Matchmaking revisited

The TF2 team seems to have won back a few hearts and spleens with the latest major update, and hearty congratulations to them. They do, however, seem dedicated to maintaining their image as somewhat behind the times, rather like a dotty old aunt still supporting the Carter administration. Case in point: matchmaking, which went from nonexistent to hilariously bad with the previous major update. Some notes, then, on the problem and possible fixes:

Fixing the wrong problem

The matchmaking system took as its central tenets two points that must have been prestidigitated from one of the dev team’s less wholesome orifices:

  1. Thou shalt not be forced to attack/defend twice in a row. (Also, thou mayest strike whichever doth not apply.)
  2. Thou shalt not be autobalanced.

Nearly every problem with matchmaking is a consequence of either or both of these axioms.

Balance problems aggregate over time

TF2 of yore had a system of (somewhat) occasional auto-shuffling: If a team failed spectacularly on its turn, teams would be shuffled before the next round. There would be switching after the fact, of course, but this is nicely preempted in the current system, by disallowing manual switching. Losing streaks were thus occasionally broken up.

The current system, however, practically rewards teams for doing too well. Winning teams retain more players, thus keeping more of whichever players were helpful, while losing teams tend to lose whichever members are savvy enough to grasp how bad they are while being (too slowly) backfilled by whatever lunkheads are low enough priority to slot into an ongoing game. Both will regress, over time, to the mean, but the time taken can be quite sufficient to ruin an evening’s play. It doesn’t help that, subsequent to Tenet the first, above, teams carry across server resets.

Autobalance is necessary

Nobody likes to be autobalanced. Under the old system, you would invariably be autobalanced into a team of five snipers, three spies and zero self-awareness. Autobalance doesn’t need to be a chore, however, if the system actually works.

As outlined above, backfilling from the global queue just does not work for the losing team. The late joiners will tend toward the average, while the winning team already tends toward better than average (not necessarily by much). Autobalancing a player from the other team, then backfilling either one on an equal footing, is much more likely to even the balance.

Teams might profitably be pre-assorted

If, once more, we disregard Tenet the first, above, we might even do better. Casual mode already has a useful proxy for the skill of the player: their so-called XP. (Note that, as an erstwhile role-player, I object to the use of this terminology outside the context of an actually improving character.) TF2 might even keep track of an Elo-like rating “under the hood” (useful, as well, for seeding Competitive mode), though the dev team might object to hidden variables.

Whichever rating one uses, applying it to matchmaking would smooth out at least some of the rather large disparities in skill between teams. If teams start out closer in skill, the mid-game might run smoother and require fewer adjustments. At least one might be spared the occasional “Benny Hill”-like embarrassing rout.

Mid-round quitting should be curbed

Although there are obvious exceptions (quitting immediately after joining springs to mind; matchmaking has a nasty habit of dropping one into unwinnable states), which should be catalogued and addressed individually, mid-round quits diminish the game. Losing a round is, after all, part of the game, too, and solid players should stick with their team ’till the bitter end.

To this end, I propose that a “flakiness index”, or FiX, be kept on players. Staying for whole rounds decreases the FiX; quitting mid-round increases it. Having a high FiX wouldn’t necessarily decrease a player’s matchmaking priority, but might usefully be employed to steer flaky players toward each other, thus keeping them out of solid players’ increasingly thinning hair.

Idling should be punished

The irritation caused by idling on one’s team is simply difficult to overstate; thoughts inevitably go toward Chinese water torture and extended bagpipe music sessions. The current idle timeout is a preposterously generous five minutes or so, and even sneezing in the proximity of the mouse is enough to forestall disconnects. Additionally, vote-kicking idlers is made more difficult by many players reflexively F2-ing—probably because of abuse idle-kicking.

Balancing the rest of matchmaking is probably impossible without addressing the problem of idling. Something along the lines of four spawntimes would be more reasonable than five minutes, and some criterion for determining actual contribution should be though up, as well. Exiting spawn would seem a reasonable minimum, except for the inevitable in-spawn sniping and healing, and one wouldn’t want to get this wrong.

To effectively prevent idling, it should carry at least some matchmaking penalty, as well. As priority penalties are rather short-lived and easily circumvented by simply—yes—idling some more, recurrent idlers might be flagged. Indeed, idling might beneficially be combined with the FiX described above.

Class limits might at least be acknowledged

We have all seen the five-sniper push, the four-spy jink and the six-scout stampede, and we all hate to be on the same team as one. If the TF2 team had five minutes to spare, they might at least add a pop-up warning when the fifth player tries to go sniper. If they felt really generous, they might add “Bye, 007 Felicia” as a vote-kick reason.

Other candidates include fourth medic (who are you healing, exactly?) and sixth engineer, perhaps excluding defense.

Pre-mades might need to be sacrificed

It may prove impossible to implement all this and keep premades together. This isn’t entirely a disaster, as many enjoy murdering their friends in interesting ways as much as, if not more than, collaborating with them. In addition, premades are, for an endless number of reasons, usually disproportionately effective, thus dominating casual matches, as well as less than helpful to outsiders on the same nominal team. “Casual” simply isn’t an appropriate descriptor for a six-person murder squad, and they would be infinitely better served on a server dedicated to that sort of thing.

The even less excusable

And finally, iron out some baffling deficiencies:

  1. Rounds starting before all players are loaded.
  2. Backfilling one team before the other.
  3. Pre-match countdowns resetting to 120.

FPS playerbases are notoriously unpleasable, and TF2 not so far from the norm in this regard. However, real, working matchmaking has been a staple of online shooters for a decade at this point. Demanding parity shouldn’t be unreasonable.

Then you can go on to rebalancing the Pyro.

Leave a Reply

Your email address will not be published. Required fields are marked *