Pricing a win: floors, clamps, and the unit you count in

Make every win worth something without turning the ladder into a slot machine.

The Elo and uncertaintypages explain why a strong player's win paid nothing. This one is the design decision: given that payout = K · (S − E), what do you change so a win always feels like progress — and what does each lever cost you?

1 · The three levers

Clamp the expected score. Cap E so a near-certain win can never be priced at ~0. Symmetric clamp: E ∈ [1−cap, cap]. A 0.90 cap makes any win at least K·0.1.
Change K. Scales every payout up or down, but also makes ratings (and tier movement) more volatile. The blunt instrument.
Change the unit. Multiply the whole ladder — seeds, payouts, tier edges — by 10. Cosmetic to the maths, but +3 becomes +30 and the number feels worth chasing.

The distinction that matters. Scaling the whole ladder ×10 is a denomination change: every ratio is identical, tier bands are 10× wider, nothing moves faster. Scaling only the payout ×10 is really "raise K ×10": tiers fly by and cold streaks hurt ten times as much. Pick deliberately.

2 · Design a payout

Design a payout

Your rating · 1720

Opponent average · 1000

K-factor · 32

Expected-score clamp

Unit scale

Raw E = 0.984, clamped E = 0.900

Win

+3

Draw

-13

Loss

-29

current Convocados

clamp only

Rank Points

Guaranteed minimum for a win at the current K and clamp: +3 pts. At 1720 vs 1000 with K=32, "none" pays 0; the 0.90 clamp lifts the win; ×10 turns the whole ladder into Rank Points and the floor becomes worth chasing.

At the Convocados case (1720 vs 1000, K=32), watch the win cell:none pays 0; the 0.90 clamp lifts it;×10 turns the whole ladder into Rank Points and the floor becomes a satisfying +30 RP.

3 · What a floor costs

A guaranteed payout is not free. If a win always pays, a strong player who keeps playing weak fields accrues points for expected results — that is farming. Mitigations:

Prefer a clamp over a flat floor: it preserves the shape of the curve (even games still pay the most) and only bounds the extremes.
Keep soft resets so farming within one Season doesn't compound across Seasons.
Price tier edges from the seeded distribution, so a +30 RP floor is small against a realistic seed spread — movement stays honest.

This is the same instinct Josh Menke argues for in his GDC talk: the rating should be fair, but the rank is a player-experience object, and an experience that shows "+0" after a win is a bug in the experience, not the maths.

Check yourself

A favourite's win rounds to zero. The cleanest fix is to…

Bound the expected score. Clamping E keeps the curve's shape and only removes the near-zero tail.

Scaling the whole ladder ×10 changes only…

How large it reads. Whole-ladder scaling is a denomination change; every ratio is untouched.

The main risk of a guaranteed minimum payout is…

Farming weak opponents. A floor pays for expected wins, so weak fields become a points source.

Primary sources

For the design layer, watchJosh Menke's GDC talk on skill, matchmaking and ranking— especially the "Ranking Systems" and "Hybrid Systems" chapters. For the maths of the levers, theElo article's K-factor sectionis enough.