How We Grade TFT Comps: Tiers, Avg Place and Win Share

What counts as a comp board

We only count boards that look like a real team: the player reached level 7 or higher and ended the game with at least 7 units on the board. Players knocked out early, before they built a team, are left out.

That is why the average comp board places better than 4.5. Across an eight-player lobby the average place is always 4.5, but the players who go out first rarely reach level 7 with a full board. Most of the bottom places drop out of the data, so the boards that remain place better than 4.5 on average.

Every board is the team the player had when their game ended. A 1st-place board is what the winner had at the very end; a board that went out in 6th is what that player had when they were knocked out.

How we find the comps

To find which comps people play, we look at how boards end up. Boards with exactly the same units are counted together, and for this step a unit list seen fewer than 2 times is left out. Then boards are grouped: starting from the most common board, every board that shares at least 6 units with it joins that group, and the next most common board left over starts the next one.

A unit is core when it is on more than 50% of the group's boards (30% for 5-cost units), and flex when it is on more than 15%. A group becomes a comp with at least 100 boards and an average place better than 5.00. Groups built around the same carry and top trait, or the same reroll unit, are listed as one comp; open a row to see its variations.

This step only decides which comps exist, their units, items and carry. It does not decide their numbers.

Which boards count for a comp

Finished boards flatter a comp. A player who lasted long enough to add the last expensive units usually placed well, while a board that was knocked out halfway through looks different and would not join the group above. Measuring a comp only on the boards it was found from leaves out most of the games where it went badly.

So a comp's numbers come from every comp board that was playing it, finished or not. A board counts for a comp when it has the pieces you commit to early, never the late ones:

the comp's carry holds at least 2 completed items, and the comp's main trait is active (at its first level or higher);
for a reroll comp: the rerolled unit is at least 2-star, and the comp's main trait is active when it has one.

A board that fits several comps counts once, for the comp whose usual board it shares the most units with. Inside a comp, each board counts for the variation it is closest to. A board that fits no comp still counts in the scope's totals (the average comp board, Play and Win Share), just not for any comp.

Every number on a comp comes from these boards: Avg, Top 4, Win %, Play, Win Share, Boards, how often each item is built and the top 4 vs bottom 4 comparison.

The stats on each comp

Avg (adjusted average place): The comp's average final place, adjusted for sample size (next section). Lower is better. The list is sorted by it and the tier comes from it.
Raw Avg: The plain average place of the comp's boards, with no adjustment.
Top 4: How often the comp's boards finished 1st to 4th.
Win %: How often the comp's boards finished 1st: its wins divided by its own boards.
Win Share: The comp's share of all 1st places: its wins divided by the wins of every comp board in the scope. Popular comps have a bigger win share simply because more people play them, so it says how often a comp wins lobbies, not how strong it is.
Play: The comp's boards divided by every comp board in the scope.
Boards: How many boards counted for the comp (see "Which boards count for a comp" above): its sample size.

Why the average is adjusted

A comp seen on a few hundred boards can look great or terrible by luck alone. So we pull every comp's average toward the scope's average comp board, as if the comp had 200 extra boards that finished exactly average. A comp with 200 boards ends up halfway between its own average and the scope average; a comp with 2,000 boards keeps 91% of its own.

Example: Yunara Reroll has 700 boards with a raw average of 4.63. It keeps 78% of its own result, so its adjusted average is 4.58.

The adjustment keeps a lucky small comp from topping the list. It does not make a small sample certain: a comp with few boards can still move a lot from one day to the next.

The scope average

The patch, rank and time window you pick together make a scope. Each scope has its own average comp board: the average place of all its comp boards, the same boards Play and Win Share divide by. Comps are graded against that number, so a grade compares a comp with the other boards from the same patch, rank and time window.

On Patch 18.5 · Diamond+ · 181.5k boards it is 4.39. The Comps page shows it above the list on wide screens.

Tiers

The tier compares a comp's adjusted average place with the scope average. Lower places are better, so "better" means a lower number.

Tier	Avg vs the scope average	Here	Comps
S	more than 0.40 better	under 3.99	3
A	0.20 to 0.40 better	under 4.19	5
B	between 0.20 better and 0.10 worse	under 4.49	14
C	0.10 worse or more	4.49 or worse	14

"Here" and "Comps" are for Patch 18.5 · Diamond+ · 181.5k boards, all 36 comps counted, hidden ones included.

Hidden comps and "Show all"

Some comps are hidden until you tap "Show all": comps played on fewer than 0.3% of the comp boards in the scope, and comps whose adjusted average is 0.25 or more worse than the scope average (4.64 or worse on Patch 18.5 · Diamond+ · 181.5k boards, where 5 of 36 comps are hidden). They are real comps, just rare or weak in that scope.

Patch, rank and time

Patch picks the game version. Rank keeps only boards from players at that rank or higher, using the rank they had when we collected the match. Time keeps only games from the last few days of the patch; "All time" uses the whole patch. Every combination is its own scope with its own average comp board, so the same comp can get a different tier in a different scope.

What the numbers can't tell you

These are results, not causes. Boards that win often carry more expensive units and better items because their players lasted longer, not only because those units won the game. Match data has no unit positions, so nothing here says where to place units.

How we grade TFT comps