1"use strict";(self.webpackChunk_kaggle_workspace=self.webpackChunk_kaggle_workspace||[]).push([["3144"],{70865(e,n,t){t.d(n,{d:()=>v});var r=t(65412),a=t(55837),i=t(14874),o=t(96540),l=t(52575);function c(e,n){return n||(n=e.slice(0)),Object.freeze(Object.defineProperties(e,{raw:{value:Object.freeze(n)}}))}function u(){var e=c(["\n ","\n"]);return u=function(){return e},e}function d(){var e=c(["\n ","\n"]);return d=function(){return e},e}function s(){var e=c(["\n ","\n\n position: ",";\n box-shadow: ",";\n z-index: ",";\n\n top: 0px;\n padding-top: 8px;\n background-color: ",";\n\n transition: background-color 0.3s ease-in-out;\n"]);return s=function(){return e},e}function m(){var e=c(["\n ","\n"]);return m=function(){return e},e}var f=l.default.div(u(),function(e){return e.theme.km.spacing.withContentPadding}),p=l.default.div(d(),function(e){return e.theme.km.spacing.withContentBody}),h=l.default.div(s(),function(e){return e.theme.km.spacing.withContentPadding},function(e){return e.$isScrolled?"sticky":"auto"},function(e){return e.$isScrolled&&e.$showShadow?e.theme.km.elevation.boxShadow[2]:"0"},function(e){return e.$isScrolled?100:"auto"},function(e){return e.theme.km.color.background.normal}),g=l.default.div(m(),function(e){return e.theme.km.spacing.withContentBody}),v=function(e){var n=e.contentHeader,t=e.onBottomSentinalIsScrolled,l=e.children,c=e.style,u=e.showShadow,d=e.showDivider,s=o.useContext(r._).setHideNavigationTopBar,m=(0,a.l)(),v=(0,a.l)();return o.useEffect(function(){return s(m.isScrolled),function(){return s(!1)}},[m.isScrolled]),o.useEffect(function(){t&&t(v.isScrolled)},[v.isScrolled,t]),o.createElement(o.Fragment,null,o.createElement(a.a,{ref:m.sentinelRef}),o.createElement(f,{style:c},o.createElement(p,null,n)),void 0!==d&&d&&o.createElement(i.c,null),o.createElement(a.a,{ref:v.sentinelRef}),o.createElement(h,{$isScrolled:v.isScrolled,$showShadow:void 0===u||u,style:c},o.createElement(g,null,o.Children.map(l,function(e){return o.isValidElement(e)&&e.props.hasOwnProperty("isScrolled")&&e.props.isScrolled!==v.isScrolled?o.cloneElement(e,{isScrolled:v.isScrolled}):e}))))}},29125(e,n,t){t.d(n,{W:()=>p,n:()=>g});var r=t(83399),a=t(66302),i=t(53490),o=t(96540),l=t(52575);function c(e,n){return n||(n=e.slice(0)),Object.freeze(Object.defineProperties(e,{raw:{value:Object.freeze(n)}}))}function u(){var e=c(["\n /*\n A simple alernative could be to use a pure CSS solution on the section\n container, but UX desires a viewport-wide divider that breaks layout flow.\n */\n background: linear-gradient(\n 0deg,\n "," 0%,\n "," 100%\n );\n border-bottom: 1px solid ",";\n position: absolute;\n bottom: 0;\n height: 120px;\n width: calc(100vw - ","px);\n /*\n Applies a left offset equal to the distance between the left edge of an\n element's normal layout, and the left edge of the viewport.\n */\n left: calc(50% - 50vw);\n"]);return u=function(){return e},e}function d(){var e=c(["\n /* Parent of SectionGradient needs position set for absolute positioning. */\n position: relative;\n /*\n Allow the second-to-last element to stay above the section bottom gradient\n (the last element).\n */\n & > *:nth-last-child(2) {\n position: relative;\n z-index: ",";\n }\n"]);return d=function(){return e},e}function s(){var e=c(["\n ","\n /*\n We use padding for the gradient (which renders inside the section),\n and margin for the offset between sections.\n */\n padding-bottom: 60px;\n margin-bottom: 64px;\n\n @media "," {\n margin-bottom: 40px;\n }\n @media "," {\n padding-bottom: 40px;\n }\n"]);return s=function(){return e},e}var m={light:{border:i.T6,gradientStart:i.MU,gradientEnd:"".concat(i.MU,"00")},dark:{border:i.dW,gradientStart:i._i,gradientEnd:"".concat(i.iF,"00")}},f=l.default.div(u(),function(e){return m[e.theme.name].gradientStart},function(e){return m[e.theme.name].gradientEnd},function(e){return m[e.theme.name].border},function(e){return e.$scrollbarWidth}),p=function(){var e=(0,r.y)();return o.createElement(f,{$scrollbarWidth:e})},h=(0,l.css)(d(),function(e){return e.theme.km.zIndex.above}),g=l.default.section(s(),h,a.vE.TABLET,a.vE.PHONE)},13683(e,n,t){t.d(n,{C:()=>a});var r,a=((r={}).GLOBAL="global",r.RIGHT_RAIL="right-rail",r)},1600(e,n,t){t.d(n,{o:()=>a});var r=t(13683),a=function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:r.C.GLOBAL;return"".concat("kaggle-portal-root","-").concat(e)}},6746(e,n,t){t.d(n,{s:()=>s,Y:()=>m});var r=t(13683),a=t(1600),i=t(96540),o=function(e){var n=e.name,t=e.className,r=e.style;return i.createElement("div",{id:(0,a.o)(n),className:t,style:r})},l=t(66302);function c(e,n){(null==n||n>e.length)&&(n=e.length);for(var t=0,r=Array(n);t<n;t++)r[t]=e[t];return r}function u(){var e,n,t=(e=["\n /* Only shrink content on large screens. */\n @media "," {\n width: ",";\n }\n"],n||(n=e.slice(0)),Object.freeze(Object.defineProperties(e,{raw:{value:Object.freeze(n)}})));return u=function(){return t},t}var d=t(52575).default.div(u(),l.Ch.LG2,function(e){return e.$isOpen?"calc(100% - 340px)":"100%"}
1),s=i.createContext({pushPanelOpen:!1}),m=function(e){var n,t=e.children,a=i.useRef(null),l=function(e){if(Array.isArray(e))return e}(n=i.useState(!1))||function(e){var n,t,r=null==e?null:"u">typeof Symbol&&e[Symbol.iterator]||e["@@iterator"];if(null!=r){var a=[],i=!0,o=!1;try{for(r=r.call(e);!(i=(n=r.next()).done)&&(a.push(n.value),2!==a.length);i=!0);}catch(e){o=!0,t=e}finally{try{i||null==r.return||r.return()}finally{if(o)throw t}}return a}}(n)||function(e){if(e){if("string"==typeof e)return c(e,2);var n=Object.prototype.toString.call(e).slice(8,-1);if("Object"===n&&e.constructor&&(n=e.constructor.name),"Map"===n||"Set"===n)return Array.from(n);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return c(e,2)}}(n)||function(){throw TypeError("Invalid attempt to destructure non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}(),u=l[0],m=l[1];return i.useEffect(function(){var e,n=null==(e=a.current)?void 0:e.childNodes[0];if(!n)throw Error("Could not find portal root with name ".concat(r.C.RIGHT_RAIL," in DOM."));var t=new MutationObserver(function(e,t){m(n.hasChildNodes())});return t.observe(n,{attributes:!1,childList:!0,subtree:!1}),function(){t.disconnect()}},[]),i.createElement(s.Provider,{value:{pushPanelOpen:u}},i.createElement(d,{ref:a,$isOpen:u},i.createElement(o,{name:r.C.RIGHT_RAIL}),t))}},81199(e,n,t){t.r(n),t.d(n,{default:()=>ee});var r=t(94582),a=t(40529),i=t(70865),o=t(13749),l=t(88003),c=t(29125),u=t(66302),d=t(6926),s=t(37958),m=t(64866),f=t(42556),p=t(96540),h=t(52575);function g(e,n){return n||(n=e.slice(0)),Object.freeze(Object.defineProperties(e,{raw:{value:Object.freeze(n)}}))}function v(){var e=g(["\n position: relative;\n padding-bottom: 20px;\n\n @media "," {\n margin-top: ","px;\n }\n"]);return v=function(){return e},e}function b(){var e=g(["\n position: relative;\n width: 540px;\n height: 420px;\n margin-left: 40px;\n\n @media "," {\n width: 100%;\n max-width: 386px;\n height: auto;\n margin-left: 0;\n }\n"]);return b=function(){return e},e}function E(){var e=g(["\n position: absolute;\n top: 124px;\n left: 258px;\n z-index: ",";\n\n @media "," {\n display: none;\n }\n"]);return E=function(){return e},e}function x(){var e=g(["\n display: flex;\n gap: 12px;\n align-self: center;\n"]);return x=function(){return e},e}function w(){var e=g(["\n align-items: center;\n\n h1 {\n font-size: 56px;\n margin-bottom: 36px;\n }\n"]);return w=function(){return e},e}function y(){var e=g(["\n overflow-wrap: break-word;\n max-width: 100%;\n"]);return y=function(){return e},e}var k=h.default.div(v(),u.Ch.PHONE,d.k3),S=h.default.div(b(),u.vE.PHONE),O=h.default.img(E(),function(e){return e.theme.km.zIndex.default},u.vE.PHONE),C=h.default.div(x()),z=(0,h.default)(s.d)(w()),j=h.default.div(y()),A=function(){return p.createElement(k,null,p.createElement(c.W,null),p.createElement(z,{title:"SimsBench",subtitle:p.createElement(j,null,p.createElement(m.cR,{emphasis:"medium"},"Measuring frontier models on long-running, goal-oriented work by having them author agents that climb real Kaggle simulation competition leaderboards.")),condensed:!1,wrap:"wrap",enforceMinHeight:!1,meta:p.createElement(S,null,p.createElement(O,{src:"/static/images/blobs/blob_blue_yellow_together.png",alt:"",width:"281",height:"281"})),actions:p.createElement(C,null,p.createElement(f.$,{emphasis:"medium",leadingIcon:"article",target:"_blank",href:"/blog/introducing-sims-bench"},"Read Our Blog"))}))},N=t(50240);function P(e,n){return n||(n=e.slice(0)),Object.freeze(Object.defineProperties(e,{raw:{value:Object.freeze(n)}}))}function _(){var e=P(["\n display: flex;\n flex-direction: column;\n"]);return _=function(){return e},e}function B(){var e=P(["\n display: grid;\n grid-template-columns: repeat(4, 1fr);\n column-gap: 24px;\n margin-top: 30px;\n\n @media "," {\n grid-template-columns: repeat(2, 1fr);\n }\n\n @media "," {\n grid-template-columns: 1fr;\n }\n"]);return B=function(){return e},e}function I(){var e=P(["\n display: flex;\n flex-direction: column;\n padding-left: 0px;\n padding-top: 36px;\n padding-right: 16px;\n padding-bottom: 36px;\n gap: 16px;\n\n @media "," {\n padding: 16px 0 20px;\n }\n"]);return I=function(){return e},e}function L(){var e=P(["\n display: flex;\n align-items: center;\n gap: 24px;\n"]);return L=function(){return e},e}var R=h.default.div(_()),T=h.default.div(B(),u.vE.TABLET,u.vE.PHONE),H=h.default.div(I(),u.vE.TABLET),$=h.default.div(L()),M=function(e){var n=e.illustrationName,t=e.title,r=e.description;return p.createElement(H,null,p.createElement($,null,p.createElement(N.v,{name:n,alt:"",width:50,height:50}),p.createElement(m.mM,{as:"h3"},t)),p.createElement(m.jz,{emphasis:"medium"},r))},K=function(){return p.createElement(R,null,p.createElement(m.cR,{style:{marginTop:"16px",marginBottom:"24px"},emphasis:"medium"},"SimsBench evaluates the long-running, under-specified, goal-oriented capabilities of frontier models by having them iteratively author and submit agents to Kaggle simulation competitions, then grading them against a frozen, human-calibrated leaderboard."),p.createElement(m.cR,{emphasis:"medium"},"Each run measures how high a model can climb on a real, human-graded competition leaderboard given a competition description, a starter agent, the Kaggle CLI, and a fixed compute budget per agent. Scores are normalized as the model's best agent rating divided by the best human rating, then averaged across all eight currently completed simulation competitions."),p.createElement(T,null,p.createElement(M,{illustrationName:"benchmarks/game-arena-environment",title:"Competitions",description:"Closed Kaggle simulation competitions cloned with a frozen, calibrated leaderboard of human submissions used as benchmarks."}),p.createElement(M,{illustrationName:"benchmarks/game-arena-harness",title:"Harness",description:"A sandboxed Kaggle CLI loop in which a frontier model authors, tests, and submits agents over many iterations."}),p.createElement(M,{illustrationName:"benchmarks/game-arena-visualizers",title:"Evaluation",description:"A two-phase scheduler runs each agent against ~200 calibrated human benchmarks for a tight rating estimate."}),p.createElement(M,{illustrationName:"benchmarks/game-arena-leaderboards",title:"Score",description:"The model's best rating divided by the best human rating, averaged across all currently completed simulation competitions."})))}
1,G=t(6746),W=t(3330),F=t(14874),U=t(26260);function D(e,n){return n||(n=e.slice(0)),Object.freeze(Object.defineProperties(e,{raw:{value:Object.freeze(n)}}))}function Y(){var e=D(["\n display: flex;\n flex-direction: column;\n"]);return Y=function(){return e},e}function q(){var e=D(["\n display: flex;\n flex-direction: row;\n align-items: center;\n gap: 8px;\n margin-bottom: 16px;\n"]);return q=function(){return e},e}function V(){var e=D(["\n display: flex;\n flex-direction: column;\n gap: 12px;\n padding: 32px;\n border: 1px solid ",";\n border-radius: ",";\n background-color: ",";\n"]);return V=function(){return e},e}var X=h.default.div(Y()),J=h.default.div(q()),Q=h.default.div(V(),function(e){return e.theme.km.color.outline.normal},function(e){return e.theme.km.borderRadius.m},function(e){return e.theme.km.color.background.normal}),Z=function(){return p.useEffect(function(){(0,U.HF)(l.g.BENCHMARKS,"sims-bench-home-page-view")},[]),p.createElement(G.Y,null,p.createElement(a._,null,p.createElement(i.d,{showShadow:!1,contentHeader:p.createElement(A,null)}),p.createElement(o.L,{dataTestId:"sims-bench-homepage"},p.createElement(X,null,p.createElement(J,null,p.createElement(W.I,{icon:"bar_chart"}),p.createElement(m.f_,{as:"h2"},"Leaderboard")),p.createElement(Q,null,p.createElement(m.f_,{as:"h3"},"Coming soon"),p.createElement(m.cR,{emphasis:"medium"},"Frontier models are running against the first wave of cloned Kaggle simulation competitions. Aggregate and per-competition scores will appear here once the first runs complete."))),p.createElement(F.c,{style:{margin:"40px 0px"}}),p.createElement(X,null,p.createElement(m.f_,{as:"h2"},"What is SimsBench?"),p.createElement(K,null)))))};let ee=function(){return p.createElement(r.b,{pageName:"sims-bench",highlightedNavBarPage:"benchmarks",showTopNavOverhang:!0,title:"SimsBench | Kaggle",element:p.createElement(Z,null)})}}}]); 2//# sourceMappingURL=SimsBench.4f90e7022bf32358.js.map
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.