Here’s an uncomfortable claim: most of the item win rate stats you check before locking in a build are close to meaningless, and trusting them is quietly capping your climb. Not because the sites are wrong about the numbers — because the numbers answer a different question than the one you’re asking. A page that says an item wins 54% of games isn’t telling you the item is strong. It’s telling you the item tends to get bought by players who were already winning. That distinction is the entire reason your win rate stalls even when you copy the “best” build every game.
The Number That Sold You on a “54% Win Rate” Item
Picture a stat page showing Passive: Gain bonus max HP based on your current HP, stacking over time.
Shen Shen top buying 
Heartsteel Heartsteel 3000 gold
A data-driven breakdown by analyst Tomas Svojanovsky dug into that specific case — Heartsteel on Shen — and found the raw number was doing something sneakier than advertising item power. It was capturing when and by whom the item tends to get bought, not what the item does for an average game. Players who are already stomping their lane buy more legendary items faster, snowball harder, and finish games in a way that inflates whatever they happened to complete first. The item didn’t win the game. The winning game bought the item.
This isn’t a Shen-specific quirk. It’s how raw win rate works for every item on every stat site you use, including the ones behind your own in-game build suggestions. The number is real. What it measures is not what you think it measures.
What Selection Bias Actually Means
Selection bias, in this context, is simple: an item’s win rate is contaminated by the reasons players chose to buy it in the first place. Nobody selects items at random. You buy Passive: Gain bonus max HP based on your current HP, stacking over time. +30% Armor Penetration. Passive: Abilities slow targets by 30%. Heartsteel Heartsteel 3000 gold 
Serylda's Grudge Serylda's Grudge 3000 gold
Statisticians call the resulting inflation the “win-more” effect: a metric that looks predictive but is actually just describing situations that were already trending toward a win. It’s the same trap as saying “teams that get a Baron kill win 78% of the time,” without acknowledging that teams ahead enough to safely take Baron were already winning before they touched it.
- A raw win rate treats every purchase as equivalent, whether it happened at 2-0 with a full wave crashed or at 0-3 as a desperation buy.
- It doesn’t separate “this item caused the win” from “this item is what winning players happen to buy.”
- It can’t tell you what would have happened if the same player, in the same losing game, had bought something else instead.
None of this means item stat pages are useless — they’re a real signal about what’s popular and generally functional in the current patch. It means treating the top number as gospel, without asking what state the buyer was in, is how a “good” build stops correlating with your actual results.
Purchase Win Probability — The Metric Stat Sites Hide
There’s a fix for the win-more problem, and it already exists: purchase win probability. Instead of asking “what percentage of games with this item were won,” it asks “what was this player’s estimated chance of winning at the exact moment they bought it.” Svojanovsky’s breakdown applied this to the same Heartsteel-on-Shen case and found the purchase win probability sat at roughly 53.66% — noticeably closer to a coin flip than the flashy 54.15% headline number, once you strip out the fact that the buyer was often already favored to win.
The gap between those two numbers is the selection bias, quantified. A related metric, win probability added, goes a step further and measures how much a team’s odds actually shifted between the purchase and the end of the game — a much more honest read on whether the item did anything.
| Metric | What it actually measures | Where it misleads |
|---|---|---|
| Raw win rate | % of games won when item was bought | Mixes in game state at purchase time |
| Purchase win probability | Estimated win odds at the moment of purchase | Requires a live model — most public stat pages don’t show it |
| Win probability added | Change in win odds from purchase to game end | Rare, but the closest thing to “did this item help” |
Most public stat pages you check mid-draft don’t expose purchase win probability — they show the raw number because it’s simpler to compute and reads better as a headline. That’s not malicious, but it means the tool you’re actually using to make a decision is the least reliable of the three.
Timing Beats Choice: The Real Predictor of a Stalled Win Rate
If raw win rate is noisy, what actually correlates with winning more consistently? Hitting your power spike on schedule. On patch 26.13, tracked data from 
Yorick Yorick top ‘s Serylda’s Grudge into Spear of Shojin into Trinity Force path shows a 52.9% win rate across 227 tracked matches, per build-tracking site MetaBot.GG. That’s a solid, unremarkable-looking number for a specific three-item combination — but the more useful part of that data isn’t which three items were involved, it’s that the build represents a complete power spike, and completing it changes what Yorick can do in a fight rather than nudging a stat up by a few percentage points.
Every legendary item you finish is a spike: a moment where your kill threat, survivability, or wave clear jumps a tier. Season 2026’s itemization changes — the crit damage multiplier reverting to 200%, the reworked Spellblade passive on items like Passive: Spellblade. Attacks deal bonus physical damage based on AD.
Essence Reaver Essence Reaver 3000 gold
This is also why “the wrong item” costs you far less than most players assume, and a late item costs you far more. A slightly suboptimal legendary completed on time to contest a dragon fight beats the theoretically stronger item finished three minutes after the window closed. Stat pages can’t show you that trade-off, because they only record what you bought — never when you needed it.
Why the Average Can’t See Your Specific Game
Every number discussed so far — raw win rate, purchase win probability, even the Yorick spike data — is still an aggregate. It describes thousands of games that are not your game. Your enemy laner, your gold lead, your team’s poke composition, and the exact minute you’re deciding what to buy next are specifics no league-wide average can encode, no matter how the stat is calculated.
That’s the actual gap between “what the internet says is a good item” and “what you should buy right now.” It’s also the gap tools like buildzcrank are built to close — reading your live match state instead of a patch-wide average, so the recommendation reflects the fight you’re actually in rather than the fight 50,000 other players happened to be in when they bought the same item. An average is a reasonable starting point. It stops being useful the moment your specific game diverges from it, which is most games, most of the time.
None of this is an argument against using stat sites. It’s an argument for knowing what number you’re actually looking at before you build a habit around it — the same reasoning behind why static build guides fail more often than players expect: they’re built on the same averaged, decontextualized numbers.
How to Read Item Stats Without Getting Fooled
You don’t need to abandon stat sites — you need a checklist for what to trust and what to discount before it shapes your build.
Reading item win rate without getting fooled
-
Check the sample size first
A win rate from 200 games swings wildly; a win rate from 20,000 doesn't. If the site shows a match count, treat anything under a few hundred games as noise.
-
Ask what state the buyer was likely in
Situational or expensive luxury items (like Heartsteel) skew toward players already ahead. Core, cheap first-completion items skew less.
-
Weigh your role and matchup over the global number
A tier list average blends every matchup together. Your specific lane opponent and team composition move the real answer more than the headline stat does.
-
Track your own completion timing, not just your item choice
If you're consistently finishing your first legendary item 2+ minutes after opponents in your bracket, that's a bigger fix than swapping which item you buy.
-
Treat the number as a hypothesis, not an answer
Use stat pages to shortlist reasonable options for your role, then let the live game — gold lead, enemy comp, your team's win conditions — make the final call.
Frequently Asked Questions
Why does an item show a high win rate if it isn't actually that strong?
Because raw win rate mixes in the state of the game at the moment of purchase. Items favored by players who are already ahead — bigger gold leads, bigger snowball potential — inherit part of that lead's win rate, a pattern known as selection bias or the 'win-more' effect.
What is purchase win probability?
It's an estimate of a player's win odds at the exact moment they bought an item, rather than the outcome of the whole game. It strips out most of the selection bias baked into raw win rate, though most public stat sites don't display it directly.
Does item timing really matter more than item choice?
In most cases, yes. Completing a legendary item on schedule represents a real power spike; a strong item finished late misses the fight it was meant to win. Data from builds like Yorick's patch 26.13 three-item path (52.9% WR, 227 games via MetaBot.GG) reflects a completed spike more than a specific item's individual strength.
Should I stop using sites like U.GG or lolalytics?
No — they're a reasonable starting shortlist for your role and patch. The mistake is treating the top-line win rate as a final answer instead of a hypothesis you adjust for your matchup, sample size, and live game state.
How does AI-based build advice avoid this bias?
By reading your specific match state — gold, enemy composition, objectives taken — instead of a single averaged number computed across every game on the patch. It doesn't remove the need for good decisions, but it removes the blind spot of comparing your game to an average that doesn't resemble it.
Item win rate isn’t a lie in the sense that the number is fake — it’s a lie in the sense that it answers a question you never asked. The honest question isn’t “which item has the highest win rate,” it’s “given my role, my matchup, and my current game state, which item and which timing actually move my odds.” Stat sites are a fine place to start that search. They’re a bad place to end it.
If you want a build reference that already accounts for role and patch context rather than a flat leaderboard number, buildzcrank’s real-time recommendations adjust as your match develops — and pairing that with a look at how live game-state data actually shapes build advice will change how you read every stat page you open next.