A guide to the numbers
What does a good result actually tell you?
A high finish rate is a reason to investigate a team. It is not a promise about your next run.
Read one real Comp row
Miasmaw + Drumire
Worked example from one dated public sample. Observed ; 30-day sample.
- 10-win finish
- 37 / 101 = 36.6%
- Observed runs
- 152
- Players
- 126
- Play rate
- 3.1%
The finish denominator counts terminal observations, not every run in the group. Play rate uses the eligible board sample instead.
- Read the denominator before the percentage. With few observations, one extra result can move the percentage substantially. A larger sample can still be biased. Neither is a controlled experiment.
- Check players as well as runs. One player can contribute several runs. More rows do not necessarily mean more independent evidence.
- Separate popularity from success. Play rate describes representation in this sample. Finish rate describes outcomes among terminal observations.
- Do not reverse-engineer the tier from the displayed raw rate. The tier score balances players and shrinks small samples toward the baseline. Two groups can have different raw rates and a different score order.
The terminal-observation rule in this calculation is not the game's life-loss rule. Read the calculation method and qualification limits.
Compare like with like
Before comparing teams
- Use the same requested window and rank band.
- Check the effective window: a small 30-day sample can widen to 90 days.
- Read the observation date, not just the page's current date.
- Look for missing history or patch attribution notes.
When a number is absent
Unavailable is not zero. A missing phase, trainer observation, or rate does not prove that the choice fails.
Keep the missing field unresolved. Use a documented reference rule for a planning decision instead of inventing a statistic.
Public histories are selective. Skill, opponents, earlier choices, and which runs are visible can all affect a group. A successful companion unit may be a consequence of reaching later rounds, not the cause.
Meta uses another denominator
Entity and pair observations can repeat across rounds. Trainer observations describe stored public runs with a reported result. They are not shop offers or pick rates.
The Meta tier list ranks raw observed percentages within each category. Do not compare a trainer's tier directly with a Batomon's tier, or treat a pair's rate as measured synergy.
Search and tier filters change which entries you see. They do not assign fresh tiers to the remaining entries. The display floor is not a statistical reliability guarantee.
Compare Meta entries →Turn the observation into a question
Open the Comp's phases. Ask which units recur, which alternatives appear, and whether the printed abilities support the interaction you want to try.
Then make an editable plan. Change one decision at a time and review your own completed runs. This does not isolate causality, but it makes your next question more concrete.