변경 내역

Winrate has a new design on web and desktop. Every page sits on one shared design system, and most pages were rebuilt.

Navigation

• Reworked layouts for phones and tablets across the site

• One slim nav bar with spotlight search

• Search lists champions, pages, items and runes before summoners

Players

• New player page, with a Habits tab that ranks 24 habits for each player
• Player SQL tab: ask in plain words and the SQL writes itself
• Players section with Search, Leaderboard, One Tricks and Tier Distribution under one header

Champions

• New champion page with similar champions in the header and a merged item tree
• Tier list as role ladders, each role ranked on its own games
• Items as shop shelves, and a patch notes table that marks every changed champion
• Item and rune pages with Overview, Champions and Trends tabs
• Query catalog queries as tabs under the Champions header

Pro

• A rebuilt Pro section: a feed of the latest series with draft ratings, team standings, player compare and pro champion pages

Matches and tools

• Match page with Lane, Teams and Stats tabs
• Draft Assistant fits on one screen, with tiers
• Item Recommender built around the lane matchup

AI

• An ask bar replaces the AI sidebar and /chat, and can ask about a match in place

Desktop

• Home, Game plan, Live and Review rebuilt on the same panels as the website
• Replay picture-in-picture and recording were removed

Billing

• Settings show cancelled and payment-failed states
• A failed payment keeps paid features while Stripe retries

Every model behind the live and post-game tools was retrained on a new training pipeline, alongside a batch of player page fixes.

Models

• Win probability v2: one calibrated model for both the live overlay and the match page
• Draft v2 takes each player's recent form into account
• Item recommender retrained on the new pipeline

Players

• A new player's whole match history is fetched on their first visit, with the last two patches shown first
• Leaderboard drops Master+ players who haven't played ranked in 14 days and ranks by tier, division, then LP
• Leaderboard picks each player's main role and champion by exact game counts
• Player search puts an exact game name match first
• Fixed matchup pair counts and the champion rank rule
• The player page's header and side cards are translated, and League terms read in their gaming sense in every language

The AI chat runs on a new agent, and Winrate's data is available to AI assistants outside the site.

What changed

• Rebuilt AI chat with a list of your chats and memory across conversations
• Answers come from running queries against Winrate's match data
• Chats that stall now retry instead of waiting forever
• The free chat model is now GLM 5.3 Flash
• AI assistants like Claude can connect to Winrate's SQL and player lookups over MCP at /mcp

The desktop app follows a game from champion select to the post-game review, with win probability at the centre of every phase.

What's new

• Live tab and web /live built around a win probability chart that updates through the game
• Review a finished game on the same chart, with its biggest plays
• Game plan before the game starts, with rune and build pickers and Import runes to send a rune page to the League client
• Draft places every champion you could pick on a predicted win chance scale, in the Draft tab and at /tools/draft
• Home shows your recent form and rank climb

Player pages are more resilient while match history is syncing, with several fixes across champion analysis as well.

What changed

• Show available player overview data before the full match sync finishes
• Automatically retry stale player syncs and improve recovery from interrupted updates
• Allow player sync to continue when a timeline contains an unknown lane opponent
• Resolve champion names even when Riot returns different capitalization
• Stop showing the previous champion or role's AI summary while a new one loads
• Limit the Sleeper Builds query to first-item purchases for a consistent comparison

Live item suggestions now handle inventory changes after the bot lane quest more consistently.

What changed

• Keep completed boots accounted for after they move out of the main inventory
• Stop recommending another pair of boots once completed boots have been detected
• Exclude items already in your inventory from the recommendation list
• Keep the next purchase slot in sync with quest boots so recommendations use the correct build stage

The new query catalog brings common champion research questions into dedicated pages.

What's included

• Sleeper builds for exploring less common item choices
• Best champions to main and champion depth analyses
• Elo skew to compare champion performance across ranks
• Experience-adjusted win rates for a different perspective on champion strength

This round of updates improves page loading and smooths out several desktop issues.

What changed

• Reduced the scripts and catalog data loaded up front on the website
• Deferred loading of analytics and the command dialog to lighten the initial page load
• Fixed desktop startup detection when the League client is already running
• Fixed desktop API communication and improved how request errors are displayed

Champion pages now include a dedicated Builds view for exploring your setup in more detail.

What changed

• Compare starter items and item choices at each purchase slot
• Explore primary and secondary runes alongside stat shards
• Select items and runes to narrow the build data around your choices
• See win rates, game counts, and confidence intervals to help judge the strength of each option

This update groups together a set of fixes across desktop, draft, live surfaces, and AI behavior.

What changed

• Fixed desktop sign-in with a browser-to-app flow and cleaned up follow-on desktop auth and cookie issues
• Fixed desktop AI auth so chat works more reliably in the desktop app
• Fixed a draft prediction mismatch that could lead to incorrect recommendations
• Made anonymous vs personalized draft recommendations clearer and improved personalized draft behavior
• Added a live lane matchup kill-delta-by-level view for faster in-game context
• Fixed tier list patch filtering and slider commit behavior
• Preserved more UI state across navigation, including pro board search and some table sorting
• Added soft read-only behavior for chat threads viewed by signed-in non-owners
• Updated AI model labeling and pricing badges, and made MiMo V2 Flash free

This update expands language support across the product so more shared screens feel consistent when you switch locales.

What changed

• Expanded localization across more shared UI surfaces instead of only a small set of pages
• Added Spanish support for player insight metadata and advice
• Added broader language coverage for the app, including German, French, Japanese, Korean, Brazilian Portuguese, Russian, Thai, Turkish, Simplified Chinese, and Traditional Chinese
• Improved the language selector with clearer locale indicators

This update focuses on smaller but important UI improvements, especially on mobile screens where a few key surfaces felt cramped.

What changed

• Reworked the Draft tool on mobile so both teams stay visible side by side instead of forcing team switching
• Improved mobile draft controls and champion search so picks and bans are faster to edit
• Tightened several mobile layouts across champion and build surfaces for cleaner spacing and better readability

This update smooths out a few frustrating player-profile dead ends and makes profile pages behave more consistently.

What changed

• Added a direct action to recover player profiles when search results come back empty
• Improved the flow from search into newly available player pages
• Updated profile lookup behavior so new profiles appear more reliably after they are added
• Fixed profile links in desktop surfaces so they open the correct player page more consistently
• Fixed expanded table rows getting stuck when moving between different player views

This update refines the Live tab after the initial release, with a focus on readability during games and more trustworthy recommendations.

What changed

• Reworked the Live tab layout so team info, win probability, and item suggestions are easier to read at a glance
• Improved the chart so game momentum is shown more clearly from your side of the match
• Tightened the model inputs behind live win probability to better reflect the current game state
• Improved item suggestion accuracy for bot-lane builds

This update focused on making the web experience feel lighter and more reliable across content and data-heavy pages.

What changed

• Improved loading behavior on article, feature, and changelog pages
• Cleaned up filter controls on pro pages for a more consistent browsing experience
• Added stronger protections around heavy query usage to keep shared tools responsive
• Fixed a few web compatibility and accessibility rough edges

Live game view now includes a dedicated win-probability timeline generated from client snapshots.

What changed

• Added match.predictLiveWinProbability route for live game snapshots
• New objective and team-state aggregation from live client events and inventory
• Added a per-minute live win probability chart with gold-diff overlays
• Added ally-side win probability output so results match the current player perspective
• Switched scaling feature lookups to recent patch data from ClickHouse

The desktop pre-game experience was redesigned to make setup decisions faster during champ select.

What changed

• New champion header with compact estimated win-rate display
• Added starter item section alongside summoner spell recommendations
• Added quick rune-page import action directly in the Runes panel
• Added win-rate-by-game-length chart to the pre-game view
• Updated item recommendation ranking logic to use new model scoring

The pro section got a major pass across filters, detail pages, and ranking views.

What changed

• Added multi-select filters for regions, leagues, splits, stages, and positions
• Improved filter defaults and min-games controls for more consistent ranking output
• Player detail now includes champion hero cards, distribution widgets, and sortable champion rows
• Champion detail overview now emphasizes top players/teams with clearer win-rate visuals

The AI experience now opens into persistent thread IDs instead of a single transient chat entrypoint.

What changed

• /chat now creates or resumes a thread and redirects to /chat/{threadId}
• prompt query params can seed the first message in a new thread
• Thread sidebar groups recent chats by date and supports new/delete actions
• Home search and command bar "Ask AI" actions now deep-link into threaded chat

Winrate now has a dedicated /desktop page for app distribution.

What changed

• New desktop landing page with direct download for the latest Windows installer
• Added navigation links from web surfaces to the new desktop page
• Added quick links to changelog, support, and pricing from the desktop page

Desktop releases now support automatic update checks and in-app install prompts.

What changed

• App checks for updates on startup in production builds
• Update toast shows download progress and supports install + restart
• Signed updater artifacts are published from CI to the desktop update bucket
• Replay PiP handling was tightened to reduce duplicate windows during launch

We rebuilt /articles to feel like a proper docs hub instead of a bare list.

What changed

• Sticky left sidebar with sections: Models, Changelog, Articles
• Shared background styling to match Home and Pricing
• New routes:
  • /articles/models
  • /articles/changelog

Added patch notes browsing with links to affected champion pages.

Features

• Patch notes organized by champion and item changes
• Quick links to champion pages to see stat changes
• Item tier list showing current meta builds

Dedicated tier list page with filtering options.

Features

• Filter by position, tier, and patch
• Sort by win rate, pick rate, or ban rate
• Adjustable minimum games threshold
• Links to individual champion pages

Launched premium subscriptions to support the site.

Premium features

• Player-specific draft predictions based on your match history
• Advanced Data Explorer tools
• Priority support

Win probability model trained on ranked match timeline data.

What it shows

• Probability of each team winning at each point in the game
• Updates based on gold leads, objectives, and kills
• Visible on match analysis page timeline

Trained role-specific models to recommend items based on current game state.

How it works

• Models consider gold, inventory, enemy composition, and game time
• Separate models for each role to capture position-specific patterns
• Recommendations update as game state changes
• Shows expected value of each item choice

SQL console on player pages with queries pre-filtered to that player's data.

Features

• Query templates for common analysis patterns
• AI assistant understands the player context
• Results scoped to the player's matches automatically

New match analysis page with in-depth game breakdown.

What's included

• Event timeline showing kills, objectives, and key moments
• Performance scores comparing each player to their tier average
• Gold and XP graphs over game duration
• Replay tab for launching replays from the desktop app

Draft recommendations powered by ML models trained on ranked match data.

How it works

• Enter your champion pool and the current draft state
• Model scores each champion based on team composition and matchups
• Recommendations update as picks and bans change
• Premium users get personalized predictions based on their match history

Launched the Winrate desktop app for Windows.

Features

• All web features available locally
• League client detection for automatic player lookup
• Replay file launching from match history
• Faster startup with native performance

Added authentication with Google OAuth.

What you get

• Save your linked summoner accounts
• Persistent settings across devices
• Required for premium features and subscriptions

Full SQL access to the ClickHouse match database.

Features

• Write and run custom queries against match, player, and timeline data
• AI assistant to help write queries in natural language
• Export results as CSV
• Save and share queries

Trained embeddings on champion co-play patterns to surface champions with similar playstyles.

How it works

• Champions that players tend to main together cluster in embedding space
• Each champion page shows the most similar champions based on this model
• Useful for finding new champions to learn based on your pool

Added dedicated pages for each champion with detailed statistics.

What's included

• Win rate and pick rate breakdowns by tier and position
• Matchup tables showing performance against other champions
• Skill order and ability leveling data
• Build paths and item choices