Which Weapons and Items Do Players Really Use?
Compare how often each weapon or item was picked with how often it was on offer, and only then look at how well it does. Pick counts settle quickly because every player adds to them, while win rates for a rarely picked weapon stay a blur for a long time.
What do pick counts tell you that win rates cannot?
Iron Pasture is an invented farm defence game, and these numbers are made up for this example. In each of 400 runs, the player picks one of four weapons at the start. All four are always on offer.
| Weapon | Picked | Share of offers | Runs won | Win rate | Very likely between |
|---|---|---|---|---|---|
| Pitchfork | 251 | 63% | 128 | 51% | 43% and 59% |
| Shears | 96 | 24% | 53 | 55% | 43% and 67% |
| Bell | 41 | 10% | 19 | 46% | 29% and 65% |
| Rusty Hook | 12 | 3% | 9 | 75% | 40% and 93% |
The Hook jumps out: 75%, the best in the game, and nobody uses it. It is tempting to call it a hidden monster. But that 75% rests on 12 runs, and the true win rate could be anywhere from 40% to 93%. All four ranges overlap, so as far as win rates go, this table says nothing yet.
The pick share is a different kind of number. It rests on all 400 offers. The Hook was passed over 388 times out of 400, and that is solid. So the finding here is not "the Hook is strong". The finding is "almost nobody picks the Hook, and we do not know why".
The 12 who chose it may also be your most skilled players, so that rate may say more about them than about the Hook.
The Pitchfork at 63% needs a second look as well. It is the first in the list. If you always show it first, you cannot tell popularity from convenience. Shuffle the order per run and count again.
Why does an item go unused?
There are three common causes, and each one looks different in the data.
| Cause | What you see | What to try |
|---|---|---|
| Nobody knows what it does | Few picks, win rate fine or unknown | A clearer description, or a moment that shows it |
| It is simply bad | Few picks, low win rate over many plays | Better numbers, or remove it |
| It is a pain to use | Picked or owned, then rarely used | Fewer steps to use it |
The Hook turned out to be the first. Its card said "2 less damage", and never said that it pulls enemies towards you.
Items have a version of this. Every run hands out three bombs, and 244 of the 400 runs ended with at least one unused. That is 61%, very likely between 56% and 66%. The maker opened the game and counted: using a bomb takes two menu steps in the middle of a fight. So he moved bombs to a single button.
Change only one option at a time, and read the win rate check before you touch any damage numbers. Our win rate recipe takes a list of plays, with what was on offer, and reports both numbers.
Paste this to your AI
Show me what players really use in my game. The event log is in [file or folder].
1. List every weapon, item or ability a player can choose. Draw the moment of choosing and the moment of using each one as a Mermaid state diagram, with how many clicks or steps it takes, read from the code.
2. Check whether the log records what was on offer, not only what was picked. If it does not, add that first and tell me, since a pick share needs it.
3. Use get_recipe with win-rate-check if Spielerando is connected. For each option, show: times offered, times picked, share, plays, wins, win rate, and the 95% ranges for both shares.
4. Say which option's numbers are solid and which rest on too few plays to judge. Do not call anything strong or weak from a win rate with a wide range.
5. For each rarely used option, say which fits best: nobody knows what it does, it is bad, or it is a pain to use. Show what points that way, and say when you cannot tell.
Also tell me if the options are always shown in the same order. Suggest one change only.
Or skip this and connect the MCP: your AI then finds all of this by itself.Connect the MCP
How to check it worked
- Count an offer by hand. Play one run and check that all the options were logged.
- Use the item in a hard fight. Count the button presses.
- Shuffle the list for a day and see whether the favourite stays the favourite.
- Compare after a change. Did the pick share move, and did anything else fall?