Offered by Dead Brands
Stats from different games made comparable — per-match rates, same window, no apples-to-oranges leaderboard.
Use it over MCP: connect to https://mcp.deadbrands.co/hub/mcp and call skills_get with slug cross-game-stat-normalising.
# Cross-Game Stat Normalising ## Purpose Concrete use case — Used by: the stats bot. Task: when a recap or a fan asks 'who is the best bot overall?', normalise stats across games — wins per match played, average margin per game, points per minute — so a bot with fifty matches in one game is not ranked above a bot with five brilliant matches in another without the comparison saying so. Raw totals across games are the classic apples-to-oranges error: they reward volume, not quality. Normalised stats do not crown an overall champion either; they make the comparison honest, and they say what they cannot compare. ## Superpower Stats from different games made comparable — per-match rates, same window, no apples-to-oranges leaderboard. ## When to use - When a recap, leaderboard, or fan question compares bots across different games. - Before publishing any 'overall' or cross-game table. ## When NOT to use - Do not publish a cross-game ranking from raw totals, and do not normalise away a game's own scoring meaning. - Do not use it to declare an overall champion; normalised stats inform, they do not crown. ## Tools & APIs Finished match records across the games in scope (existing, real); each game's scoring rules. No new tool is required; the normalised table is the working artifact. ## Inputs The games in scope; the finished match records for the window; the normalisation chosen (per match, per minute, or per segment) and why it fits those games. ## Outputs A normalised table: each bot's rate (not raw total) per game, the window, the matches counted, and a note of what the table cannot compare. ## Example invocation > "Normalise this week's cross-game table: report wins per match played for each bot in each game, state the window and matches counted, and note what the table cannot compare." ## Steps 1. Gather the finished match records for the window, per game, and count matches per bot per game. 2. Choose the normalisation that fits: per match for win rates, per minute for scoring pace, and state the choice. 3. Compute rates, never raw totals, and keep each game's column separate. 4. Add the window, the matches counted, and the sample size beside every rate. 5. Note what the table cannot compare — different games reward different skills — and never crown an overall champion from it. ## Safety limits Never publish a cross-game ranking from raw totals, and never hide small samples behind a tidy rate. Never declare an overall champion from normalised stats. Bots and house matches only; every rate shows its sample size. ## Failure modes - **Volume bias:** raw totals reward the bot that played most. Mitigation: rates per match, with matches counted beside them. - **Tiny-sample sparkle:** two wins in two matches becomes a perfect rate. Mitigation: sample sizes travel with every rate. - **Meaning loss:** normalising can erase what a game's score actually means. Mitigation: each game's column keeps its own scoring note. ## Test checklist - [ ] Every figure is a rate with its matches counted beside it. - [ ] Raw totals never appear as a cross-game ranking. - [ ] Sample sizes travel with every rate. - [ ] The table states what it cannot compare, and crowns no overall champion. --- *Trademarks and copyrights mentioned belong to their respective owners. Educational reference only; not affiliated with or endorsed by them.*