Nice Article on TrueSkill Related to Halo Reach

There are jobs that bring you face to face with mankind at its ugliest: cops, social workers, garbage men. Add video game engineers to that list. At Bungie, the studio behind the Halo games, including the upcoming Halo: Reach (available exclusively for the Xbox 360 on Sept. 14th), the developers who work on competitive multiplayer have seen it all. There are the cowards; the players who pull off a streak of successful matches and then retire to keep their standings at the top of the leaderboards. Some players even pay others to repeatedly take the fall, artificially boosting their rank. More common are the virtual psychopaths—the type who toss grenades at friend and foe alike, or callously bash in a teammate's head to steal his rocket launcher. While most players aren't trying to cheat their own video game, a solid minority are always hunting for exploits, creating an arms race that pits dozens of developers against an insurgency of thousands.

For its last hurrah in the Halo universe, Bungie is trying to clean up the mess that it helped create—after all, Doom may have invented the deathmatch, but it was Halo 2 that brought online multiplayer gaming to the larger console market. Halo 2 became a reason for gamers to buy subscriptions for Microsoft's Xbox Live online network, while also spawning an awful new gameplay reality: competitive matches dominated by trash-talking teen and tween-age players, who slaughter "noobs" with a range of dirty tactics, such as setting up camp where players enter the game, and picking them off the instant they appear. Halo 3 was designed to break up these one-sided turkey shoots, with the help of Microsoft's new (at the time) matchmaking system, TrueSkill. After playing a relatively small number of matches, players were assigned a skill rating between 1 and 50, and sent up against similarly-rated enemies. The goal was for a given player or team to average an even split of wins and losses, making for closer, more competitive games, and fewer massacres. As Bungie soon found out, TrueSkill was a brilliant set of algorithms, and, as Luke Timmins, Engineering Lead on Halo: Reach put it, "a marvelous, efficient way to sort populations based on skill." Unfortunately, it also created brand new exploits, inspired a rash of unsavory behavior, and, in some cases, quantified just how bad some players were. For Halo: Reach to have the kind of online longevity that its predecessors have had—Bungie hopes it will be played for another five years, at least—the studio's engineers had to pick up where Halo 3 and TrueSkill left off, and face some of the ugliest truths about the often-ugly world of multiplayer gaming.
 

The Truth Hurts

Bungie has nothing but respect for TrueSkill. During my visit to the studio's offices in Kirkland, WA, a trio of engineers described Microsoft's set of matchmaking algorithms with an almost spiritual reverence, particularly when it comes to TrueSkill's ability to begin nailing down a user's skill after as few as 10 games. In fact, for most players, TrueSkill was a little too good. "The only people who want to be told what their skill level is are people with a very high skill level," said Senior Designer Tyson Green. If TrueSkill said you were a 32, that was usually that. After another 100 matches, that number might budge by a point or two, if at all. It might even inch down. Bungie's answer for Halo 3 was to tack on an experience point system, which tried to help distract less fortunate players from Microsoft's brutal, quantitative honesty. "With Halo 3 we took that beautiful engine and poured sand into it, dampening TrueSkill, to create the appearance of progression," Timmins says.

For Reach, the engineering team is still using TrueSkill to match up players, but they've turned rankings into an ongoing competition. There are tons of different multiplayer game modes available, but only those players who fight in the straightforward Arena matches (which can be free-for-alls, or team vs. team affairs) are ranked. And rankings earned there last for a season, which is one month long. When each season ends, stats are archived, but rankings are wiped, and players must repeat the process of qualifying for a division (there are five, ranging from Steel to Onyx) and competing for the top scores. So a player who pays for a boosted ranking would have to do so quickly, and on a regular basis. Likewise, anyone who rockets to the top and then quits will have their bragging rights buried in the archive.

The seasonal rankings also provide something more ephemeral than raw, semi-permanent TrueSkill ratings: hope. "It's giving you more chances to win," Green says. "Every season is a new chance of being rated in a newer, better division, or at least better than your friends." TrueSkill being the robust system that it is, not to mention the Arena algorithm that Bungie layered on top of it (more on that in a second), chances are you'll wind up finishing the next season in roughly the same spot as the last. But as lies go, it's an elegant one.

Your Code Will Be Cracked

Soon after

was launched, TrueSkill's troublesome accuracy took a hit, as

players paid, cheated or simply played dirty to hike their skill ratings, often at the expense of anyone unlucky enough to be teamed up with them. Through crowd-sourced trial-and-error, dedicated players reverse-engineered a good deal of the game's matchmaking system, and developed savant-like playing styles that usually favored lone wolves. For

, Bungie created its own algorithm to help determine rankings in Arena, which emphasizes team play. And the studio released the formula online, even explaining the more advanced mathematical elements in a blog post. This was on April 30th, just before the

Beta went live, allowing players to try out an unfinished version of the multiplayer portion of the game months before the September release.

The posting of the Arena formula was an open challenge to players to utterly abuse it, to tailor their gameplay to the code and burn through the rankings during the 17-day beta period. According to Green, the equation survived intact. Despite elements that seemed risky, like downplaying deaths (a player has to die three times to lose the points gained with a single kill), when

comes out next month, the formula will be unchanged. Naturally, there are scattered complaints in blog comments and on message boards about the team-oriented elements—the high price of betrayals, a.k.a. friendly fire, is maybe the biggest gripe. But Bungie believes that the formula did its job, discouraging grenade-spamming and antisocial antics, while still harnessing TrueSkill's spooky powers of population sorting. "It fixes the shortcomings of

, but still does justice to the Ferrari engine that is TrueSkill," Timmins says.

Skill Is Overrated

According to Green, Reach's most lasting contribution to multiplayer gaming might have nothing to do with the delicate, borderline psychic math that assesses player talent. Skill level is crucial to setting up competitive matches, but as developers add more cooperative modes to online shooters, other factors take precedence. In Reach you can specify whether you want to be teamed up with similarly chatty, "boisterous," lone wolves, or with quiet, polite, team players. Meanwhile, Bungie is crunching its own numbers behind the scenes—if you're searching for someone to play through one of the story-based missions, for example, the system might favor players who haven't dropped out of cooperative sessions midway through. The result, in theory, is a kind of eHarmony for Halo players, creating teams that mesh well, whether matched-up members are going to fight computer-controlled aliens in modes like Firefight, or wade into the Arena. This is especially important for games that are hoping to open up multiplayer to older gamers, who may have been turned off by punitive brushes with competitive matches in the past, and who don't have the array of always-on friends to team up with. "I could be matched into a Firefight game where everyone is exactly equal to me in skill, but it's no fun, because maybe they're trash talking all the time, or they think it's fun to backstab all the time," Green says.

Skill, and TrueSkill, will still play a major role in lining up teams to fight each other. And to avoid potential exploits, players won't be able to specify the characteristics of opponents (no "randomly" going up against the one other guy online who speaks Tongan). But with much of the gaming industry embracing cooperative multiplayer as an even bigger draw than hardcore-oriented competitive modes, Reach could set an example for how to make online gaming an inclusive party, instead of a virtual Fight Club for tweens. "Especially in the cooperative world, we need to evolve the state of the art," Green says. "There are more ways of matching people up than skill. All of that is secondary. Really, it's about finding other players you can have fun with. "

www.popularmechanics.com/technology/digital/gaming/halo-reach-preview-1955640