You also need to train your eye on spotting potential improvements in whatever you see. Read submission texts carefully (prefer detailed ones), follow forum threads, because there people describe how exactly they work, how the games work, what aspects one must consider. This is also widely described in this Wiki, the links are provided in proper places.

Many useful details can be found if you pick someone else's TAS and try matching its time yourself from scratch. You will notice the places where he is ahead, work out how it was achieved, replicate techniques, and (hopefully) even improve the existing record.

To free your mind, try getting rid of routine, use the best tools you can find, because they afford abstraction, when you can automate some actions or calculations, loosing mental resources for thinking over gameplay, route, tricks - for inventing ideas.

Read more how to work inventively and effectively.

It is recommended to place a backup copy on the internet (User files, Microstorage, DropBox) or on a portable disk such as a floppy or USB drive. Computer crashes are not unheard of. SVN hostings such as http://code.google.com are very useful too.

Periodically verify that your movie plays back correctly. Desyncs may occur, depending on the game and emulator.

If taking damage can save time (damage boosts, invincibility periods, passing an obstacle or an enemy), you must find all places where it can be used, spend all your hitpoints before the place where you can refill them and test different options to spend them, one can save more time than the other.

If a player's death throws him back within some segment, it can be used to get there quicker. On the opposite, if death respawns you at the same place, you can sacrifice lives as long as you have them.

Different items and weapons can give you important advantages: killing enemies faster, moving faster. Plan your use of them, spend as much as you can between refill points.

Read more about lag reduction.

It's up to developer to choose the entropy resource, that can not be foreseen by a regular player. Frame numbers, objects positions, CPU cycles, input - any of these can be used as RNGs. It can require much effort to reverse engineer a game to figure out what it uses and what it outputs. That means, you can still try trial and error method (see below), but examining the game code, while giving more powers, requires deeper knowledge.

Read this guide on luck manipulation, and it's advanced subpage.

Test runs should be based entirely on route and strategy. All long dialogs in a test run should be cleared using the emulator's autofire buttons. Try to work out the game mechanics so you know whether delaying button presses changes whether attacks are critical hits, or misses, or such. Try changing previous button presses if attacks don't change because sometimes the results of attacks are determined further back.

If the RNG hates you (e.g. nothing changes no matter what you do), give up. Not all games are created equal.

Once you have worked out the game mechanics, carefully construct the route through the run. The strategy is often going to change, depending on the circumstances. Try not to be too committal, although that might work sometimes. You may have to create backups and branch the movie to test two different strategies. The important thing is that you know why such strategy can work.

Try to shoot for uncommon or better, but not rare, probabilities. Assuming delaying button presses changes attack results, 1/10 is huge for a TAS. 1/20 is not far off. Generally, the lower the probability, the greater the justification. If you obsess over 1/100, it should be very worthwhile. Probabilities worse than 1/300 shouldn't even be considered, unless you know what you are doing.

Do not overdo an attempt to pull off a strategy. If it seems impossible or unreasonably improbable after testing, alter your strategy and move on if possible.

The more you redo a section, the better it becomes. But at first you have no interest in redoing something. If you notice a mistake in what's already done, dare to go back and fix it. If it involves redoing of the further actions, redo them, and you will likely find mistakes unnoticeable on the first glance. This happens because every time you redo a section, you become more familiar with how it works, abstract of the old impression of it, or have found new aspects farther in the run that are still applicable to the earlier sections.

However, if you do not feel any potential at all, there is no sense in endless overdoing, because this way you can never finish. Even highly optimized runs can still have non-applied improvements, if the author considers it is not worth a new attempt, while it can still meet publication standard.

Basic traits to pay attention to:

• Avoid bumping the character on a wall that he is jumping over.
• Avoid stopping unnecessarily.
• See if jumping into a shaft (that you need to go down) is faster than just sliding off the edge.
• When faced with an enemy, try to go around or through it, or kill it, without stopping.
• See if performing some action changes the way a boss or enemy behaves (if this is a micro rather than a macro effect, this is known as "luck manipulation")

If it is faster to take damage (for example, by running into an enemy instead of going around it or trying to kill it), take damage. As a stylistic consideration, you may choose to take no damage whenever taking damage saves time, in order to make the run look better. In either case, be consistent in always taking damage to save time or taking no damage. Do not take damage if you can avoid it without wasting time (see “Be interesting” below).

It is the same issue concerning using death to save time.

This applies also to auto-scrolling levels where you can entertain the audience instead of just staying at the front of the screen. See the Super Mario 3 video for a perfect example.

Be different. Take faster routes or shortcuts that no one takes, whether undesirable or impossible in real play. If you think of a route that seems blocked but would speed up your play significantly, find out how to go through, over or under it. If you want some radical shortcut or strategy to happen, find out how to make it happen.

If something that relies on chance usually does not happen, make it happen. The game is only as random as you are. For example, you can win lotteries, deliver critical hits, and make bosses behave how you want them to behave. Take care, as the game is flexible only to a certain extent. See Luck Manipulation for how to do this.

In games that have a user-controllable camera that does not affect the actual playing, try to control the camera so that it provides the most interesting or beautiful angles at all time instead of going with default settings.

Push actions to the limit. If you can destroy a target as soon or as late as possible without losing time, do so. If you can make your jumps as short as possible without losing time, do so. Do not have slow reactions; act as soon as possible.

Do not sleep. You are supposed to be the master of the game, not the slave of the game. Aim for the impossible. Think outside the box and do not give up if your ideas do not work right away. Just because you cannot get something to work right away does not mean that it is not possible.

TAS records always appear unbeatable before you try to beat them. So try to beat them.

Similarly, even if a certain move may be the fastest one, hearing the same sound effect all the time can quickly become annoying. Especially if said move is used to move faster than just walking. Find ways to shorten repetitive parts.

If exploiting the camera can allow the player to pass through walls or similar bugs, do so. However, do not turn the game's video into a kaleidoscope, the novelty wears off quickly. A game may have less lag if the camera focuses on less material, but no one wants to watch a video of pure black, balance the performance gains with watchability.