Introduction
"My take on Team Fortress 2 and what Valve has done wrong and right."
That's what this post was originally supposed to be. However, Valve has done so many things right it would be nearly impossible to delineate all of them. That being said, while I do try to avoid stating the obvious, there's a lot to be learned from looking at good design, and that learning can go a long way when analyzing problems. Expect to see some of both.
Most of the game's issues are emergent: they slowly come to light over the course of many months of gameplay by millions of players. Valve couldn't have been expected to have anticipated most of them. After all, emergent gameplay defies expectations quite nearly by its definition.
Case in point: In CEVO's first Team Fortress 2 tournament (shortly after its release in 2007), teams seldom used the demoman as part of their regular lineup on push maps, instead preferring to run a sniper. Since then, however, the demoman's potential has been more fully realized. He has been nerfed more than any other class since the game's launch, yet he continues to be perceived as one of the strongest classes in the game.
Another example: Team Fortress 2 launched with six maps, and the one that Valve was most proud of was Hydro. However, Hydro has proven to be the least popular of the six maps that the game launched with.
Before we begin
One thing I'd like to point out is that having a game that is "balanced" and having a game that is "fun" are two entirely different things. It seems that the two are often confused because it is difficult to achieve "fun" without some level of balance. "Fun," to some degree, does imply balance. However, just because a game element is "balanced" does not mean that it enhances gameplay. Case in point: the sandman is, by most accounts, "balanced" in its current capacity (possibly bar the issue of uber stunning). However, it detracts from gameplay experience of all participants. The sandman achieves balance by ruining everyone's day.
If I had one piece of advice for Valve, it would be this: prioritize making the game fun ahead of making it balanced. Balance is important, but you'll generally find that most "fun" game elements are fun partially because they help preserve balance. Even if they don't, it's comparatively easy to tweak a fun game element to the point where it's balanced. It's not so easy to take something that is balanced and tweak it in the direction of fun.
It's also important to realize that "fun" does not mean "amusing." The rocket launcher is a good example of a weapon that is "fun." It can be used in a myriad of ways. You don't need exceptional aim to hit people with the rockets' splash damage, but the weapon still has an insanely high skill ceiling and climbing toward that ceiling has very noticeable and tangible benefits. The sandvich, on the other hand, is a good example of a weapon that is "amusing." As a healing item, it's more of a tactical decision than a toy to play with, and as such it does not allow the player to experience "fun" on the same level as the rocket launcher.
Maps
Several months into the lifespan of Team Fortress 2, Valve released a map called Badlands. It's a fantastic map that has proven to be the most popular push map for both public and competitive play. There are several things to learn from Badlands, aside from the fact that Valve apparently possesses a peerless ability to construct good push maps.
All of Badlands' features contribute to one thing that makes it an incredibly face-paced and exciting map to play: It's not prone to turtling. The middle control point has three points of entry from either side of the map, meaning that it is impossible for a single sentry or demoman to single-handedly stop enemies from advancing. The Spire control points and the yards that surround them are open enough to prevent anyone (defender or attacker) from being able to secure the control point from a position of relative invulnerability. The open nature of the map reduces the effectiveness of spam. The only portion of the map that forces players to take an indoor route is the final control point, and turtling strategies here often backfire because the control point can be captured in a single second.
A push map should be a constant tug of war (fitting the "push" moniker), with teams always focused on the goal of advancing to expand their own territory rather than turtling to preserve the status quo. Badlands executes this perfectly.
One thing to note is that a map which doesn't lend itself to turtling is not a map where the engineer is useless. In fact, many players in public settings have noted that it is more fun to play the engineer offensively than defensively. By making a map not prone to turtling, you discourage engineers from building level 3 sentries that require the engineer to stay at home to tend (boring engineer play) and encourage them to do things like place level 1 sentries in sneaky flanking positions to catch foes unawares (fun engineering, in my eyes). In a public environment, it's very possible to play an effective engineer on Badlands simply by placing sentries in unexpected locations and making use of the engineer's shotgun and pistol. Level 1 sentries don't last long but they get kills and allow the engineer to be more mobile, contributing to his team wherever the action on the map is taking place.
Returning to the topic of Hydro (a map that Valve took great pride in before it turned out to be a colossal flop), the thing that they emphasized as the map's greatest virtue was its dynamic nature. I strongly agree that a dynamic map is one that is fun to replay. Push maps put into practice what Hydro attempted to achieve: the action shifts around the map much more quickly without separating the game into "rounds" that arbitrarily halt a team's forward progress. A dynamic map is one where the action shifts at a regular pace.
The idea that maps should be dynamic needn't apply exclusively to push maps. Good attack/defend maps should, similarly, be dynamic. It's for this reason that you don't see A/D maps featuring only one control point: attacking the same part of the map for the entire round would be tremendously boring and repetitive. It's evident that Valve has tried to minimize repetitiveness within A/D maps by choosing maps with multiple stages, like Dustbowl, Goldrush, Egypt, and Hoodoo.
Unfortunately, these maps share one glaring flaw: The defending team wins a majority of the time. How is this a flaw? On an attack/defend map, there is only one way for action on the map to shift: the attackers need to gain territory. A map where the defenders win most of the time is a map that is slow-paced and static. (This goes hand in hand with turtling interfering with fluid gameplay.) It's evident that Valve tried to mitigate this with the Payload gametype by giving the attacking team a roaming objective rather than a static control point, but the attacking team still runs into the same problem of being stuck at easily-defensible choke points on maps like Goldrush.
Here's a heat map of Goldrush showing where player deaths occur most frequently. Note that the vast, vast majority of the map's deaths occur where the attackers try to break past the map's first choke point.
If Valve wants to create maps that are dynamic, the importance of the attacking team (the team that moves the action forward) having an advantage needs to be strongly emphasized. If the attacking team is winning, the action is shifting around the map, providing fresh experiences for both teams. See Gravelpit for a good example; it's the only stock map where the attackers win a majority of the time. If the defending team is winning, both teams are forced to do battle at the same choke points, which can get stale quite quickly.
It seems that 32-player servers are a relatively common sight, which is a bit of a problem for Valve. None of Valve's maps were really built with 32 players in mind, and most maps tend to scale up more awkwardly than they scale down. My suggestion to Valve is this: Allow the community to create those 32-player maps. Frontier and Cashworks are examples of maps that work as well as can be expected in a 32-player environment, and they work well in that capacity largely because they were specifically engineered for it. Rather than relearn how to design maps so that they can accommodate 32 players comfortably, Valve can leave this task to custom mappers (who have already demonstrated a willingness and ability to construct maps of this variety).
It should already be apparent by now, but arena is not a tremendously popular game mode. (For a look at just how unpopular arena mode is, check the Team Fortress 2 stats page.) Valve could stop making Arena maps and most people would not mind at all. It's my hope that they discontinue the trend of releasing new arena maps with each update and devote their time to creating content that more people can appreciate.
Three out of the four most popular maps in public setting are attack/defend maps. (The other is 2fort.) I may be guilty of faulty application of Occam's razor here, but I think that the reason that attack/defend is such a popular game mode in a public setting is that putting the focus on a single objective makes it relatively easy for teams to play as a cohesive unit. On CTF and push maps, you run into issues where a team must defend their own objective and attack the enemy objective. This can lead to fragmented teams, with half the team staying at their own point to defend and the other half trying to attack the enemy point, each frustrated by the fact that not all of their team is joining in the effort. Attack/defend doesn't suffer from these problems because the focus is on one objective. The entire red team defends, and the entire blu team attacks.
There is a map that preserved this appealing aspect of attack/defend while presenting it in a form similar to that of a traditional push map. I am speaking of pl_waste, a tug of war payload map. A single payload cart starts in center of the symmetrical map, and both teams try to push the payload away from their base and into the enemy base. It gives you the wonderful back-and-fourth dynamic where the action is constantly shifting around the map while still allowing both teams to focus on the singular objective of a payload cart. I really feel that this tug-of-war payload game mode has more potential than the "payload race" mode introduced with the Valve map Pipeline.
Although removing split objectives from a map in a public setting may lead to greater cooperative play, at high levels of organized play most players prefer traditional 5 CP push to tug-of-war payload because it is faster paced and places more emphasis on planning and strategy by forcing teams to consider two objectives at once. The beauty of the tug-of-war payload game type is that maps made for this format can often be modified to satisfy those who crave traditional push maps as well. Waste, though originally a tug-of-war payload map, has also been released in the form of a 5 CP push map.
Changes to existing game modes
One of the biggest problems with CTF seems to be that the definition of a "win" makes it especially prone to turtling. With capture limits, it becomes tactically feasible to turtle to avoid a loss and force a stalemate. If a team is down 2-0 and the capture limit is 3, the team that is lagging behind can't reasonably expect to make three captures. Instead, they will probably elect to attempt to avoid a loss by forcing a stalemate. This isn't fun for anyone.
As it is, CTF seems to be mostly a deathmatch, at least on public servers. Valve has expressed that this was not their intention, but in the case of 2fort I think it's best to work with the player mentality than against it. Simply create a single round with no capture limit and a defined time limit (10 minutes seems like a good number, although 30 minutes could work if the desire is for a single round during each map rotation). The goal of both teams is to get as many captures during that time window as possible. That way, the only way to avoid losing is to score at least as many points as the opposing team, encouraging players to focus on capturing the enemy flag just as much as defending their own.
One of the things that set arena mode back several paces was the introduction of first blood: arena as we once knew it no longer exists. The crit buff drastically alters the pace of smaller games, where it can allow a single player to kill entire teams in some situations. First blood needn't be removed entirely, but an optional cvar to disable it would certainly be welcome.
Attack/defend maps are best enjoyed with stopwatch enabled, and thankfully Valve already has a cvar to enable stopwatch mode. Unfortunately, it's only possible to use it when tournament mode is enabled, making it infeasible for public play. Players shouldn't be denied access to this setting just because they choose to play in a public setting.
For those who don't understand stopwatch mode, I'll attempt to briefly summarize how it works. There are two rounds, so each team has one round where they attack and one round where they defend. The winner of those two games is decided based on which team captured more points. If both teams captured the same number of points, the win goes to the team that took the least amount of time to capture their points. What this means is that teams that defend well are rewarded with lower win requirements when it is their turn to attack, and teams that attack well are rewarded with more breathing room when they get to defend.
At a bare minimum, the game needs an optional stopwatch cvar that isn't bound to tournament mode. Whether or not Valve wants to make stopwatch the default for attack/defend maps is a different matter entirely.
Non-gameplay issues
The process for creating and managing demos could be improved. As things currently stand, I need to manually type in a file name every single time I want to record a demo. It would be nice if the syntax existed for me to create scripts to automate the demo-naming process. Strings to include in the file name could be the date and time of the recording, the map the recording was made on, the class being played at the start of the recording, or the server IP address. I'm perfectly okay with doing all of the recording on my client as long as I can script it so that it only requires a single keystroke. There are individuals working on external programs to accomplish this, but it really seems like something that ought to be built in the game (and it seems like it would be relatively simple to implement it).
When the new unlockables first shipped, swapping your loadout was a relatively painless process. There was an equip command that made it possible to create scripts to change your loadout for you, and changing between weapons was as simple as using a keybind for many people. In addition, it was possible to change your loadout from anywhere inside your spawn room (without having to touch the resupply cabinets). For some unknown reason, Valve did away with both of these wonderful features, making the process of changing the loadout unnecessarily painful and cumbersome. Even worse is the fact that some maps have forward spawns that do not have resupply cabinets. There are workarounds for the resupply cabinet issue (swapping classes), but it seems quite silly when the previous system was so much easier.
Right now, it's not possible to change your viewmodels from right to left handed (or visa versa) when you're connected to a server. The goal of this was to defeat scripts that would be used to swap the viewmodels in mid-combat, but a much less restrictive solution would be to limit viewmodel swapping to the spawn room. Left-handed viewmodels opened up the possibility of truly balanced horseshoe-shaped push maps, but if players have to disconnect from the server to change their viewmodels, it puts the team turning right at a disadvantage. (Similarly, if a player got team autobalanced on such a map, it would be frustrating for them to have to disconnect to change their viewmodels).
When scoring a kill, the target of the kill is more important than the method of execution. Killing someone with a bodyshot achieves the same effect as killing them with a headshot and in some cases deliberately aiming for the head when a bodyshot would suffice is foolish. The current points system encourages this kind of foolish behavior. Killing an enemy with a backstab or a headshot should not award two points. Awarding bonus points to scouts who kill medics that are actively healing someone is a step in the right direction. Another idea is to give all classes a one-point bonus for killing medics who are holding uber, giving them more of an incentive to focus on important targets.
Minor gameplay issues
When a burning player uses the medic call, it's safe to assume that they would like to be extinguished. As such, it would be nice if using the medic call while burning notified all teammates with extinguishing capabilities. This includes medics, air blast pyros, and jarate snipers. Also, awarding an assist point to pyros and jarate snipers who are kind enough to extinguish their teammates seems like it would provide a greater incentive for them to do their teammates the favor. (Medics already get points for healing people.)
Most people use the medic call as a means of rendezvousing with a friendly medic. There are HUD icons to assist in this, but they only work one way: the medic can see the calling player, but the caller can't see the medic. The ability to see HUD icons for medics would greatly assist players trying to locate heals (and reduce the frustration of medics chasing down teammates who are running away from them, oblivious to their presence).
Class-specific issues
Scout
The scout is similar to the sniper in that he relies on hitscan weapons rather than projectiles and performs best at taking out individual targets rather than engaging large groups. Jarate is a really nice unlock for the sniper because it gives him a throwable item with an area of effect. (That's almost the definition of a "grenade.") The scout would have done better with something like the jarate over the sandman. The sandman is a bit like a concussion grenade, but it manages to preserve the worst aspects of the concussion grenade while taking away what might have made it useful as an item (area of effect and knockback that could be used for a third jump, which Valve ended up including in the Force-A-Nature).
The Force-A-Nature suffers from the same problem that the sandman does, albeit to a lesser extent. Robin Walker has already stated that they are working with the Force-A-Nature's knockback code, so hopefully the gun's issues will not be long-lasting. As it is, the Force-A-Nature's knockback doesn't seem to be working as intended, a good example being that it breaks a player's forward momentum even when it is fired at their back.
Another gripe about the Force-A-Nature's is that there is only one thing that affects knockback: range. It seems a little bit wrong that a 9-damage shot at close range has the same knockback as a 108-damage shot at the same range. The knockback should in part be determined by the number of pellets that hit the target. (Things could be simplified further if the knockback was simply based on damage, since damage is a function of distance and pellet count.) This rewards scouts for aiming well and landing shots at close range without penalizing rocket jumping soldiers who just happened to get clipped by a single pellet on the way up.
Pyro
The pyro is often criticized for being a novice class with a relatively low skill ceiling. This is understandable, given that it is entirely possible for a pyro to perform well on a highly populated server by charging enemies head-on and dealing damage with the flamethrower, a practice commonly referred to as "W+M1." However, the pyro's skill ceiling is actually higher than one might estimate it to be. Reflecting rockets, airblast/axtinguisher combos, and borrowed rocket jumps are all skills that take quite a bit of practice to master. Even the backburner has a learning curve to scale, such as discovering the ideal ambush locations on each map and gaining the situational awareness needed to use it effectively as an ambush weapon. The problem with the pyro is that these more "advanced" techniques (airblast acrobatics and ambushing with the backburner), while effective, aren't drastically more effective in most settings than the W+M1 approach. The problem is not that the skill ceiling is low, but that the incremental gains at higher levels are relatively marginal and the time required to master them to the point where they become more effective than W+M1 serves as a short-term deterrent to becoming a more advanced pyro.
There have been a number of proposed solutions to this issue, many of them horribly complex. I believe that the solution is relatively simple: return the flamethrower's damage dropoff to what it was when the game first launched. Players who started playing post-pyro update may not know this, but the pyro's flamethrower has been considerably buffed since launch. When the pyro update came, it received two buffs: Players had a choice between an airblast or 100% crits from behind, and the second buff came in the form of a reduction to the weapon's damage dropoff.
Pre-update, the pyro's usefulness was limited to situations where he could ambush foes and catch them unawares. W+M1 tactics were seldom effective, motivating players to find creative ways to make use of the pyro. The beauty of restoring the damage dropoff to its original levels is that it does not decrease the pyro's effectiveness at high levels. An airblast pyro is still able to use the flamethrower to set people on fire to set them up for axtinguisher combos, and a backburner pyro completely overrides damage dropoff when he is attacking from behind. (Critical hits ignore damage dropoff.) Airblast pyros also have the option of torching juggled foes once they have gotten close enough to deal maximum damage with the flamthrower. The only time that damage dropoff affects the flamethrower is when attacking at range (from the front in the case of the backburner). This isn't exactly good practice, so it's a modification that reduces the effectiveness of low-level pyros, giving them a greater incentive to try out more advanced techniques. For some people, it may be as simple as discovering that the backburner is more effective when directed at the backs of enemies. For others, it means using an enemy rocket to jump close enough to set them on fire, airblast them against a wall, and whack them with the axtinguisher.
Returning the flamethrower's damage dropoff to its original state isn't really "nerfing" the pyro; it's simply removing a buff. Valve has, in the past, removed buffs that had an adverse affect on the game (such as the backburner's +50 health). Although it's not as apparent, over time it seems that the flamethrower damage buff has adversely affected the game, and it doesn't seem unreasonable for Valve to roll it back.
The pyro issue is one that really doesn't relate to balance at all. At high levels of play, the pyro is one of the lesser-used classes, so it can hardly be considered "overpowered." (If anything, balance suggests that the pyro needs a tweak that would make him more viable in this format.) However, the pyro should be styled so that players have ample incentive to learn and improve. I believe that doing so would allow many players to discover new and and fun ways to use the pyro.
Demoman
The demoman has turned out to be one of the most powerful classes in the game, and Valve has described the sticky bomb launcher as being a bit "out of whack." However, I don't believe that the problem lies solely with the sticky bomb launcher.
Power aside, one of the reasons that players turn to the sticky bomb launcher so frequently is that the grenade launcher has several things that constrain its usefulness in close combat. Probably one of the biggest issues to note is that the grenade launcher has an obscenely long base reload time. It takes 1.24 seconds to insert a grenade into the grenade launcher's chamber, compared with 1.10 seconds for a sticky. That means that if both of your clips are empty, the sticky bomb launcher will allow you to fire sooner than the grenade launcher.
The grenade launcher's high base reload time has also had another side-effect that I'm pretty sure was unintended. Although reloading the first grenade takes a relatively long amount of time, each subsequent grenade only takes .62 seconds to insert. (That's faster than the rocket launcher.)
Here's a graph illustrating the reload times associated with the grenade launcher:
The issue is that because the incremental cost of each additional grenade is so low, it's most efficient to reload four grenades at a time. Reloading four grenades at a time means shooting four grenades at a time. This means that the grenade launcher is prone to be empty when it is not full, and an empty grenade launcher really only leaves the demoman with the sticky bomb launcher due to the long base reload of the grenade launcher. It also means that the grenade launcher is prone to be rather spammy: if the demoman is firing four grenades at a time, he doesn't have to worry about whether any given grenade hits since odds are that at least several of them will be close enough to deal damage.
What I propose is that Valve slightly tweak the grenade launcher (which will lead into a corresponding tweak to nerf the sticky bomb launcher). The solution is simple: decrease the base reload time and increase the reload time of each additional grenade. For example, if the graph looks like this:
Here, you do save a little bit of time by reloading more than one grenade at a time, but the total time saved is marginal enough to make it more viable for demomen to shoot one grenade at a time. It also doesn't increase the effectiveness of the grenade launcher as a spam weapon: although it reloads the first grenade more quickly, each additional grenade is slow enough that it still takes a full 3.10 seconds to completely fill the chamber from zero to four grenades. In fact, if Valve wanted to, they could reduce the base reload time but increase the subsequent reload time so that it takes more than 3.10 seconds to completely fill the chamber, making the modification less of a "buff" and more a "change in the direction of discouraging spam." (The graph above is just an example of what could be done; if playtesting reveals that drastically reducing the reload time for the first grenade causes the demoman to become overpowered, Valve can test until they find a happy medium between the two extremes.)
We can also modify the sticky bomb launcher in the same manner that we modified the grenade launcher. With the grenade launcher, the goal was to make it more viable in combat without affecting its effectiveness as a spam weapon (or possibly decreasing it). With the sticky bomb launcher, we want to decrease its viability as a combat weapon without hindering the player's ability to use it as a defensive weapon.
Right now, the sticky bomb launcher reloads the first grenade in 1.10 seconds, and each subsequent sticky takes .66 seconds. The time to reload a full clip is:
1.10 + (7 * .66 ) = 6.38 seconds
It's possible for us to "nerf" the sticky bomb launcher by giving it a higher base reload time without affecting the amount of time it takes to fill an entire clip:
1.30 + (7 * .64) = 6.38 seconds
A reload time of 1.3 doesn't kill the sticky bomb launcher's usefulness as a rocket launcher replacement, but it does make it more difficult for the demoman to keep his clip full enough during combat to use it as an assault weapon during prolonged engagements or against large groups of enemies. On the other hand, a demoman who uses his sticky bomb launcher to lay traps and guard choke points will still take the same amount of time to fill his sticky bomb launcher (assuming he lays eight stickies at a time before refilling).
This is just one example of how Valve could "fix" the demoman. Note that while I did provide a method for weakening the sticky bomb launcher as a direct combat weapon, the same mechanic (reload time) to strengthen the grenade launcher as a combat weapon. If the demoman receives a nerf to one weapon but an improvement to another it shouldn't drastically alter the balance of the game. And remember, we're not messing with things like the weapons' range or damage, but the simple mechanic of reload time, which is less likely to be a factor in one-on-one situations and more likely to impact the demoman's decisions when fighting a series of engagements against multiple foes.
(Acknowledgement/Disclaimer: I'd like to thank TF2wiki.net and the people who maintain it for providing the weapon statistics used in the illustration above. On the same note, I also apologize for the fact that some of the statistics I parrot are probably inaccurate. Many of the points made above still hold even if the numbers supporting them are slightly off.)
Heavy
The heavy is currently the least-played class in the game. This isn't a recent phenomenon; the only time the heavy saw any real popularity was immediately after his class update. I'm strongly convinced that the lack of heavies deployed on public servers points to an issue of "fun" rather than "balance." The European competitive scene has demonstrated that it is possible for the heavy to be a useful class (albeit one with less flexibility than the soldier). The main issue is that people simply do not enjoy playing heavy.
One major issue that makes it less fun to play the heavy is that he moves quite slowly, making the process of getting from place to place quite laborious, especially compared to other classes. (The only classes other than heavy that have lower-than-average running speed are the soldier and demomen, and they have propulsion techniques that more than make up for their slow running speed.) One existing relationship that the heavy has with speed is that he can sacrifice speed for weapon response time by spinning up his minigun. The heavy can bypass one-second windup of the minigun, but it comes at the cost of his mobility which is significantly reduced when the minigun is spun up. What if the exchange could work in the other direction, allowing the heavy to gain speed at the expensive of a higher windup time? Putting it into practice, when the heavy runs around with his melee weapon out, the minigun has an effective wind-up time of (roughly) two seconds: it takes about one second to switch to the minigun, and another second to spin it up. In exchange for this added minigun latency, let the heavy run faster when he has his fists/gloves out. It would make the heavy slightly less boring to play (especially during the initial march to the middle point on push maps) and although it's a buff, I don't think we need to worry about upsetting balance when the heavy is the least-used class in the game.
One issue that the heavy suffers from is his reliance on a medic. Valve stated this prior to their release of the heavy update. The heavy gains two things from a medic: In combat, the medigun effectively reduces damage against him by 24 DPS. Out of combat, the medigun enables the heavy to run around with 450 HP. Without the medic, he loses both of these defensive boons, and unfortunately he gains neither with the sandvich. The sandvich is more of a medkit replacement than a medic replacement. If Valve really wants the heavy to be less medic-reliant, they need to find a better way to emulate the function of the medigun. If they want to do it by modifying an existing weapon, allowing the sandvich to buff the heavy would be a good place to start. Another option that Valve has is to give the heavy a new weapon. They could start from scratch, or revisit an old idea. Ludmilla was a scrapped unlockable idea that replaced the minigun's secondary fire with 75% damage and 25% healing. If Valve cooked up a new weapon for the heavy that enhanced gameplay and didn't introduce any serious balance issues, I don't think anyone would mind horribly if the heavy received a new weapon bundled with the next class pack as a freebie (give it to everyone without forcing them to unlock it).
Engineer
As a player who has spent many hours playing the engineer in the context of a public server, I have this to share: The amount of fun I have playing the class is in many cases inversely proportional to the level of my sentry. The 2fort basement and places like it are some of the most boring places to maintain a sentry, despite the"effectiveness" of a level 3 sentry in such a location. Plopping a level 3 sentry on a less-trafficked portion of Dustbowl and getting most of your points off teleports is seldom entertaining. Spending your time on the front lines, surprising foes with 1 sentries you while ruthlessly shotgun down enemies is an incredibly rewarding way to play the engineer.
Summarizing the engineer: building sentries is fun. Maintaining sentries is boring. In the interest of making the engineer more fun to play, I'd like to see unlockables that favor building, rather than maintenance. One suggestion I've heard suggested more than once is a wrench that halves build time but doubles the amount of time it takes to upgrade or repair a gun. Another idea is to give engineers the option of having buildings that can be constructed twice as fast but have half as much HP.
The engineer is already the king of static defense; he doesn't need to get any better at it. Giving the engineer more mobile options should make the game more interesting for both the engineer and non-engineer classes.
You may notice that I didn't to go into detail about a number of the game's classes. I really have nothing to say about them other than that they are wonderful in their current state and Valve would be hard-pressed to do a better job with them than they already have. I'm not going to pretend that they're perfect; as an example, the game probably could do without the Razorback. However, when adding new weapons, it's probably best to err on the side of uselessness (see the Razorback) than to release a potential fun-killer like the Sandman.
Conclusion
Truthfully, the aim of this post is advocacy, not education. Chances are that you won't gain a whole lot of factual knowledge from reading it, as it's mostly analysis with very little data presentation. However, it may change the way you think about Team Fortress 2 as a game, and that's really all I could reasonably expect.
If you want to discuss this post, head over to the forums and share any comments, criticisms, or other thoughts.
Inspired by 18 pages on TF2 by Chro of TF2F.
