaptation. In neither cases "rewarding mediocrity" is a real concern.

The bad example of adaptation can be seen in quite a few FPS. Before adaptation there were different skill settings from beginner, easy, normal, veteran, nightmare. Now the game constantly assesses how well I am doing and as a somewhat experienced player it increases the difficulty level quickly. Now there are the basic enemies around the corner, who I had plenty of fun slaughtering early on in the game, which are now equipped with x-ray vision and one-shot-to-kill handguns. While the player may still manage to progress by being much more careful it just seems unfair and not fun anymore. So basic enemies should always be easy to overcome, no matter what.

A good example is the A.I. director from Left 4 Dead. Quite often the teams are unbalanced and it stops being enjoyable if you know that your team does not stand a chance at all. Sometimes right at that moment the A.I. notices and throws in an extra tank. This can often mess up even an organized team and now it immediately feels more balanced. Also the basic zombies are always easy to kill, the adaptation changes their number not their individual strength.

A simple way around automatic adaptation is to continually look how the player progresses into the game and then at the end of a level their commander just asks: "You exceeded my expectations, do you want a tougher challenge next time?"

LTSpice is free as in beer and works nicely even with more complicated problems. There is only a windows version available, but Linux support with wine should not be a problem. http://www.linear.com/designtools/software/

For simple circuits SolveElec runs on windows and mac, has a very nice user interface and is a good tool for teaching. http://www.physicsbox.com/indexsolveelec2en.html

A dirty trick that is still around in even the newest X86 processors and causing problems for hardware engineers and has security implications. The first xbox was hacked partially because of the A20 gate (pdf description of vulnerability) and maybe Intel will stop using the A20 gate in the upcoming Nehalem generation.

Why don't you just read the website you are citing:

Parent is right: money is not the argument, that is worth the switch. Software companies, Microsoft included want students to learn MS Office, Adobe, Matlab, Autodesk Inventor, etc. Some companies even give their student versions of really expensive software packages away for free, just have a look at Autodesk.

For the students it is of great value, if they are able to work efficiently with open source software. Just a few days ago I helped someone to switch from Endnote to Zotero+Jabref. It was quite a pain to convert from the Endnote format to something more open like the Bibtex format and there are several websites which show you 10 different hacks how to do it somehow.

With open source the file format is always documented, at least in the code itself. So if you want to work with your reference in 5 years without upgrading Endnote to Windows 8 this is the only sane choice.

For science in general it is necessary to check your results carefully and be able to reproduce other people's work somehow. How are you going to judge a paper claiming: "We simulated bla with this $$$ software package and it looks marvelous"?

Besides file formats and reproducibility in my opinion it is in most cases better to teach something that can be useful for the next 5-20 years, instead of some fast moving target. Software vendors often change their products and break backwards compatibility (Labview is great, but going back 2 versions is a no go) not because they invented this new must have feature but to sell the next version. If your students can do statistical analysis in Gnumeric and R they are well equipped for advanced work and do not have to worry about all the errors in Excel (statistics in Excel).

This would be easy to implement and some games show a similar behavior. Still this is not widespread because it just does not add to the gameplay.
A longer timeout for alertness would just mean, that you need to wait in a dark corner for a long time until the enemy finally goes back to "no alert". I certainly don't want to be punished for a tiny mistake by having to wait forever.
In a simulation rather than a game the enemy should not only react by increasing the alertness but calling for additional patrols, etc. which would reduce your chance of success significantly.

It is a fine balance between "not noticing a dead body" and "call for everybody to eliminate you".