Ranking Decay and Schedule Planning in a Tennis Career

The first playable version of our tennis career had no decay. Points accumulated. You entered tournaments, you won matches, the number went up, and the number never went down.

It took about four hours of testing to notice the problem, and considerably longer to admit what it was. The ranking was not a ranking. It was a progress bar with a tennis theme.

Why a cumulative total kills a career game

A number that only rises cannot carry a decision, because every option is non-negative. Play a tournament, gain something. Play a smaller one, gain less. Skip it, gain nothing. There is no branch anywhere in that tree where a reasonable player would hesitate.

Worse, it makes the second half of a career inert. Once the total is large, individual results stop moving it in any way you can feel, so the exact period when a real career becomes most interesting — when a player is defending a position rather than building one — becomes the period when the game has the least to say.

Real tennis rankings do not work that way and have not for decades. Points are dropped 52 weeks after being awarded, and the table is recalculated weekly. A player's ranking is a description of the last year, not of their life.

Putting that into the game was the single change that turned a scoreboard into a career.

Three decay shapes, and the one we shipped

We built and tested three.

Cliff decay. Points sit at full value for a fixed period and then vanish entirely on a set date. This is closest to how the real system behaves and it is the version we shipped.

Linear taper. Points lose a fixed fraction of their value each in-game week, fading to zero over the window. Smoother, arguably more elegant, and rejected.

Half-life decay. Points lose a proportion of their remaining value each week, so they decline steeply then trail off. Rejected faster than the taper.

The taper and the half-life were rejected for the same reason, and it is not the reason we expected. Both were better behaved mathematically — no discontinuities, no bad weeks caused by the calendar. But neither was legible. A player could not look at their ranking and answer "what am I defending, and when?" because under continuous decay you are always defending everything a little bit and nothing in particular.

Cliff decay is lumpier and occasionally feels unfair. It also produces a sentence a player can say out loud: in six weeks I lose the points from that semi-final, so I need a result before then. That sentence is the game. We took the ugly model that produced it over the elegant models that did not.

What changed in testing once points expired

The behaviour change was immediate and larger than anything else we have added to that build.

Before decay, testers picked tournaments by size. The biggest event they could enter, every time, because there was no cost to anything.

After decay, three habits appeared within a session or two:

• They started reading the calendar backwards. Sessions began with a look at what was about to expire rather than at what was available.
• They began skipping big events deliberately. A large tournament in a week where they had nothing to defend became less attractive than a medium one in a week where they had a lot to defend, which is exactly the trade real scheduling involves.
• They started talking about seasons in pairs. "This season" stopped being the unit. The unit became this season and the shadow last season casts over it.

None of that was taught by a tutorial. It emerged from one rule.

There was a fourth effect we liked less: a subset of testers found the first expiry genuinely demoralising, because it arrived without warning in a career that had been going well. We added a visible expiry schedule in response — you can see what is due to come off and when. We did not soften the mechanic.