diff --git a/docs/assets/index-DVN-FNIK.js b/docs/assets/index-DVN-FNIK.js new file mode 100644 index 0000000..e0a103b --- /dev/null +++ b/docs/assets/index-DVN-FNIK.js @@ -0,0 +1,11 @@ +var e=Object.create,t=Object.defineProperty,n=Object.getOwnPropertyDescriptor,r=Object.getOwnPropertyNames,i=Object.getPrototypeOf,a=Object.prototype.hasOwnProperty,o=(e,t)=>()=>(t||(e((t={exports:{}}).exports,t),e=null),t.exports),s=(e,i,o,s)=>{if(i&&typeof i==`object`||typeof i==`function`)for(var c=r(i),l=0,u=c.length,d;li[e]).bind(null,d),enumerable:!(s=n(i,d))||s.enumerable});return e},c=(n,r,o)=>(o=n==null?{}:e(i(n)),s(r||!n||!n.__esModule||!a.call(n,`default`)?t(o,`default`,{value:n,enumerable:!0}):o,n));(function(){let e=document.createElement(`link`).relList;if(e&&e.supports&&e.supports(`modulepreload`))return;for(let e of document.querySelectorAll(`link[rel="modulepreload"]`))n(e);new MutationObserver(e=>{for(let t of e)if(t.type===`childList`)for(let e of t.addedNodes)e.tagName===`LINK`&&e.rel===`modulepreload`&&n(e)}).observe(document,{childList:!0,subtree:!0});function t(e){let t={};return e.integrity&&(t.integrity=e.integrity),e.referrerPolicy&&(t.referrerPolicy=e.referrerPolicy),t.credentials=e.crossOrigin===`use-credentials`?`include`:e.crossOrigin===`anonymous`?`omit`:`same-origin`,t}function n(e){if(e.ep)return;e.ep=!0;let n=t(e);fetch(e.href,n)}})();var l=o((e=>{var t=Symbol.for(`react.transitional.element`),n=Symbol.for(`react.portal`),r=Symbol.for(`react.fragment`),i=Symbol.for(`react.strict_mode`),a=Symbol.for(`react.profiler`),o=Symbol.for(`react.consumer`),s=Symbol.for(`react.context`),c=Symbol.for(`react.forward_ref`),l=Symbol.for(`react.suspense`),u=Symbol.for(`react.memo`),d=Symbol.for(`react.lazy`),f=Symbol.for(`react.activity`),p=Symbol.for(`react.view_transition`),m=Symbol.iterator;function h(e){return typeof e!=`object`||!e?null:(e=m&&e[m]||e[`@@iterator`],typeof e==`function`?e:null)}var g={isMounted:function(){return!1},enqueueForceUpdate:function(){},enqueueReplaceState:function(){},enqueueSetState:function(){}},_=Object.assign,v={};function y(e,t,n){this.props=e,this.context=t,this.refs=v,this.updater=n||g}y.prototype.isReactComponent={},y.prototype.setState=function(e,t){if(typeof e!=`object`&&typeof e!=`function`&&e!=null)throw Error(`takes an object of state variables to update or a function which returns an object of state variables.`);this.updater.enqueueSetState(this,e,t,`setState`)},y.prototype.forceUpdate=function(e){this.updater.enqueueForceUpdate(this,e,`forceUpdate`)};function b(){}b.prototype=y.prototype;function ee(e,t,n){this.props=e,this.context=t,this.refs=v,this.updater=n||g}var x=ee.prototype=new b;x.constructor=ee,_(x,y.prototype),x.isPureReactComponent=!0;var te=Array.isArray;function ne(){}var S={H:null,A:null,T:null,S:null},re=Object.prototype.hasOwnProperty;function C(e,n,r){var i=r.ref;return{$$typeof:t,type:e,key:n,ref:i===void 0?null:i,props:r}}function ie(e,t){return C(e.type,t,e.props)}function w(e){return typeof e==`object`&&!!e&&e.$$typeof===t}function ae(e){var t={"=":`=0`,":":`=2`};return`$`+e.replace(/[=:]/g,function(e){return t[e]})}var T=/\/+/g;function oe(e,t){return typeof e==`object`&&e&&e.key!=null?ae(``+e.key):t.toString(36)}function se(e){switch(e.status){case`fulfilled`:return e.value;case`rejected`:throw e.reason;default:switch(typeof e.status==`string`?e.then(ne,ne):(e.status=`pending`,e.then(function(t){e.status===`pending`&&(e.status=`fulfilled`,e.value=t)},function(t){e.status===`pending`&&(e.status=`rejected`,e.reason=t)})),e.status){case`fulfilled`:return e.value;case`rejected`:throw e.reason}}throw e}function ce(e,r,i,a,o){var s=typeof e;(s===`undefined`||s===`boolean`)&&(e=null);var c=!1;if(e===null)c=!0;else switch(s){case`bigint`:case`string`:case`number`:c=!0;break;case`object`:switch(e.$$typeof){case t:case n:c=!0;break;case d:return c=e._init,ce(c(e._payload),r,i,a,o)}}if(c)return o=o(e),c=a===``?`.`+oe(e,0):a,te(o)?(i=``,c!=null&&(i=c.replace(T,`$&/`)+`/`),ce(o,r,i,``,function(e){return e})):o!=null&&(w(o)&&(o=ie(o,i+(o.key==null||e&&e.key===o.key?``:(``+o.key).replace(T,`$&/`)+`/`)+c)),r.push(o)),1;c=0;var l=a===``?`.`:a+`:`;if(te(e))for(var u=0;u{t.exports=l()})),d=o((e=>{function t(e,t){var n=e.length;e.push(t);a:for(;0>>1,a=e[r];if(0>>1;ri(c,n))li(u,c)?(e[r]=u,e[l]=n,r=l):(e[r]=c,e[s]=n,r=s);else if(li(u,n))e[r]=u,e[l]=n,r=l;else break a}}return t}function i(e,t){var n=e.sortIndex-t.sortIndex;return n===0?e.id-t.id:n}if(e.unstable_now=void 0,typeof performance==`object`&&typeof performance.now==`function`){var a=performance;e.unstable_now=function(){return a.now()}}else{var o=Date,s=o.now();e.unstable_now=function(){return o.now()-s}}var c=[],l=[],u=1,d=null,f=3,p=!1,m=!1,h=!1,g=!1,_=typeof setTimeout==`function`?setTimeout:null,v=typeof clearTimeout==`function`?clearTimeout:null,y=typeof setImmediate<`u`?setImmediate:null;function b(e){for(var i=n(l);i!==null;){if(i.callback===null)r(l);else if(i.startTime<=e)r(l),i.sortIndex=i.expirationTime,t(c,i);else break;i=n(l)}}function ee(e){if(h=!1,b(e),!m){if(n(c)!==null)m=!0,x||(x=!0,ie());else{var t=n(l);t!==null&&T(ee,t.startTime-e)}}}var x=!1,te=-1,ne=5,S=-1;function re(){return g?!0:!(e.unstable_now()-St&&re());){var o=d.callback;if(typeof o==`function`){d.callback=null,f=d.priorityLevel;var s=o(d.expirationTime<=t);if(t=e.unstable_now(),typeof s==`function`){d.callback=s,b(t),i=!0;break b}d===n(c)&&r(c),b(t)}else r(c);d=n(c)}if(d!==null)i=!0;else{var u=n(l);u!==null&&T(ee,u.startTime-t),i=!1}}break a}finally{d=null,f=a,p=!1}i=void 0}}finally{i?ie():x=!1}}}var ie;if(typeof y==`function`)ie=function(){y(C)};else if(typeof MessageChannel<`u`){var w=new MessageChannel,ae=w.port2;w.port1.onmessage=C,ie=function(){ae.postMessage(null)}}else ie=function(){_(C,0)};function T(t,n){te=_(function(){t(e.unstable_now())},n)}e.unstable_IdlePriority=5,e.unstable_ImmediatePriority=1,e.unstable_LowPriority=4,e.unstable_NormalPriority=3,e.unstable_Profiling=null,e.unstable_UserBlockingPriority=2,e.unstable_cancelCallback=function(e){e.callback=null},e.unstable_forceFrameRate=function(e){0>e||125o?(r.sortIndex=a,t(l,r),n(c)===null&&r===n(l)&&(h?(v(te),te=-1):h=!0,T(ee,a-o))):(r.sortIndex=s,t(c,r),m||p||(m=!0,x||(x=!0,ie()))),r},e.unstable_shouldYield=re,e.unstable_wrapCallback=function(e){var t=f;return function(){var n=f;f=t;try{return e.apply(this,arguments)}finally{f=n}}}})),f=o(((e,t)=>{t.exports=d()})),p=o((e=>{var t=u();function n(e){var t=`https://react.dev/errors/`+e;if(1{function n(){if(!(typeof __REACT_DEVTOOLS_GLOBAL_HOOK__>`u`||typeof __REACT_DEVTOOLS_GLOBAL_HOOK__.checkDCE!=`function`))try{__REACT_DEVTOOLS_GLOBAL_HOOK__.checkDCE(n)}catch(e){console.error(e)}}n(),t.exports=p()})),h=o((e=>{var t=f(),n=u(),r=m();function i(e){var t=`https://react.dev/errors/`+e;if(1De||(e.current=Ee[De],Ee[De]=null,De--)}function O(e,t){De++,Ee[De]=e.current,e.current=t}var Ae=Oe(null),je=Oe(null),Me=Oe(null),Ne=Oe(null);function Pe(e,t){switch(O(Me,t),O(je,e),O(Ae,null),t.nodeType){case 9:case 11:e=(e=t.documentElement)&&(e=e.namespaceURI)?up(e):0;break;default:if(e=t.tagName,t=t.namespaceURI)t=up(t),e=dp(t,e);else switch(e){case`svg`:e=1;break;case`math`:e=2;break;default:e=0}}ke(Ae),O(Ae,e)}function Fe(){ke(Ae),ke(je),ke(Me)}function Ie(e){var t=e.memoizedState;t!==null&&(sh._currentValue=t.memoizedState,O(Ne,e)),t=Ae.current;var n=dp(t,e.type);t!==n&&(O(je,e),O(Ae,n))}function Le(e){je.current===e&&(ke(Ae),ke(je)),Ne.current===e&&(ke(Ne),sh._currentValue=Te)}var Re,ze;function Be(e){if(Re===void 0)try{throw Error()}catch(e){var t=e.stack.trim().match(/\n( *(at )?)/);Re=t&&t[1]||``,ze=-1)`:-1i||c[r]!==l[i]){var u=` +`+c[r].replace(` at new `,` at `);return e.displayName&&u.includes(``)&&(u=u.replace(``,e.displayName)),u}while(1<=r&&0<=i);break}}}finally{Ve=!1,Error.prepareStackTrace=n}return(n=e?e.displayName||e.name:``)?Be(n):``}function Ue(e,t){switch(e.tag){case 26:case 27:case 5:return Be(e.type);case 16:return Be(`Lazy`);case 13:return e.child!==t&&t!==null?Be(`Suspense Fallback`):Be(`Suspense`);case 19:return Be(`SuspenseList`);case 0:case 15:return He(e.type,!1);case 11:return He(e.type.render,!1);case 1:return He(e.type,!0);case 31:return Be(`Activity`);case 30:return Be(`ViewTransition`);default:return``}}function We(e){try{var t=``,n=null;do t+=Ue(e,n),n=e,e=e.return;while(e);return t}catch(e){return` +Error generating stack: `+e.message+` +`+e.stack}}var Ge=Object.prototype.hasOwnProperty,Ke=t.unstable_scheduleCallback,qe=t.unstable_cancelCallback,Je=t.unstable_shouldYield,Ye=t.unstable_requestPaint,Xe=t.unstable_now,Ze=t.unstable_getCurrentPriorityLevel,Qe=t.unstable_ImmediatePriority,$e=t.unstable_UserBlockingPriority,et=t.unstable_NormalPriority,tt=t.unstable_LowPriority,nt=t.unstable_IdlePriority,rt=t.log,it=t.unstable_setDisableYieldValue,at=null,ot=null;function st(e){if(typeof rt==`function`&&it(e),ot&&typeof ot.setStrictMode==`function`)try{ot.setStrictMode(at,e)}catch{}}var ct=Math.clz32?Math.clz32:dt,lt=Math.log,ut=Math.LN2;function dt(e){return e>>>=0,e===0?32:31-(lt(e)/ut|0)|0}var ft=256,pt=262144,mt=4194304;function ht(e){var t=e&42;if(t!==0)return t;switch(e&-e){case 1:return 1;case 2:return 2;case 4:return 4;case 8:return 8;case 16:return 16;case 32:return 32;case 64:return 64;case 128:return 128;case 256:case 512:case 1024:case 2048:case 4096:case 8192:case 16384:case 32768:case 65536:case 131072:return e&-e;case 262144:case 524288:case 1048576:case 2097152:return e&3932160;case 4194304:case 8388608:case 16777216:case 33554432:return e&62914560;case 67108864:return 67108864;case 134217728:return 134217728;case 268435456:return 268435456;case 536870912:return 536870912;case 1073741824:return 0;default:return e}}function gt(e,t,n){var r=e.pendingLanes;if(r===0)return 0;var i=0,a=e.suspendedLanes,o=e.pingedLanes;e=e.warmLanes;var s=r&134217727;return s===0?(s=r&~a,s===0?o===0?n||(n=r&~e,n!==0&&(i=ht(n))):i=ht(o):i=ht(s)):(r=s&~a,r===0?(o&=s,o===0?n||(n=s&~e,n!==0&&(i=ht(n))):i=ht(o)):i=ht(r)),i===0?0:t!==0&&t!==i&&(t&a)===0&&(a=i&-i,n=t&-t,a>=n||a===32&&n&4194048)?t:i}function _t(e,t){return(e.pendingLanes&~(e.suspendedLanes&~e.pingedLanes)&t)===0}function vt(e,t){t&8&&(t|=t&32);var n=e.entangledLanes;if(n!==0)for(e=e.entanglements,n&=t;0n;n++)t.push(e);return t}function St(e,t){e.pendingLanes|=t,t!==268435456&&(e.suspendedLanes=0,e.pingedLanes=0,e.warmLanes=0)}function Ct(e,t,n,r,i,a){var o=e.pendingLanes;e.pendingLanes=n,e.suspendedLanes=0,e.pingedLanes=0,e.warmLanes=0,e.expiredLanes&=n,e.entangledLanes&=n,e.errorRecoveryDisabledLanes&=n,e.shellSuspendCounter=0;var s=e.entanglements,c=e.expirationTimes,l=e.hiddenUpdates;for(n=o&~n;0`u`||window.document===void 0||window.document.createElement===void 0),Rn=!1;if(Ln)try{var zn={};Object.defineProperty(zn,"passive",{get:function(){Rn=!0}}),window.addEventListener(`test`,zn,zn),window.removeEventListener(`test`,zn,zn)}catch{Rn=!1}var Bn=null,Vn=null,Hn=null;function Un(){if(Hn)return Hn;var e,t=Vn,n=t.length,r,i=`value`in Bn?Bn.value:Bn.textContent,a=i.length;for(e=0;e=Sr),Tr=` `,Er=!1;function Dr(e,t){switch(e){case`keyup`:return br.indexOf(t.keyCode)!==-1;case`keydown`:return t.keyCode!==229;case`keypress`:case`mousedown`:case`focusout`:return!0;default:return!1}}function Or(e){return e=e.detail,typeof e==`object`&&`data`in e?e.data:null}var kr=!1;function Ar(e,t){switch(e){case`compositionend`:return Or(t);case`keypress`:return t.which===32?(Er=!0,Tr):null;case`textInput`:return e=t.data,e===Tr&&Er?null:e;default:return null}}function jr(e,t){if(kr)return e===`compositionend`||!xr&&Dr(e,t)?(e=Un(),Hn=Vn=Bn=null,kr=!1,e):null;switch(e){case`paste`:return null;case`keypress`:if(!(t.ctrlKey||t.altKey||t.metaKey)||t.ctrlKey&&t.altKey){if(t.char&&1=t)return{node:n,offset:t-e};e=r}a:{for(;n;){if(n.nextSibling){n=n.nextSibling;break a}n=n.parentNode}n=void 0}n=ei(n)}}function ni(e,t){return e&&t?e===t?!0:e&&e.nodeType===3?!1:t&&t.nodeType===3?ni(e,t.parentNode):`contains`in e?e.contains(t):e.compareDocumentPosition?!!(e.compareDocumentPosition(t)&16):!1:!1}function ri(e){e=e!=null&&e.ownerDocument!=null&&e.ownerDocument.defaultView!=null?e.ownerDocument.defaultView:window;for(var t=$r(e.document);t instanceof e.HTMLIFrameElement;){try{var n=typeof t.contentWindow.location.href==`string`}catch{n=!1}if(n)e=t.contentWindow;else break;t=$r(e.document)}return t}function ii(e){var t=e&&e.nodeName&&e.nodeName.toLowerCase();return t&&(t===`input`&&(e.type===`text`||e.type===`search`||e.type===`tel`||e.type===`url`||e.type===`password`)||t===`textarea`||e.contentEditable===`true`)}var ai=Ln&&`documentMode`in document&&11>=document.documentMode,oi=null,si=null,ci=null,li=!1;function ui(e,t,n){var r=n.window===n?n.document:n.nodeType===9?n:n.ownerDocument;li||oi==null||oi!==$r(r)||(r=oi,`selectionStart`in r&&ii(r)?r={start:r.selectionStart,end:r.selectionEnd}:(r=(r.ownerDocument&&r.ownerDocument.defaultView||window).getSelection(),r={anchorNode:r.anchorNode,anchorOffset:r.anchorOffset,focusNode:r.focusNode,focusOffset:r.focusOffset}),ci&&Qr(ci,r)||(ci=r,r=Jf(si,`onSelect`),0>=o,i-=o,oa=1<<32-ct(t)+i|n<h?(g=d,d=null):g=d.sibling;var _=p(i,d,s[h],c);if(_===null){d===null&&(d=g);break}e&&d&&_.alternate===null&&t(i,d),a=o(_,a,h),u===null?l=_:u.sibling=_,u=_,d=g}if(h===s.length)return n(i,d),j&&ca(i,h),l;if(d===null){for(;hg?(_=h,h=null):_=h.sibling;var y=p(a,h,v.value,l);if(y===null){h===null&&(h=_);break}e&&h&&y.alternate===null&&t(a,h),s=o(y,s,g),d===null?u=y:d.sibling=y,d=y,h=_}if(v.done)return n(a,h),j&&ca(a,g),u;if(h===null){for(;!v.done;g++,v=c.next())v=f(a,v.value,l),v!==null&&(s=o(v,s,g),d===null?u=v:d.sibling=v,d=v);return j&&ca(a,g),u}for(h=r(h);!v.done;g++,v=c.next())v=m(h,a,g,v.value,l),v!==null&&(e&&(_=v.alternate,_!==null&&h.delete(_.key===null?g:_.key)),s=o(v,s,g),d===null?u=v:d.sibling=v,d=v);return e&&h.forEach(function(e){return t(a,e)}),j&&ca(a,g),u}function _(e,r,o,c){if(typeof o==`object`&&o&&o.type===T&&o.key===null&&o.props.ref===void 0&&(o=o.props.children),typeof o==`object`&&o){switch(o.$$typeof){case w:a:{for(var l=o.key;r!==null;){if(r.key===l){if(l=o.type,l===T){if(r.tag===7){n(e,r.sibling),c=a(r,o.props.children),_o(c,o),c.return=e,e=c;break a}}else if(r.elementType===l||typeof l==`object`&&l&&l.$$typeof===me&&lo(l)===r.type){n(e,r.sibling),c=a(r,o.props),_o(c,o),c.return=e,e=c;break a}n(e,r);break}t(e,r),r=r.sibling}o.type===T?(c=qi(o.props.children,e.mode,c,o.key),_o(c,o),c.return=e,e=c):(c=Ki(o.type,o.key,o.props,null,e.mode,c),_o(c,o),c.return=e,e=c)}return s(e);case ae:a:{for(l=o.key;r!==null;){if(r.key===l){if(r.tag===4&&r.stateNode.containerInfo===o.containerInfo&&r.stateNode.implementation===o.implementation){n(e,r.sibling),c=a(r,o.children||[]),c.return=e,e=c;break a}n(e,r);break}t(e,r),r=r.sibling}c=Xi(o,e.mode,c),c.return=e,e=c}return s(e);case me:return o=lo(o),_(e,r,o,c)}if(we(o))return h(e,r,o,c);if(xe(o)){if(l=xe(o),typeof l!=`function`)throw Error(i(150));return o=l.call(o),g(e,r,o,c)}if(typeof o.then==`function`)return _(e,r,go(o),c);if(o.$$typeof===le)return _(e,r,Fa(e,o),c);vo(e,o)}return typeof o==`string`&&o!==``||typeof o==`number`||typeof o==`bigint`?(o=``+o,r!==null&&r.tag===6?(n(e,r.sibling),c=a(r,o),c.return=e,e=c):(n(e,r),c=Ji(o,e.mode,c),c.return=e,e=c),s(e)):n(e,r)}return function(e,t,n,r){try{ho=0;var i=_(e,t,n,r);return mo=null,i}catch(t){if(t===ro||t===ao)throw t;var a=Hi(29,t,null,e.mode);return a.lanes=r,a.return=e,a}}}var bo=yo(!0),xo=yo(!1),So=!1;function Co(e){e.updateQueue={baseState:e.memoizedState,firstBaseUpdate:null,lastBaseUpdate:null,shared:{pending:null,lanes:0,hiddenCallbacks:null},callbacks:null}}function wo(e,t){e=e.updateQueue,t.updateQueue===e&&(t.updateQueue={baseState:e.baseState,firstBaseUpdate:e.firstBaseUpdate,lastBaseUpdate:e.lastBaseUpdate,shared:e.shared,callbacks:null})}function To(e){return{lane:e,tag:0,payload:null,callback:null,next:null}}function Eo(e,t,n){var r=e.updateQueue;if(r===null)return null;if(r=r.shared,W&2){var i=r.pending;return i===null?t.next=t:(t.next=i.next,i.next=t),r.pending=t,t=zi(e),Ri(e,null,n),t}return Fi(e,r,t,n),zi(e)}function Do(e,t,n){if(t=t.updateQueue,t!==null&&(t=t.shared,n&4194048)){var r=t.lanes;r&=e.pendingLanes,n|=r,t.lanes=n,Tt(e,n)}}function Oo(e,t){var n=e.updateQueue,r=e.alternate;if(r!==null&&(r=r.updateQueue,n===r)){var i=null,a=null;if(n=n.firstBaseUpdate,n!==null){do{var o={lane:n.lane,tag:n.tag,payload:n.payload,callback:null,next:null};a===null?i=a=o:a=a.next=o,n=n.next}while(n!==null);a===null?i=a=t:a=a.next=t}else i=a=t;n={baseState:r.baseState,firstBaseUpdate:i,lastBaseUpdate:a,shared:r.shared,callbacks:r.callbacks},e.updateQueue=n;return}e=n.lastBaseUpdate,e===null?n.firstBaseUpdate=t:e.next=t,n.lastBaseUpdate=t}var ko=!1;function Ao(){if(ko){var e=Ja;if(e!==null)throw e}}function jo(e,t,n,r){ko=!1;var i=e.updateQueue;So=!1;var a=i.firstBaseUpdate,o=i.lastBaseUpdate,s=i.shared.pending;if(s!==null){i.shared.pending=null;var c=s,l=c.next;c.next=null,o===null?a=l:o.next=l,o=c;var u=e.alternate;u!==null&&(u=u.updateQueue,s=u.lastBaseUpdate,s!==o&&(s===null?u.firstBaseUpdate=l:s.next=l,u.lastBaseUpdate=c))}if(a!==null){var d=i.baseState;o=0,u=l=c=null,s=a;do{var f=s.lane&-536870913,p=f!==s.lane;if(p?(q&f)===f:(r&f)===f){f!==0&&f===qa&&(ko=!0),u!==null&&(u=u.next={lane:0,tag:s.tag,payload:s.payload,callback:null,next:null});a:{var m=e,h=s;f=t;var g=n;switch(h.tag){case 1:if(m=h.payload,typeof m==`function`){d=m.call(g,d,f);break a}d=m;break a;case 3:m.flags=m.flags&-65537|128;case 0:if(m=h.payload,f=typeof m==`function`?m.call(g,d,f):m,f==null)break a;d=C({},d,f);break a;case 2:So=!0}}f=s.callback,f!==null&&(e.flags|=64,p&&(e.flags|=8192),p=i.callbacks,p===null?i.callbacks=[f]:p.push(f))}else p={lane:f,tag:s.tag,payload:s.payload,callback:s.callback,next:null},u===null?(l=u=p,c=d):u=u.next=p,o|=f;if(s=s.next,s===null){if(s=i.shared.pending,s===null)break;p=s,s=p.next,p.next=null,i.lastBaseUpdate=p,i.shared.pending=null}}while(1);u===null&&(c=d),i.baseState=c,i.firstBaseUpdate=l,i.lastBaseUpdate=u,a===null&&(i.shared.lanes=0),sd|=o,e.lanes=o,e.memoizedState=d}}function Mo(e,t){if(typeof e!=`function`)throw Error(i(191,e));e.call(t)}function No(e,t){var n=e.callbacks;if(n!==null)for(e.callbacks=null,e=0;ea?a:8;var o=E.T,s={};s.types=o===null?null:o.types,E.T=s,gc(e,!1,t,n);try{var c=i(),l=E.S;l!==null&&l(s,c),typeof c==`object`&&c&&typeof c.then==`function`?hc(e,t,Za(c,r),jd(e)):hc(e,t,r,jd(e))}catch(n){hc(e,t,{then:function(){},status:`rejected`,reason:n},jd())}finally{D.p=a,o!==null&&s.types!==null&&(o.types=s.types),E.T=o}}function ac(){}function oc(e,t,n,r){if(e.tag!==5)throw Error(i(476));var a=sc(e).queue;ic(e,a,t,Te,n===null?ac:function(){return cc(e),n(r)})}function sc(e){var t=e.memoizedState;if(t!==null)return t;t={memoizedState:Te,baseState:Te,baseQueue:null,queue:{pending:null,lanes:0,dispatch:null,lastRenderedReducer:_s,lastRenderedState:Te},next:null};var n={};return t.next={memoizedState:n,baseState:n,baseQueue:null,queue:{pending:null,lanes:0,dispatch:null,lastRenderedReducer:_s,lastRenderedState:n},next:null},e.memoizedState=t,e=e.alternate,e!==null&&(e.memoizedState=t),t}function cc(e){var t=sc(e);t.next===null&&(t=e.alternate.memoizedState),hc(e,t.next.queue,{},jd())}function lc(){return Pa(sh)}function uc(){return L().memoizedState}function dc(){return L().memoizedState}function fc(e){for(var t=e.return;t!==null;){switch(t.tag){case 24:case 3:var n=jd();e=To(n);var r=Eo(t,e,n);r!==null&&(Pd(r,t,n),Do(r,t,n)),t={cache:Ba()},e.payload=t;return}t=t.return}}function pc(e,t,n){var r=jd();n={lane:r,revertLane:0,gesture:null,action:n,hasEagerState:!1,eagerState:null,next:null},_c(e)?vc(t,n):(n=Ii(e,t,n,r),n!==null&&(Pd(n,e,r),yc(n,t,r)))}function mc(e,t,n){hc(e,t,n,jd())}function hc(e,t,n,r){var i={lane:r,revertLane:0,gesture:null,action:n,hasEagerState:!1,eagerState:null,next:null};if(_c(e))vc(t,i);else{var a=e.alternate;if(e.lanes===0&&(a===null||a.lanes===0)&&(a=t.lastRenderedReducer,a!==null))try{var o=t.lastRenderedState,s=a(o,n);if(i.hasEagerState=!0,i.eagerState=s,Zr(s,o))return Fi(e,t,i,0),G===null&&Pi(),!1}catch{}if(n=Ii(e,t,i,r),n!==null)return Pd(n,e,r),yc(n,t,r),!0}return!1}function gc(e,t,n,r){if(r={lane:2,revertLane:Pf(),gesture:null,action:r,hasEagerState:!1,eagerState:null,next:null},_c(e)){if(t)throw Error(i(479))}else t=Ii(e,n,r,2),t!==null&&Pd(t,e,2)}function _c(e){var t=e.alternate;return e===N||t!==null&&t===N}function vc(e,t){Qo=Zo=!0;var n=e.pending;n===null?t.next=t:(t.next=n.next,n.next=t),e.pending=t}function yc(e,t,n){if(n&4194048){var r=t.lanes;r&=e.pendingLanes,n|=r,t.lanes=n,Tt(e,n)}}var bc={readContext:Pa,use:hs,useCallback:I,useContext:I,useEffect:I,useImperativeHandle:I,useLayoutEffect:I,useInsertionEffect:I,useMemo:I,useReducer:I,useRef:I,useState:I,useDebugValue:I,useDeferredValue:I,useTransition:I,useSyncExternalStore:I,useId:I,useHostTransitionStatus:I,useFormState:I,useActionState:I,useOptimistic:I,useMemoCache:I,useCacheRefresh:I,useEffectEvent:I},xc={readContext:Pa,use:hs,useCallback:function(e,t){return fs().memoizedState=[e,t===void 0?null:t],e},useContext:Pa,useEffect:Gs,useImperativeHandle:function(e,t,n){n=n==null?null:n.concat([e]),Us(4194308,4,Zs.bind(null,t,e),n)},useLayoutEffect:function(e,t){return Us(4194308,4,e,t)},useInsertionEffect:function(e,t){Us(4,2,e,t)},useMemo:function(e,t){var n=fs();t=t===void 0?null:t;var r=e();if($o){st(!0);try{e()}finally{st(!1)}}return n.memoizedState=[r,t],r},useReducer:function(e,t,n){var r=fs();if(n!==void 0){var i=n(t);if($o){st(!0);try{n(t)}finally{st(!1)}}}else i=t;return r.memoizedState=r.baseState=i,e={pending:null,lanes:0,dispatch:null,lastRenderedReducer:e,lastRenderedState:i},r.queue=e,e=e.dispatch=pc.bind(null,N,e),[r.memoizedState,e]},useRef:function(e){var t=fs();return e={current:e},t.memoizedState=e},useState:function(e){e=Ds(e);var t=e.queue,n=mc.bind(null,N,t);return t.dispatch=n,[e.memoizedState,n]},useDebugValue:$s,useDeferredValue:function(e,t){return nc(fs(),e,t)},useTransition:function(){var e=Ds(!1);return e=ic.bind(null,N,e.queue,!0,!1),fs().memoizedState=e,[!1,e]},useSyncExternalStore:function(e,t,n){var r=N,a=fs();if(j){if(n===void 0)throw Error(i(407));n=n()}else{if(n=t(),G===null)throw Error(i(349));q&127||Ss(r,t,n)}a.memoizedState=n;var o={value:n,getSnapshot:t};return a.queue=o,Gs(ws.bind(null,r,o,e),[e]),r.flags|=2048,Vs(9,{destroy:void 0},Cs.bind(null,r,o,n,t),null),n},useId:function(){var e=fs(),t=G.identifierPrefix;if(j){var n=sa,r=oa;n=(r&~(1<<32-ct(r)-1)).toString(32)+n,t=`_`+t+`R_`+n,n=es++,0<\/script>`,o=o.removeChild(o.firstChild);break;case`select`:o=typeof r.is==`string`?s.createElement(`select`,{is:r.is}):s.createElement(`select`),r.multiple?o.multiple=!0:r.size&&(o.size=r.size);break;default:o=typeof r.is==`string`?s.createElement(a,{is:r.is}):s.createElement(a)}}o[Mt]=t,o[Nt]=r;a:for(s=t.child;s!==null;){if(s.tag===5||s.tag===6)o.appendChild(s.stateNode);else if(s.tag!==4&&s.tag!==27&&s.child!==null){s.child.return=s,s=s.child;continue}if(s===t)break a;for(;s.sibling===null;){if(s.return===null||s.return===t)break a;s=s.return}s.sibling.return=s.return,s=s.sibling}t.stateNode=o;a:switch(np(o,a,r),a){case`button`:case`input`:case`select`:case`textarea`:r=!!r.autoFocus;break a;case`img`:r=!0;break a;default:r=!1}r&&_l(t)}}return B(t),t.subtreeFlags&=-33554433,vl(t,t.type,e===null?null:e.memoizedProps,t.pendingProps,n),null;case 6:if(e&&t.stateNode!=null)e.memoizedProps!==r&&_l(t);else{if(typeof r!=`string`&&t.stateNode===null)throw Error(i(166));if(e=Me.current,ba(t)){if(e=t.stateNode,n=t.memoizedProps,r=null,a=pa,a!==null)switch(a.tag){case 27:case 5:r=a.memoizedProps}e[Mt]=t,e=!!(e.nodeValue===n||r!==null&&!0===r.suppressHydrationWarning||ep(e.nodeValue,n)),e||_a(t,!0)}else e=lp(e).createTextNode(r),e[Mt]=t,t.stateNode=e}return B(t),null;case 31:if(n=t.memoizedState,e===null||e.memoizedState!==null){if(r=ba(t),n!==null){if(e===null){if(!r)throw Error(i(318));if(e=t.memoizedState,e=e===null?null:e.dehydrated,!e)throw Error(i(557));e[Mt]=t}else xa(),!(t.flags&128)&&(t.memoizedState=null),t.flags|=4;B(t),e=!1}else n=Sa(),e!==null&&e.memoizedState!==null&&(e.memoizedState.hydrationErrors=n),e=!0;if(!e)return t.flags&256?(Go(t),t):(Go(t),null);if(t.flags&128)throw Error(i(558))}return B(t),null;case 13:if(r=t.memoizedState,e===null||e.memoizedState!==null&&e.memoizedState.dehydrated!==null){if(a=ba(t),r!==null&&r.dehydrated!==null){if(e===null){if(!a)throw Error(i(318));if(a=t.memoizedState,a=a===null?null:a.dehydrated,!a)throw Error(i(317));a[Mt]=t}else xa(),!(t.flags&128)&&(t.memoizedState=null),t.flags|=4;B(t),a=!1}else a=Sa(),e!==null&&e.memoizedState!==null&&(e.memoizedState.hydrationErrors=a),a=!0;if(!a)return t.flags&256?(Go(t),t):(Go(t),null)}return Go(t),t.flags&128?(t.lanes=n,t):(n=r!==null,e=e!==null&&e.memoizedState!==null,n&&(r=t.child,a=null,r.alternate!==null&&r.alternate.memoizedState!==null&&r.alternate.memoizedState.cachePool!==null&&(a=r.alternate.memoizedState.cachePool.pool),o=null,r.memoizedState!==null&&r.memoizedState.cachePool!==null&&(o=r.memoizedState.cachePool.pool),o!==a&&(r.flags|=2048)),n!==e&&n&&(t.child.flags|=8192),bl(t,t.updateQueue),B(t),null);case 4:return Fe(),e===null&&Wf(t.stateNode.containerInfo),t.flags|=67108864,B(t),null;case 10:return Oa(t.type),B(t),null;case 19:if(Jo(t),r=t.memoizedState,r===null)return B(t),null;if(a=!!(t.flags&128),o=r.rendering,o===null){if(a)xl(r,!1);else{if(Y!==0||e!==null&&e.flags&128)for(e=t.child;e!==null;){if(o=Yo(e),o!==null){for(t.flags|=128,xl(r,!1),e=o.updateQueue,t.updateQueue=e,bl(t,e),t.subtreeFlags=0,e=n,n=t.child;n!==null;)Gi(n,e),n=n.sibling;return qo(t,Ko.current&1|2),j&&ca(t,r.treeForkCount),t.child}e=e.sibling}r.tail!==null&&Xe()>_d&&(t.flags|=128,a=!0,xl(r,!1),t.lanes=4194304)}}else{if(!a){if(e=Yo(o),e!==null){if(t.flags|=128,a=!0,e=e.updateQueue,t.updateQueue=e,bl(t,e),xl(r,!0),r.tail===null&&r.tailMode!==`collapsed`&&r.tailMode!==`visible`&&!o.alternate&&!j)return B(t),null}else 2*Xe()-r.renderingStartTime>_d&&n!==536870912&&(t.flags|=128,a=!0,xl(r,!1),t.lanes=4194304)}r.isBackwards?(o.sibling=t.child,t.child=o):(e=r.last,e===null?t.child=o:e.sibling=o,r.last=o)}if(r.tail!==null){e=r.tail;a:{for(n=e;n!==null;){if(n.alternate!==null){n=!1;break a}n=n.sibling}n=!0}return r.rendering=e,r.tail=e.sibling,r.renderingStartTime=Xe(),e.sibling=null,o=Ko.current,o=a?o&1|2:o&1,r.tailMode===`visible`||r.tailMode===`collapsed`||!n||j?qo(t,o):(n=o,O(zo,t),O(Ko,n),Bo===null&&(Bo=t)),j&&ca(t,r.treeForkCount),e}return B(t),null;case 22:case 23:return Go(t),Ro(),r=t.memoizedState!==null,e===null?r&&(t.flags|=8192):e.memoizedState!==null!==r&&(t.flags|=8192),r?n&536870912&&!(t.flags&128)&&(B(t),t.subtreeFlags&6&&(t.flags|=8192)):B(t),n=t.updateQueue,n!==null&&bl(t,n.retryQueue),n=null,e!==null&&e.memoizedState!==null&&e.memoizedState.cachePool!==null&&(n=e.memoizedState.cachePool.pool),r=null,t.memoizedState!==null&&t.memoizedState.cachePool!==null&&(r=t.memoizedState.cachePool.pool),r!==n&&(t.flags|=2048),e!==null&&ke($a),null;case 24:return n=null,e!==null&&(n=e.memoizedState.cache),t.memoizedState.cache!==n&&(t.flags|=2048),Oa(M),B(t),null;case 25:return null;case 30:return t.flags|=33554432,B(t),null}throw Error(i(156,t.tag))}function Cl(e,t){switch(da(t),t.tag){case 1:return e=t.flags,e&65536?(t.flags=e&-65537|128,t):null;case 3:return Oa(M),Fe(),e=t.flags,e&65536&&!(e&128)?(t.flags=e&-65537|128,t):null;case 26:case 27:case 5:return Le(t),null;case 31:if(t.memoizedState!==null){if(Go(t),t.alternate===null)throw Error(i(340));xa()}return e=t.flags,e&65536?(t.flags=e&-65537|128,t):null;case 13:if(Go(t),e=t.memoizedState,e!==null&&e.dehydrated!==null){if(t.alternate===null)throw Error(i(340));xa()}return e=t.flags,e&65536?(t.flags=e&-65537|128,t):null;case 19:return Jo(t),e=t.flags,e&65536?(t.flags=e&-65537|128,e=t.memoizedState,e!==null&&(e.rendering=null,e.tail=null),t.flags|=4,t):null;case 4:return Fe(),null;case 10:return Oa(t.type),null;case 22:case 23:return Go(t),Ro(),e!==null&&ke($a),e=t.flags,e&65536?(t.flags=e&-65537|128,t):null;case 24:return Oa(M),null;case 25:return null;default:return null}}function wl(e,t){switch(da(t),t.tag){case 3:Oa(M),Fe();break;case 26:case 27:case 5:Le(t);break;case 4:Fe();break;case 31:t.memoizedState!==null&&Go(t);break;case 13:Go(t);break;case 19:Jo(t);break;case 10:Oa(t.type);break;case 22:case 23:Go(t),Ro(),e!==null&&ke($a);break;case 24:Oa(M)}}function Tl(e,t){try{var n=t.updateQueue,r=n===null?null:n.lastEffect;if(r!==null){var i=r.next;n=i;do{if((n.tag&e)===e){r=void 0;var a=n.create,o=n.inst;r=a(),o.destroy=r}n=n.next}while(n!==i)}}catch(e){Z(t,t.return,e)}}function El(e,t,n){try{var r=t.updateQueue,i=r===null?null:r.lastEffect;if(i!==null){var a=i.next;r=a;do{if((r.tag&e)===e){var o=r.inst,s=o.destroy;if(s!==void 0){o.destroy=void 0,i=t;var c=n,l=s;try{l()}catch(e){Z(i,c,e)}}}r=r.next}while(r!==a)}}catch(e){Z(t,t.return,e)}}function Dl(e){var t=e.updateQueue;if(t!==null){var n=e.stateNode;try{No(t,n)}catch(t){Z(e,e.return,t)}}}function Ol(e,t,n){n.props=Oc(e.type,e.memoizedProps),n.state=e.memoizedState;try{n.componentWillUnmount()}catch(n){Z(e,t,n)}}function kl(e,t){try{var n=e.ref;if(n!==null){switch(e.tag){case 26:case 27:case 5:var r=e.stateNode;break;case 30:var i=e.stateNode,a=Di(e.memoizedProps,i);(i.ref===null||i.ref.name!==a)&&(i.ref=Pp(a)),r=i.ref;break;case 7:if(e.stateNode===null){var o=new Fp(e);h(e.child,!1,Qp,o,void 0,void 0),e.stateNode=o}r=e.stateNode;break;default:r=e.stateNode}typeof n==`function`?e.refCleanup=n(r):n.current=r}}catch(n){Z(e,t,n)}}function Al(e,t){var n=e.ref,r=e.refCleanup;if(n!==null){if(typeof r==`function`)try{r()}catch(n){Z(e,t,n)}finally{e.refCleanup=null,e=e.alternate,e!=null&&(e.refCleanup=null)}else if(typeof n==`function`)try{n(null)}catch(n){Z(e,t,n)}else n.current=null}}function jl(e,t){if((e.tag===5||e.tag===27||e.tag===6)&&e.alternate===null&&t!==null)for(var n=0;n title`))),np(r,t,n),r[Mt]=e,Kt(r),t=r;break a;case`link`:if(o=Gm(`link`,`href`,a).get(t+(n.href||``))){for(s=0;sg&&(o=g,g=h,h=o);var _=ti(s,h),v=ti(s,g);if(_&&v&&(p.rangeCount!==1||p.anchorNode!==_.node||p.anchorOffset!==_.offset||p.focusNode!==v.node||p.focusOffset!==v.offset)){var y=d.createRange();y.setStart(_.node,_.offset),p.removeAllRanges(),h>g?(p.addRange(y),p.extend(v.node,v.offset)):(y.setEnd(v.node,v.offset),p.addRange(y))}}}}for(d=[],p=s;p=p.parentNode;)p.nodeType===1&&d.push({element:p,left:p.scrollLeft,top:p.scrollTop});for(typeof s.focus==`function`&&s.focus(),s=0;sn?32:n,E.T=null,n=wd,wd=null;var o=bd,s=Sd;if(X=0,xd=bd=null,Sd=0,W&6)throw Error(i(331));var c=W;if(W|=4,Zu(o.current),Uu(o,o.current,s,n),W=c,Df(0,!1),ot&&typeof ot.onPostCommitFiberRoot==`function`)try{ot.onPostCommitFiberRoot(at,o)}catch{}return!0}finally{D.p=a,E.T=r,uf(e,t)}}function pf(e,t,n){t=Qi(n,t),t=Pc(e.stateNode,t,2),e=Eo(e,t,2),e!==null&&(St(e,2),Ef(e))}function Z(e,t,n){if(e.tag===3)pf(e,e,n);else for(;t!==null;){if(t.tag===3){pf(t,e,n);break}if(t.tag===1){var r=t.stateNode;if(typeof t.type.getDerivedStateFromError==`function`||typeof r.componentDidCatch==`function`&&(yd===null||!yd.has(r))){e=Qi(n,e),n=Fc(2),r=Eo(t,n,2),r!==null&&(Ic(n,r,t,e),St(r,2),Ef(r));break}}t=t.return}}function mf(e,t,n){var r=e.pingCache;if(r===null){r=e.pingCache=new td;var i=new Set;r.set(t,i)}else i=r.get(t),i===void 0&&(i=new Set,r.set(t,i));i.has(n)||(ad=!0,i.add(n),e=hf.bind(null,e,t,n),t.then(e,e))}function hf(e,t,n){var r=e.pingCache;r!==null&&r.delete(t),e.pingedLanes|=e.suspendedLanes&n,e.warmLanes&=~n,G===e&&(q&n)===n&&(Y===4||Y===3&&(q&62914560)===q&&300>Xe()-hd?W&2?ld|=n:Vd(e,0):ld|=n,dd===q&&(dd=0)),Ef(e)}function gf(e,t){t===0&&(t=bt()),e=Li(e,t),e!==null&&(St(e,t),Ef(e))}function _f(e){var t=e.memoizedState,n=0;t!==null&&(n=t.retryLane),gf(e,n)}function vf(e,t){var n=0;switch(e.tag){case 31:case 13:var r=e.stateNode,a=e.memoizedState;a!==null&&(n=a.retryLane);break;case 19:r=e.stateNode;break;case 22:r=e.stateNode._retryCache;break;default:throw Error(i(314))}r!==null&&r.delete(t),gf(e,n)}function yf(e,t){return Ke(e,t)}var bf=null,xf=null,Sf=!1,Cf=!1,wf=!1,Tf=0;function Ef(e){e!==xf&&e.next===null&&(xf===null?bf=xf=e:xf=xf.next=e),Cf=!0,Sf||(Sf=!0,Nf())}function Df(e,t){if(!wf&&Cf){wf=!0;do for(var n=!1,r=bf;r!==null;){if(!t){if(e!==0){var i=r.pendingLanes;if(i===0)var a=0;else{var o=r.suspendedLanes,s=r.pingedLanes;a=(1<<31-ct(42|e)+1)-1,a&=i&~(o&~s),a=a&201326741?a&201326741|1:a?a|2:0}a!==0&&(n=!0,Mf(r,a))}else a=q,a=gt(r,r===G?a:0,r.cancelPendingCommit!==null||r.timeoutHandle!==-1),!(a&3)||_t(r,a)||(n=!0,Mf(r,a))}r=r.next}while(n);wf=!1}}function Of(){kf()}function kf(){Cf=Sf=!1;var e=0;Tf!==0&&hp()&&(e=Tf);for(var t=Xe(),n=null,r=bf;r!==null;){var i=r.next,a=Af(r,t);a===0?(r.next=null,n===null?bf=i:n.next=i,i===null&&(xf=n)):(n=r,(e!==0||a&3)&&(Cf=!0)),r=i}X!==0&&X!==5||Df(e,!1),Tf!==0&&(Tf=0)}function Af(e,t){for(var n=e.suspendedLanes,r=e.pingedLanes,i=e.expirationTimes,a=e.pendingLanes&-62914561;0s)break;var u=c.transferSize,d=c.initiatorType;u&&ap(d)&&(c=c.responseEnd,o+=u*(c$m){s.length=o;break}f=new Promise(jp.bind(f)),s.push(f)}}}if(0`u`?null:document;function Tm(e,t,n){var r=wm;if(r&&typeof t==`string`&&t){var i=pn(t);i=`link[rel="`+e+`"][href="`+i+`"]`,typeof n==`string`&&(i+=`[crossorigin="`+n+`"]`),ym.has(i)||(ym.add(i),e={rel:e,crossOrigin:n,href:t},r.querySelector(i)===null&&(t=r.createElement(`link`),np(t,`link`,e),Kt(t),r.head.appendChild(t)))}}function Em(e){xm.D(e),Tm(`dns-prefetch`,e,null)}function Dm(e,t){xm.C(e,t),Tm(`preconnect`,e,t)}function Om(e,t,n){xm.L(e,t,n);var r=wm;if(r&&e&&t){var i=`link[rel="preload"][as="`+pn(t)+`"]`;t===`image`&&n&&n.imageSrcSet?(i+=`[imagesrcset="`+pn(n.imageSrcSet)+`"]`,typeof n.imageSizes==`string`&&(i+=`[imagesizes="`+pn(n.imageSizes)+`"]`)):i+=`[href="`+pn(e)+`"]`;var a=i;switch(t){case`style`:a=Pm(e);break;case`script`:a=Rm(e)}if(!(vm.has(a)||(e=C({rel:`preload`,href:t===`image`&&n&&n.imageSrcSet?void 0:e,as:t},n),vm.set(a,e),r.querySelector(i)!==null||t===`style`&&r.querySelector(Fm(a))||t===`script`&&r.querySelector(zm(a))))){var o=r.createElement(`link`);np(o,`link`,e),t===`style`&&(o[Bt]=!0,o.onload=o.onerror=function(){qt(o)}),Kt(o),r.head.appendChild(o)}}}function km(e,t){xm.m(e,t);var n=wm;if(n&&e){var r=t&&typeof t.as==`string`?t.as:`script`,i=`link[rel="modulepreload"][as="`+pn(r)+`"][href="`+pn(e)+`"]`,a=i;switch(r){case`audioworklet`:case`paintworklet`:case`serviceworker`:case`sharedworker`:case`worker`:case`script`:a=Rm(e)}if(!vm.has(a)&&(e=C({rel:`modulepreload`,href:e},t),vm.set(a,e),n.querySelector(i)===null)){switch(r){case`audioworklet`:case`paintworklet`:case`serviceworker`:case`sharedworker`:case`worker`:case`script`:if(n.querySelector(zm(a)))return}r=n.createElement(`link`),np(r,`link`,e),Kt(r),n.head.appendChild(r)}}}function Am(e,t,n){xm.S(e,t,n);var r=wm;if(r&&e){var i=Gt(r).hoistableStyles,a=Pm(e);t||=`default`;var o=i.get(a);if(!o){var s={loading:0,preload:null};if(o=r.querySelector(Fm(a)))s.loading=5;else{e=C({rel:`stylesheet`,href:e,"data-precedence":t},n),(n=vm.get(a))&&Hm(e,n);var c=o=r.createElement(`link`);Kt(c),np(c,`link`,e),c._p=new Promise(function(e,t){c.onload=e,c.onerror=t}),c.addEventListener(`load`,function(){s.loading|=1}),c.addEventListener(`error`,function(){s.loading|=2}),s.loading|=4,Vm(o,t,r)}o={type:`stylesheet`,instance:o,count:1,state:s},i.set(a,o)}}}function jm(e,t){xm.X(e,t);var n=wm;if(n&&e){var r=Gt(n).hoistableScripts,i=Rm(e),a=r.get(i);a||(a=n.querySelector(zm(i)),a||(e=C({src:e,async:!0},t),(t=vm.get(i))&&Um(e,t),a=n.createElement(`script`),Kt(a),np(a,`link`,e),n.head.appendChild(a)),a={type:`script`,instance:a,count:1,state:null},r.set(i,a))}}function Mm(e,t){xm.M(e,t);var n=wm;if(n&&e){var r=Gt(n).hoistableScripts,i=Rm(e),a=r.get(i);a||(a=n.querySelector(zm(i)),a||(e=C({src:e,async:!0,type:`module`},t),(t=vm.get(i))&&Um(e,t),a=n.createElement(`script`),Kt(a),np(a,`link`,e),n.head.appendChild(a)),a={type:`script`,instance:a,count:1,state:null},r.set(i,a))}}function Nm(e,t,n,r){var a=(a=Me.current)?bm(a):null;if(!a)throw Error(i(446));switch(e){case`meta`:case`title`:return null;case`style`:return typeof n.precedence==`string`&&typeof n.href==`string`?(n=Pm(n.href),t=Gt(a).hoistableStyles,r=t.get(n),r||(r={type:`style`,instance:null,count:0,state:null},t.set(n,r)),r):{type:`void`,instance:null,count:0,state:null};case`link`:if(n.rel===`stylesheet`&&typeof n.href==`string`&&typeof n.precedence==`string`){e=Pm(n.href);var o=Gt(a).hoistableStyles,s=o.get(e);if(s||(a=a.ownerDocument||a,s={type:`stylesheet`,instance:null,count:0,state:{loading:0,preload:null}},o.set(e,s),(o=a.querySelector(Fm(e)))?o._p||(s.instance=o,s.state.loading=5):(o=vm.get(e),o||(o={rel:`preload`,as:`style`,href:n.href,crossOrigin:n.crossOrigin,integrity:n.integrity,media:n.media,hrefLang:n.hrefLang,referrerPolicy:n.referrerPolicy},vm.set(e,o)),Lm(a,e,o,s.state))),t&&r===null)throw Error(i(528,``));return s}if(t&&r!==null)throw Error(i(529,``));return null;case`script`:return t=n.async,n=n.src,typeof n==`string`&&t&&typeof t!=`function`&&typeof t!=`symbol`?(n=Rm(n),t=Gt(a).hoistableScripts,r=t.get(n),r||(r={type:`script`,instance:null,count:0,state:null},t.set(n,r)),r):{type:`void`,instance:null,count:0,state:null};default:throw Error(i(444,e))}}function Pm(e){return`href="`+pn(e)+`"`}function Fm(e){return`link[rel="stylesheet"][`+e+`]`}function Im(e){return C({},e,{"data-precedence":e.precedence,precedence:null})}function Lm(e,t,n,r){if(t=e.querySelector(`link[rel="preload"][as="style"][`+t+`]`)){if(!0!==t[Bt]){r.loading=1;return}}else t=e.createElement(`link`),t[Bt]=!0,t.onload=t.onerror=qt.bind(null,t),np(t,`link`,n),Kt(t),e.head.appendChild(t);r.preload=t,t.addEventListener(`load`,function(){return r.loading|=1}),t.addEventListener(`error`,function(){return r.loading|=2})}function Rm(e){return`[src="`+pn(e)+`"]`}function zm(e){return`script[async]`+e}function Bm(e,t,n){if(t.count++,t.instance===null)switch(t.type){case`style`:var r=e.querySelector(`style[data-href~="`+pn(n.href)+`"]`);if(r)return t.instance=r,Kt(r),r;var a=C({},n,{"data-href":n.href,"data-precedence":n.precedence,href:null,precedence:null});return r=(e.ownerDocument||e).createElement(`style`),Kt(r),np(r,`style`,a),Vm(r,n.precedence,e),t.instance=r;case`stylesheet`:a=Pm(n.href);var o=e.querySelector(Fm(a));if(o)return t.state.loading|=4,t.instance=o,Kt(o),o;r=Im(n),(a=vm.get(a))&&Hm(r,a),o=(e.ownerDocument||e).createElement(`link`),Kt(o);var s=o;return s._p=new Promise(function(e,t){s.onload=e,s.onerror=t}),np(o,`link`,r),t.state.loading|=4,Vm(o,n.precedence,e),t.instance=o;case`script`:return o=Rm(n.src),(a=e.querySelector(zm(o)))?(t.instance=a,Kt(a),a):(r=n,(a=vm.get(o))&&(r=C({},n),Um(r,a)),e=e.ownerDocument||e,a=e.createElement(`script`),Kt(a),np(a,`link`,r),e.head.appendChild(a),t.instance=a);case`void`:return null;default:throw Error(i(443,t.type))}else t.type===`stylesheet`&&!(t.state.loading&4)&&(r=t.instance,t.state.loading|=4,Vm(r,n.precedence,e));return t.instance}function Vm(e,t,n){for(var r=n.querySelectorAll(`link[rel="stylesheet"][data-precedence],style[data-precedence]`),i=r.length?r[r.length-1]:null,a=i,o=0;o title`):null)}function qm(e,t,n){if(n===1||t.itemProp!=null)return!1;switch(e){case`meta`:case`title`:return!0;case`style`:if(typeof t.precedence!=`string`||typeof t.href!=`string`||t.href===``)break;return!0;case`link`:if(typeof t.rel!=`string`||typeof t.href!=`string`||t.href===``||t.onLoad||t.onError)break;switch(t.rel){case`stylesheet`:return e=t.disabled,typeof t.precedence==`string`&&e==null;default:return!0}case`script`:if(t.async&&typeof t.async!=`function`&&typeof t.async!=`symbol`&&!t.onLoad&&!t.onError&&t.src&&typeof t.src==`string`)return!0}return!1}function Jm(e,t){return e===`img`&&t.src!=null&&t.src!==``&&t.onLoad==null&&t.loading!==`lazy`}function Ym(e){return!(e.type===`stylesheet`&&!(e.state.loading&3))}function Xm(e){return(e.width||100)*(e.height||100)*(typeof devicePixelRatio==`number`?devicePixelRatio:1)*.25}function Zm(e,t){typeof t.decode==`function`&&(e.imgCount++,t.complete||(e.imgBytes+=Xm(t),e.suspenseyImages.push(t)),e=rh.bind(e),t.decode().then(e,e))}function Qm(e,t,n,r){if(n.type===`stylesheet`&&(typeof r.media!=`string`||!1!==matchMedia(r.media).matches)&&!(n.state.loading&4)){if(n.instance===null){var i=Pm(r.href),a=t.querySelector(Fm(i));if(a){t=a._p,typeof t==`object`&&t&&typeof t.then==`function`&&(e.count++,e=nh.bind(e),t.then(e,e)),n.state.loading|=4,n.instance=a,Kt(a);return}a=t.ownerDocument||t,r=Im(r),(i=vm.get(i))&&Hm(r,i),a=a.createElement(`link`),Kt(a);var o=a;o._p=new Promise(function(e,t){o.onload=e,o.onerror=t}),np(a,`link`,r),n.instance=a}e.stylesheets===null&&(e.stylesheets=new Map),e.stylesheets.set(n,t),(t=n.state.preload)&&!(n.state.loading&3)&&(e.count++,n=nh.bind(e),t.addEventListener(`load`,n),t.addEventListener(`error`,n))}}var $m=0;function eh(e,t){return e.stylesheets&&e.count===0&&ah(e,e.stylesheets),0$m?50:800)+t);return e.unsuspend=n,function(){e.unsuspend=null,clearTimeout(r),clearTimeout(i)}}:null}function th(e){if(e.count===0&&(e.imgCount===0||!e.waitingForImages)){if(e.stylesheets)ah(e,e.stylesheets);else if(e.unsuspend){var t=e.unsuspend;e.unsuspend=null,t()}}}function nh(){this.count--,th(this)}function rh(){this.imgCount--,th(this)}var ih=null;function ah(e,t){e.stylesheets=null,e.unsuspend!==null&&(e.count++,ih=new Map,t.forEach(oh,e),ih=null,nh.call(e))}function oh(e,t){if(!(t.state.loading&4)){var n=ih.get(e);if(n)var r=n.get(null);else{n=new Map,ih.set(e,n);for(var i=e.querySelectorAll(`link[data-precedence],style[data-precedence]`),a=0;a{function n(){if(!(typeof __REACT_DEVTOOLS_GLOBAL_HOOK__>`u`||typeof __REACT_DEVTOOLS_GLOBAL_HOOK__.checkDCE!=`function`))try{__REACT_DEVTOOLS_GLOBAL_HOOK__.checkDCE(n)}catch(e){console.error(e)}}n(),t.exports=h()})),_=c(u(),1),v=g(),y={width:1200,height:900,pad:{left:76,right:46,top:55,bottom:82}},b=[[0,-1],[0,1],[-1,0],[1,0],[-1,-1],[1,-1],[-1,1],[1,1]],ee=[22,34,48,64,82,104,130,160,196],x=(e,t,n,r)=>({cx:e,cy:t,width:n,height:r}),te=(e,t)=>Math.abs(e.cx-t.cx)*2e.cx-e.width/2>=t.left&&e.cx+e.width/2<=t.right&&e.cy-e.height/2>=t.top&&e.cy+e.height/2<=t.bottom;function S(e){let t=[...e.countries,...e.models],n=Math.min(...t.map(e=>e.x)),r=Math.max(...t.map(e=>e.x)),i=Math.min(...t.map(e=>e.y)),a=Math.max(...t.map(e=>e.y)),o=(r-n)*.17,s=(a-i)*.17,c={x0:n-o,x1:r+o,y0:i-s,y1:a+s*1.22},l={left:y.pad.left,right:y.width-y.pad.right,top:y.pad.top,bottom:y.height-y.pad.bottom};return{x:e=>l.left+(e-c.x0)/(c.x1-c.x0)*(l.right-l.left),y:e=>l.bottom-(e-c.y0)/(c.y1-c.y0)*(l.bottom-l.top),bounds:l,limits:c}}function re(e,t){let n=t===`zone`?15:t===`model`?13:11;return{width:e.length*n*.59+10,height:n+7}}function C(e,t){let{x:n,y:r,bounds:i}=t,a=[...e.countries.map(e=>x(n(e.x),r(e.y),10,10)),...e.models.map(e=>x(n(e.x),r(e.y),22,22))],o=[...e.models.filter(e=>e.label).map(e=>({id:`model:${e.name}`,point:e,text:e.label,kind:`model`})),...e.countries.filter(e=>e.label).map(e=>({id:`country:${e.name}`,point:e,text:e.name,kind:`country`})),...e.zone_hulls.map(e=>({id:`zone:${e.name}`,point:{x:e.label_anchor[0],y:e.label_anchor[1]},text:e.name,kind:`zone`}))],s={};for(let e of o){let t={x:n(e.point.x),y:r(e.point.y)},o=re(e.text,e.kind),c=null;for(let e of ee){for(let[n,r]of b){let s=x(t.x+n*(e+o.width/3),t.y+r*e,o.width,o.height);if(ne(s,i)&&!a.some(e=>te(s,e))){c=s;break}}if(c)break}if(!c)throw Error(`no collision-free label location for ${e.id}`);s[e.id]={...c,anchor:t,kind:e.kind,text:e.text},a.push(c)}return s}function ie(e,t){let n=e.slice(0,-1).map(([e,n])=>({x:t.x(e),y:t.y(n)}));if(n.length<3)throw Error(`a closed hull needs three points`);let r=e=>n[(e+n.length)%n.length],i=(e,t)=>Math.hypot(e.x-t.x,e.y-t.y),a=(e,t,n)=>({x:e.x+(t.x-e.x)*n,y:e.y+(t.y-e.y)*n}),o=e=>{let t=r(e-1),n=r(e),o=r(e+1),s=Math.min(.24,14/i(t,n),14/i(n,o));return{before:a(n,t,s),current:n,after:a(n,o,s)}},s=o(0),c=`M ${s.after.x} ${s.after.y}`;for(let e=1;e<=n.length;e+=1){let t=o(e%n.length);c+=` L ${t.before.x} ${t.before.y} Q ${t.current.x} ${t.current.y} ${t.after.x} ${t.after.y}`}return`${c} Z`}function w(e){let t=S(e),n=C(e,t),r=Object.entries(n);for(let[e,n]of r)if(!ne(n,t.bounds))throw Error(`label out of bounds: ${e}`);for(let e=0;e{var t=Symbol.for(`react.transitional.element`),n=Symbol.for(`react.fragment`);function r(e,n,r){var i=null;if(r!==void 0&&(i=``+r),n.key!==void 0&&(i=``+n.key),`key`in n)for(var a in r={},n)a!==`key`&&(r[a]=n[a]);else r=n;return n=r.ref,{$$typeof:t,type:e,key:i,ref:n===void 0?null:n,props:r}}e.Fragment=n,e.jsx=r,e.jsxs=r})),T=o(((e,t)=>{t.exports=ae()}))(),oe=`wvs/`;function se(e,t){return Object.keys(e).filter(e=>!t.has(e))}function ce({active:e,geometry:t,id:n=`model-tooltip`,panel:r}){if(!e)return null;let i=e.tooltipLeft??`${Math.min(86,t.x(e.x)/y.width*100)}%`,a=e.tooltipTop??`${Math.min(82,t.y(e.y)/y.height*100)}%`,o=e.provenance.release_created&&(0,T.jsxs)(`span`,{children:[`release `,e.provenance.release_created]});return(0,T.jsxs)(`aside`,{id:n,className:`tooltip`,style:{left:i,top:a},role:`status`,children:[(0,T.jsx)(`strong`,{children:e.name}),e.tooltipScore!=null&&(0,T.jsxs)(`span`,{children:[`HLE score: `,e.tooltipScore.toFixed(3)]}),r===`secular`&&(0,T.jsxs)(`span`,{children:[`Secular-Rational: `,e.y.toFixed(3)]}),r===`self-expression`&&(0,T.jsxs)(`span`,{children:[`Self-expression: `,e.tooltipValue.toFixed(3)]}),!r&&(0,T.jsxs)(T.Fragment,{children:[(0,T.jsxs)(`span`,{children:[`Self-expression/Survival: `,e.x.toFixed(3)]}),(0,T.jsxs)(`span`,{children:[`Traditional/Secular-Rational: `,e.y.toFixed(3)]})]}),e.tooltipScore==null&&o]})}function le({model:e,placement:t,geometry:n,setActive:r,clearActive:i,markerRef:a,logo:o}){let s=n.x(e.x),c=n.y(e.y),l=e.label?t[`model:${e.name}`]:null,u=l&&Math.hypot(l.cx-s,l.cy-c)>20;return(0,T.jsxs)(`g`,{ref:a,className:`model-mark`,"data-family":e.family,"data-model":e.name,"data-x":e.x,"data-y":e.y,tabIndex:`0`,role:`button`,"aria-label":`${e.name}, ${e.family}`,"aria-describedby":`model-tooltip`,onPointerEnter:()=>r(e),onPointerLeave:i,onFocus:()=>r(e),onBlur:i,children:[u&&(0,T.jsx)(`line`,{className:`leader`,x1:s,y1:c,x2:l.cx,y2:l.cy}),(0,T.jsx)(`circle`,{className:`model-ring`,cx:s,cy:c,r:`8`,stroke:e.color}),(0,T.jsx)(`image`,{href:`${oe}${o}`,x:s-5,y:c-5,width:`10`,height:`10`,preserveAspectRatio:`xMidYMid meet`,"aria-hidden":`true`,focusable:`false`}),l&&(0,T.jsx)(`text`,{className:`model-label`,x:l.cx,y:l.cy+4,textAnchor:`middle`,children:e.label})]})}function ue(e,t,n){let r=e.map(t),i=e.map(n);if(new Set(r).size<2)return null;let a=r.reduce((e,t)=>e+t,0)/r.length,o=i.reduce((e,t)=>e+t,0)/i.length,s=r.reduce((e,t,n)=>e+(t-a)*(i[n]-o),0)/r.reduce((e,t)=>e+(t-a)**2,0),c=o-s*a,l=i.reduce((e,t,n)=>e+(t-(c+s*r[n]))**2,0),u=i.reduce((e,t)=>e+(t-o)**2,0);return{slope:s,intercept:c,n:e.length,r2:u===0?null:1-l/u}}function de({data:e,hidden:t,field:n,title:r,axisMode:i}){let a=(0,_.useMemo)(()=>e.models.filter(e=>Number.isFinite(Date.parse(e.provenance.release_created??``))).toSorted((e,t)=>e.provenance.release_created.localeCompare(t.provenance.release_created)||e.name.localeCompare(t.name)),[e]),o=(0,_.useMemo)(()=>e.models.filter(e=>Number.isFinite(e.provenance.hle_score)).toSorted((e,t)=>e.provenance.hle_score-t.provenance.hle_score||e.name.localeCompare(t.name)),[e]),s=i===`release-date`?a:o,c=s.filter(e=>!t.has(e.family)),l=e=>n===`x`?-e.x:e.y,u=e=>i===`release-date`?Date.parse(e.provenance.release_created):e.provenance.hle_score,d=ue(c,u,l),[f,p]=(0,_.useState)(null),m=1200,h=s.map(u),g=s.map(l),v=Math.min(...h),y=Math.max(...h),S=Math.min(...g),re=Math.max(...g),C=e=>96+(e-v)/(y-v||1)*1069,ie=e=>42+(re-e)/(re-S||1)*230,w=`release-${n}`,ae=`${w}-tooltip`,le=n===`y`?e.axis.y:[...e.axis.x].reverse(),de=c.map(u),fe=d&&[Math.min(...de),Math.max(...de)].map(e=>[C(e),ie(d.intercept+d.slope*e)]),pe=(0,_.useMemo)(()=>{if(i!==`release-date`)return[];let e=-1/0;return c.filter(e=>Number.isFinite(e.provenance.hle_score)).filter(t=>t.provenance.hle_score<=e?!1:(e=t.provenance.hle_score,!0))},[i,c]),me=(0,_.useMemo)(()=>{let e={left:96,right:1165,top:42,bottom:272},t=[...s.map(e=>x(C(u(e)),ie(l(e)),16,16)),...d?[x(162,51,128,18)]:[]],n={};for(let r of pe){let i={x:C(u(r)),y:ie(l(r))},a={width:r.name.length*8+12,height:18},o=null;for(let n of ee){for(let[r,s]of b){let c=x(i.x+r*(n+a.width/3),i.y+s*n,a.width,a.height);if(ne(c,e)&&!t.some(e=>te(c,e))){o=c;break}}if(o)break}if(!o)for(let n=e.top+a.height/2;n<=e.bottom-a.height/2&&!o;n+=a.height+4)for(let r=e.left+a.width/2;r<=e.right-a.width/2;r+=12){let e=x(r,n,a.width,a.height);if(!t.some(t=>te(e,t))){o=e;break}}o||=x(i.x,Math.max(e.top+a.height/2,i.y-a.height),a.width,a.height),n[r.name]={...o,anchor:i},t.push(o)}return n},[pe,s,v,y,S,re]),he=(e,t)=>{let n=t.currentTarget.closest(`.scatter-shell`).getBoundingClientRect(),r=t.currentTarget.getBoundingClientRect();p({...e,tooltipValue:l(e),tooltipScore:i===`capability`?e.provenance.hle_score:null,tooltipLeft:`${Math.min(82,(r.left-n.left)/n.width*100)}%`,tooltipTop:`${Math.min(78,(r.top-n.top)/n.height*100)}%`})};return(0,T.jsxs)(`section`,{className:`release-panel`,"data-coordinate":n,"aria-labelledby":`${w}-heading`,children:[(0,T.jsx)(`h2`,{id:`${w}-heading`,children:r}),(0,T.jsxs)(`div`,{className:`scatter-shell`,children:[(0,T.jsxs)(`svg`,{id:`${w}-svg`,viewBox:`0 0 ${m} 320`,role:`img`,"aria-labelledby":`${w}-svg-title ${w}-svg-desc`,"data-axis-mode":i,"data-panel-model-count":c.length,"data-omitted-model-count":e.models.length-s.length,"data-fit-n":d?.n??0,"data-fit-r2":d?.r2??``,"data-fit-slope":d?.slope??``,children:[(0,T.jsx)(`title`,{id:`${w}-svg-title`,children:r}),(0,T.jsxs)(`desc`,{id:`${w}-svg-desc`,children:[`Scatter plot with `,i===`release-date`?`release date`:`HLE score`,` on the horizontal axis and `,le.join(` to `),` increasing upward on the vertical axis. `,c.length,` of `,s.length,` matched models are visible from `,se(Object.groupBy(e.models,e=>e.family),t).join(`, `)||`no families`,`; `,e.models.length-s.length,` plotted models lack this x value and are omitted only here. Each white-ring logo mark is a model. `,d?`The thin line is an ordinary least squares descriptive fit to the ${d.n} currently visible matched models${d.r2===null?`; R squared is unavailable because the y values are constant`:`; R squared is ${d.r2.toFixed(2)}`}.`:`The fit is hidden because fewer than two distinct x values are visible.`,` Hover or keyboard focus a mark for model-specific details.`]}),(0,T.jsxs)(`defs`,{children:[(0,T.jsx)(`marker`,{id:`${w}-arrow`,markerWidth:`8`,markerHeight:`8`,refX:`6`,refY:`3`,orient:`auto`,children:(0,T.jsx)(`path`,{d:`M0,0 L0,6 L6,3 z`,className:`scatter-arrow`})}),(0,T.jsx)(`clipPath`,{id:`${w}-fit-clip`,children:(0,T.jsx)(`rect`,{x:96,y:42,width:1069,height:230})})]}),(0,T.jsx)(`rect`,{className:`canvas`,width:m,height:320}),(0,T.jsx)(`g`,{className:`scatter-grid`,children:Array.from({length:5},(e,t)=>(0,T.jsx)(`line`,{x1:96,x2:1165,y1:42+t*230/4,y2:42+t*230/4},t))}),(0,T.jsx)(`line`,{className:`scatter-axis`,x1:96,x2:1165,y1:272,y2:272}),(0,T.jsx)(`line`,{className:`scatter-axis`,x1:96,x2:96,y1:272,y2:42,markerEnd:`url(#${w}-arrow)`}),(0,T.jsxs)(`text`,{className:`scatter-y-label`,x:`17`,y:157,textAnchor:`middle`,transform:`rotate(-90 17 157)`,children:[le.join(` -> `),` `,`↑`]}),(0,T.jsx)(`text`,{className:`scatter-tick`,x:96,y:302,children:i===`release-date`?new Date(v).toISOString().slice(0,10):v.toFixed(1)}),(0,T.jsx)(`text`,{className:`scatter-tick`,x:1165,y:302,textAnchor:`end`,children:i===`release-date`?new Date(y).toISOString().slice(0,10):y.toFixed(1)}),(0,T.jsx)(`text`,{className:`scatter-tick`,x:88,y:46,textAnchor:`end`,children:re.toFixed(2)}),(0,T.jsx)(`text`,{className:`scatter-tick`,x:88,y:272,textAnchor:`end`,children:S.toFixed(2)}),fe&&(0,T.jsxs)(`g`,{className:`release-fit`,"data-fit-n":d.n,"data-fit-r2":d.r2??``,children:[(0,T.jsx)(`line`,{clipPath:`url(#${w}-fit-clip)`,x1:fe[0][0],y1:fe[0][1],x2:fe[1][0],y2:fe[1][1]}),` `,(0,T.jsxs)(`text`,{x:102,y:55,children:[`OLS, n=`,d.n,`, R² `,d.r2===null?`unavailable`:d.r2.toFixed(2)]})]}),!d&&(0,T.jsx)(`text`,{className:`release-fit-unavailable`,x:102,y:55,children:`Fit unavailable: fewer than two release dates`}),s.map(n=>(0,T.jsxs)(`g`,{className:`release-mark`,"data-family":n.family,"data-release-model":n.name,"data-release-date":n.provenance.release_created,"data-hle-score":n.provenance.hle_score,"data-coordinate-value":l(n),display:t.has(n.family)?`none`:`inline`,tabIndex:`0`,role:`button`,"aria-label":`${n.name}, ${i===`release-date`?n.provenance.release_created:`HLE score ${n.provenance.hle_score}`}`,"aria-describedby":ae,onPointerEnter:e=>he(n,e),onPointerLeave:()=>p(null),onFocus:e=>he(n,e),onBlur:()=>p(null),children:[(0,T.jsx)(`circle`,{className:`model-ring`,cx:C(u(n)),cy:ie(l(n)),r:`8`,stroke:n.color}),(0,T.jsx)(`image`,{href:`${oe}${e.logos[n.family]}`,x:C(u(n))-5,y:ie(l(n))-5,width:`10`,height:`10`,preserveAspectRatio:`xMidYMid meet`,"aria-hidden":`true`,focusable:`false`})]},n.name)),i===`release-date`&&pe.map(e=>(0,T.jsx)(`g`,{className:`frontier-label`,"data-frontier-model":e.name,children:(0,T.jsx)(`text`,{x:me[e.name].cx,y:me[e.name].cy+4,textAnchor:`middle`,children:e.name})},`frontier-${e.name}`))]}),(0,T.jsx)(ce,{active:f,geometry:null,id:ae,panel:n===`y`?`secular`:`self-expression`})]})]})}function fe({data:e,hidden:t,axisMode:n,setAxisMode:r}){let i=n===`release-date`?`Release date`:e.capability_x.label;return(0,T.jsxs)(`section`,{className:`release-panels`,"aria-label":`Release-date and capability scatter panels`,children:[(0,T.jsxs)(`div`,{className:`release-axis-selector`,children:[(0,T.jsxs)(`label`,{children:[`Release-panel x axis `,(0,T.jsxs)(`select`,{"aria-label":`Release-panel x axis`,"aria-controls":`release-y-svg release-x-svg`,value:n,onChange:e=>r(e.target.value),children:[(0,T.jsx)(`option`,{value:`release-date`,children:`Release date`}),(0,T.jsx)(`option`,{value:`capability`,children:e.capability_x.label})]})]}),(0,T.jsxs)(`span`,{children:[(0,T.jsx)(`a`,{href:e.capability_x.source_url,children:`Artificial Analysis HLE score`}),`, saved `,e.capability_x.fetched_utc.slice(0,10),`. `,n===`capability`?`${e.capability_x.matched_models} matched, ${e.models.length-e.capability_x.matched_models} omitted.`:`Release-date labels mark running HLE score highs among shown mapped models.`]})]}),(0,T.jsx)(de,{data:e,hidden:t,field:`y`,axisMode:n,title:`${i} vs Secular-Rational`}),(0,T.jsx)(de,{data:e,hidden:t,field:`x`,axisMode:n,title:`${i} vs Self-expression`})]})}function pe({data:e}){let t=new URLSearchParams(location.search),[n,r]=(0,_.useState)(()=>new Set(t.get(`hide`)?.split(`,`).filter(Boolean))),[i,a]=(0,_.useState)(()=>t.get(`axis`)===`capability`?`capability`:`release-date`),[o,s]=(0,_.useState)(()=>e.models.find(e=>e.name===(t.get(`tooltip`)||t.get(`focus`)))??null),c=t.get(`focus`),l=(0,_.useRef)(null),u=(0,_.useMemo)(()=>Object.groupBy(e.models,e=>e.family),[e]),{labels:d,geometry:f}=(0,_.useMemo)(()=>w(e),[e]),p=se(u,n),m=e.models.filter(e=>!n.has(e.family)).length;(0,_.useEffect)(()=>{c&&l.current?.focus()},[c]);let h=e=>{e.currentTarget.matches(`:focus`)||s(null)},g=e=>r(t=>{let n=new Set(t);return n.has(e)?n.delete(e):n.add(e),n}),v=f.x(e.median.x),b=f.y(e.median.y);return(0,T.jsxs)(T.Fragment,{children:[(0,T.jsx)(`section`,{className:`controls`,"aria-label":`Model-family visibility`,children:Object.entries(u).map(([t])=>{let r=!n.has(t);return(0,T.jsxs)(`button`,{className:`chip`,type:`button`,"aria-pressed":r,onClick:()=>g(t),children:[(0,T.jsx)(`img`,{src:`${oe}${e.logos[t]}`,alt:``,"aria-hidden":`true`}),t]},t)})}),(0,T.jsxs)(`div`,{className:`chart-shell`,children:[(0,T.jsxs)(`svg`,{viewBox:`0 0 ${y.width} ${y.height}`,role:`img`,"aria-labelledby":`map-svg-title map-svg-desc`,"data-median-x":e.median.x,"data-median-y":e.median.y,"data-visible-model-count":m,children:[(0,T.jsx)(`title`,{id:`map-svg-title`,children:`Moral Maps: Where Do Frontier Models' Cultural Values Lie?`}),(0,T.jsxs)(`desc`,{id:`map-svg-desc`,children:[`World Values Survey cultural map. Horizontal direction runs from Self-expression on the left to Survival on the right. Vertical direction runs from Traditional below to Secular-Rational above. Coloured dots are selected WVS countries, outlines are cultural regions, and white-ring logo marks are models. `,m,` model marks are visible from `,p.join(`, `)||`no families`,`. Use the family controls to hide marks and labels. Hover or keyboard focus a model for model-specific details.`]}),(0,T.jsx)(`defs`,{children:(0,T.jsx)(`marker`,{id:`arrow`,markerWidth:`8`,markerHeight:`8`,refX:`6`,refY:`3`,orient:`auto`,children:(0,T.jsx)(`path`,{d:`M0,0 L0,6 L6,3 z`,className:`arrow`})})}),(0,T.jsx)(`rect`,{className:`canvas`,width:y.width,height:y.height}),(0,T.jsxs)(`g`,{className:`grid`,children:[Array.from({length:8},(e,t)=>(0,T.jsx)(`line`,{x1:f.bounds.left+t*(f.bounds.right-f.bounds.left)/7,y1:f.bounds.top,x2:f.bounds.left+t*(f.bounds.right-f.bounds.left)/7,y2:f.bounds.bottom},`v${t}`)),Array.from({length:6},(e,t)=>(0,T.jsx)(`line`,{x1:f.bounds.left,y1:f.bounds.top+t*(f.bounds.bottom-f.bounds.top)/5,x2:f.bounds.right,y2:f.bounds.top+t*(f.bounds.bottom-f.bounds.top)/5},`h${t}`))]}),(0,T.jsx)(`line`,{className:`median`,x1:f.bounds.left,y1:b,x2:f.bounds.right,y2:b}),(0,T.jsx)(`line`,{className:`median`,x1:v,y1:f.bounds.top,x2:v,y2:f.bounds.bottom}),e.zone_hulls.map(e=>(0,T.jsx)(`path`,{className:`zone`,d:ie(e.points,f),stroke:e.color},e.name)),e.countries.map(e=>(0,T.jsxs)(`g`,{children:[(0,T.jsx)(`circle`,{className:`country`,"data-country":e.name,"data-x":e.x,"data-y":e.y,cx:f.x(e.x),cy:f.y(e.y),r:`3.5`,fill:e.color}),e.label&&(0,T.jsx)(`text`,{className:`country-label`,x:d[`country:${e.name}`].cx,y:d[`country:${e.name}`].cy+4,textAnchor:`middle`,children:e.name})]},e.name)),e.zone_hulls.map(e=>(0,T.jsx)(`text`,{className:`zone-label`,x:d[`zone:${e.name}`].cx,y:d[`zone:${e.name}`].cy+5,textAnchor:`middle`,fill:e.color,children:e.name},e.name)),Object.entries(u).map(([t,r])=>(0,T.jsx)(`g`,{"data-family":t,display:n.has(t)?`none`:`inline`,children:r.map(t=>(0,T.jsx)(le,{model:t,placement:d,geometry:f,setActive:s,clearActive:h,markerRef:t.name===c?l:null,logo:e.logos[t.family]},t.name))},t)),(0,T.jsxs)(`g`,{className:`poles`,children:[(0,T.jsx)(`line`,{x1:v,y1:`62`,x2:v,y2:f.bounds.top,markerEnd:`url(#arrow)`}),(0,T.jsx)(`line`,{x1:v,y1:f.bounds.bottom,x2:v,y2:`838`,markerEnd:`url(#arrow)`}),(0,T.jsx)(`line`,{x1:`64`,y1:b,x2:f.bounds.left,y2:b,markerEnd:`url(#arrow)`}),(0,T.jsx)(`line`,{x1:f.bounds.right,y1:b,x2:`1184`,y2:b,markerEnd:`url(#arrow)`}),(0,T.jsx)(`text`,{x:v,y:`40`,textAnchor:`middle`,children:e.axis.y[1]}),(0,T.jsx)(`text`,{x:v,y:`870`,textAnchor:`middle`,children:e.axis.y[0]}),(0,T.jsx)(`text`,{x:`25`,y:b+7,children:e.axis.x[0]}),(0,T.jsx)(`text`,{x:`1136`,y:b+7,textAnchor:`end`,children:e.axis.x[1]})]}),(0,T.jsx)(`text`,{className:`map-title`,x:f.bounds.left+8,y:f.bounds.bottom-78,children:e.title.split(` +`).map((e,t)=>(0,T.jsx)(`tspan`,{x:f.bounds.left+8,dy:t?17:0,children:e},e))}),(0,T.jsx)(`text`,{className:`map-note`,x:f.bounds.left+8,y:f.bounds.bottom-34,textAnchor:`start`,children:e.note.split(` +`).map((e,t)=>(0,T.jsx)(`tspan`,{x:f.bounds.left+8,dy:t?11:0,children:e},e))})]}),(0,T.jsx)(ce,{active:o,geometry:f})]}),(0,T.jsx)(`p`,{className:`map-explanation`,children:`Since 1981, the World Values Survey has asked people in about ninety countries the same questions. Its axes run from Traditional to Secular-Rational and from Self-expression to Survival.`}),(0,T.jsx)(fe,{data:e,hidden:n,axisMode:i,setAxisMode:a}),(0,T.jsxs)(`p`,{className:`caption`,children:[`Use the family controls to compare saved rated coordinates. The lines are weak descriptive correlations for the matched models currently shown; n and R² update with visibility. Hover or keyboard focus a mark for its values. See the `,(0,T.jsx)(`a`,{href:`https://github.com/wassname/moral-maps`,children:`code and records`}),`.`]})]})}function me(){let[e,t]=(0,_.useState)(null);return(0,_.useEffect)(()=>{fetch(`wvs/wvs_map_data.json`).then(e=>e.json()).then(t)},[]),(0,T.jsxs)(`main`,{children:[(0,T.jsx)(`h1`,{children:`Moral Maps: Where Do Frontier Models' Cultural Values Lie?`}),(0,T.jsxs)(`p`,{className:`lede`,children:[`We start with the `,(0,T.jsx)(`a`,{href:`https://www.worldvaluessurvey.org/`,children:`World Values Survey`}),`, a map of human values across about ninety countries.`]}),e&&(0,T.jsx)(pe,{data:e})]})}(0,v.createRoot)(document.getElementById(`root`)).render((0,T.jsx)(me,{})); \ No newline at end of file diff --git a/docs/img/wvs/wvs_map_iw.png b/docs/img/wvs/wvs_map_iw.png index 045ec28..cee2e13 100644 Binary files a/docs/img/wvs/wvs_map_iw.png and b/docs/img/wvs/wvs_map_iw.png differ diff --git a/docs/img/wvs/wvs_map_iw.svg b/docs/img/wvs/wvs_map_iw.svg index 7103344..f819c83 100644 --- a/docs/img/wvs/wvs_map_iw.svg +++ b/docs/img/wvs/wvs_map_iw.svg @@ -6,7 +6,7 @@ - 2026-09-17T07:39:16.812532 + 2026-09-18T01:46:25.467471 image/svg+xml @@ -40,12 +40,12 @@ z +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #c9c4b4; stroke-linecap: square"/> +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #c9c4b4; stroke-linecap: square"/> +" clip-path="url(#p890f8b3f7a)" style="fill: none; opacity: 0.9; stroke: #4e79a7; stroke-width: 1.8; stroke-linejoin: miter"/> +" clip-path="url(#p890f8b3f7a)" style="fill: none; opacity: 0.9; stroke: #9c755f; stroke-width: 1.8; stroke-linejoin: miter"/> +" clip-path="url(#p890f8b3f7a)" style="fill: none; opacity: 0.9; stroke: #e15759; stroke-width: 1.8; stroke-linejoin: miter"/> +" clip-path="url(#p890f8b3f7a)" style="fill: none; opacity: 0.9; stroke: #f28e2b; stroke-width: 1.8; stroke-linejoin: miter"/> +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #9a958a; stroke-width: 0.6; stroke-linecap: square"/> +L 273.72292 121.46959 +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #9a958a; stroke-width: 0.6; stroke-linecap: square"/> +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #9a958a; stroke-width: 0.6; stroke-linecap: square"/> - + +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #9a958a; stroke-width: 0.6; stroke-linecap: square"/> +L 244.27119 93.774912 +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #9a958a; stroke-width: 0.6; stroke-linecap: square"/> +L 88.378442 133.000492 +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #9a958a; stroke-width: 0.6; stroke-linecap: square"/> +L 96.275705 115.422669 +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #9a958a; stroke-width: 0.6; stroke-linecap: square"/> - + +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #9a958a; stroke-width: 0.6; stroke-linecap: square"/> +L 110.927272 92.538827 +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #9a958a; stroke-width: 0.6; stroke-linecap: square"/> - + +" clip-path="url(#p890f8b3f7a)" style="fill: none; stroke: #9a958a; stroke-width: 0.6; stroke-linecap: square"/> - - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + @@ -6216,548 +6348,548 @@ z " style="fill: #7c3aed"/> - - @@ -7342,292 +7474,288 @@ z " style="fill: #14b8a6"/> - - @@ -8340,2078 +8468,2454 @@ z " style="fill: #ec4899"/> - - - - - - - - +M 351.590522 135.219985 +L 354.349584 135.219985 +L 354.349584 136.498266 +L 351.590522 136.498266 +L 351.590522 135.219985 +z +M 357.627553 131.611547 +L 360.752241 131.611547 +L 360.752241 138.450141 +L 359.165991 138.450141 +L 359.165991 132.645141 +L 357.962241 132.645141 +Q 357.627553 132.645141 357.495366 132.766078 +Q 357.364584 132.887016 357.364584 133.185141 +L 357.364584 133.528266 +L 358.599272 133.528266 +L 358.599272 134.653266 +L 357.364584 134.653266 +L 357.364584 138.450141 +L 355.790991 138.450141 +L 355.790991 134.653266 +L 355.009116 134.653266 +L 355.009116 133.528266 +L 355.790991 133.528266 +L 355.790991 133.185141 +Q 355.790991 132.380766 356.239584 131.99686 +Q 356.688178 131.611547 357.627553 131.611547 +z +M 364.470366 136.235297 +Q 363.978178 136.235297 363.729272 136.402641 +Q 363.481772 136.568578 363.481772 136.894828 +Q 363.481772 137.192953 363.681459 137.36311 +Q 363.882553 137.53186 364.238334 137.53186 +Q 364.681303 137.53186 364.985053 137.214047 +Q 365.288803 136.894828 365.288803 136.415297 +L 365.288803 136.235297 +L 364.470366 136.235297 +z +M 366.875053 135.64186 +L 366.875053 138.450141 +L 365.288803 138.450141 +L 365.288803 137.720297 +Q 364.972397 138.168891 364.575834 138.374203 +Q 364.180678 138.57811 363.613959 138.57811 +Q 362.848959 138.57811 362.372241 138.132328 +Q 361.895522 137.685141 361.895522 136.973578 +Q 361.895522 136.107328 362.490366 135.703735 +Q 363.086616 135.298735 364.360678 135.298735 +L 365.288803 135.298735 +L 365.288803 135.176391 +Q 365.288803 134.802328 364.993491 134.62936 +Q 364.699584 134.454985 364.075209 134.454985 +Q 363.570366 134.454985 363.134428 134.556235 +Q 362.699897 134.657485 362.325834 134.859985 +L 362.325834 133.660453 +Q 362.832084 133.536703 363.341147 133.473422 +Q 363.851616 133.410141 364.360678 133.410141 +Q 365.692397 133.410141 366.283022 133.934672 +Q 366.875053 134.459203 366.875053 135.64186 +z +M 372.183647 133.681547 +L 372.183647 134.87686 +Q 371.678803 134.665922 371.207709 134.560453 +Q 370.738022 134.454985 370.320366 134.454985 +Q 369.871772 134.454985 369.653803 134.567485 +Q 369.437241 134.679985 369.437241 134.912016 +Q 369.437241 135.10186 369.601772 135.20311 +Q 369.766303 135.302953 370.192397 135.352172 +L 370.469428 135.391547 +Q 371.678803 135.544828 372.095053 135.896391 +Q 372.512709 136.247953 372.512709 137.000297 +Q 372.512709 137.786391 371.931928 138.182953 +Q 371.352553 138.57811 370.202241 138.57811 +Q 369.714272 138.57811 369.192553 138.500766 +Q 368.672241 138.423422 368.122397 138.270141 +L 368.122397 137.074828 +Q 368.593491 137.302641 369.087084 137.417953 +Q 369.582084 137.53186 370.091147 137.53186 +Q 370.553803 137.53186 370.785834 137.405297 +Q 371.019272 137.277328 371.019272 137.02561 +Q 371.019272 136.814672 370.858959 136.712016 +Q 370.698647 136.60936 370.219116 136.551703 +L 369.942084 136.516547 +Q 368.891616 136.38436 368.469741 136.028578 +Q 368.047866 135.672797 368.047866 134.947172 +Q 368.047866 134.165297 368.583647 133.788422 +Q 369.120834 133.410141 370.227553 133.410141 +Q 370.663491 133.410141 371.141616 133.476235 +Q 371.621147 133.540922 372.183647 133.681547 +z +M 378.642553 135.453422 +L 378.642553 138.450141 +L 377.060522 138.450141 +L 377.060522 137.962172 +L 377.060522 136.164985 +Q 377.060522 135.519516 377.032397 135.277641 +Q 377.004272 135.035766 376.933959 134.92186 +Q 376.841147 134.767172 376.682241 134.681391 +Q 376.524741 134.59561 376.322241 134.59561 +Q 375.830053 134.59561 375.548803 134.976703 +Q 375.267553 135.356391 375.267553 136.028578 +L 375.267553 138.450141 +L 373.695366 138.450141 +L 373.695366 131.611547 +L 375.267553 131.611547 +L 375.267553 134.248266 +Q 375.623334 133.817953 376.022709 133.614047 +Q 376.423491 133.410141 376.907241 133.410141 +Q 377.759428 133.410141 378.200991 133.933266 +Q 378.642553 134.454985 378.642553 135.453422 +z +" style="fill: #0ea5e9; stroke: #ffffff; stroke-width: 2"/> + - - - + - +" style="fill: #dc2626"/> - - + + - - + + + + + + + + + + + + - - - - - - - - - - - - - - - - - - - - - - - - - - + + + + + + + + + + + + + + + + + + + + + + + + + + + + + - - + + - - - - - - - - - - - - - - - - - - - - - - - - - - + + + + + + + + + + + + + + + + + + + + + + + + + + + + - - + + - - - - + + - - - - - + - - - - + + + + + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - + + - - + + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + - + diff --git a/docs/img/wvs/wvs_model_ci.md b/docs/img/wvs/wvs_model_ci.md index dd6e88b..eb7b293 100644 --- a/docs/img/wvs/wvs_model_ci.md +++ b/docs/img/wvs/wvs_model_ci.md @@ -1,66 +1,110 @@ -| model | x self-expr | y secular | x 95%CI | y 95%CI | -|:-------------------------------------|--------------:|------------:|----------:|----------:| -| qwen3.6-27b (rated) | +0.46 | +0.68 | +0.18 | +0.14 | -| qwen3.5-35b-a3b (rated) | +0.54 | +0.55 | +0.14 | +0.18 | -| qwen3.6-max-preview (rated) | +0.48 | +0.69 | +0.16 | +0.14 | -| muse-spark-1.3 (rated) | +0.43 | +0.73 | +0.18 | +0.12 | -| qwen3.8-27b (rated) | +0.43 | +0.69 | +0.16 | +0.14 | -| qwen3.8-flash (rated) | +0.42 | +0.63 | +0.14 | +0.15 | -| gemma-4-31b-it (rated) | +0.36 | +0.66 | +0.16 | +0.13 | -| qwen3.5-122b-a10b (rated) | +0.48 | +0.62 | +0.15 | +0.14 | -| grok-4.20 (rated) | +0.59 | +0.62 | +0.11 | +0.17 | -| qwen3.5-plus-20260420 (rated) | +0.59 | +0.65 | +0.12 | +0.16 | -| gemini-2.5-pro (rated) | +0.47 | +0.70 | +0.13 | +0.14 | -| grok-4.3 (rated) | +0.44 | +0.73 | +0.14 | +0.13 | -| deepseek-v4-flash (rated) | +0.57 | +0.64 | +0.11 | +0.16 | -| gpt-5.4 (rated) | +0.45 | +0.68 | +0.13 | +0.14 | -| deepseek-v4.1-flash (rated) | +0.54 | +0.60 | +0.10 | +0.16 | -| qwen3-coder-30b-a3b-instruct (rated) | +0.50 | +0.58 | +0.09 | +0.17 | -| qwen3-30b-a3b (rated) | +0.67 | +0.61 | +0.07 | +0.19 | -| gpt-6-astra (rated) | +0.46 | +0.68 | +0.15 | +0.11 | -| mistral-large-2512 (rated) | +0.65 | +0.64 | +0.08 | +0.18 | -| gemma-3-27b-it (rated) | +0.60 | +0.60 | +0.09 | +0.17 | -| llama-4-maverick (rated) | +0.64 | +0.64 | +0.08 | +0.18 | -| qwen3-coder (rated) | +0.60 | +0.62 | +0.08 | +0.18 | -| qwen3.7-flash (rated) | +0.65 | +0.59 | +0.10 | +0.16 | -| qwen3-vl-235b-a22b-instruct (rated) | +0.60 | +0.55 | +0.05 | +0.21 | -| qwen-plus (rated) | +0.60 | +0.63 | +0.07 | +0.18 | -| gpt-5.3-chat (rated) | +0.50 | +0.67 | +0.09 | +0.16 | -| deepseek-v4-pro (rated) | +0.55 | +0.73 | +0.15 | +0.10 | -| llama-4-scout (rated) | +0.62 | +0.53 | +0.05 | +0.20 | -| qwen3-vl-30b-a3b-instruct (rated) | +0.53 | +0.53 | +0.06 | +0.19 | -| qwen3.5-397b-a17b (rated) | +0.54 | +0.65 | +0.08 | +0.17 | -| qwen3.5-plus-02-15 (rated) | +0.54 | +0.66 | +0.08 | +0.17 | -| glm-5.3 (rated) | +0.55 | +0.64 | +0.12 | +0.13 | -| qwen3.7-plus (rated) | +0.51 | +0.67 | +0.10 | +0.15 | -| qwen3-30b-a3b-instruct-2507 (rated) | +0.65 | +0.54 | +0.05 | +0.20 | -| gpt-5.5 (rated) | +0.42 | +0.76 | +0.17 | +0.07 | -| qwen3.6-35b-a3b (rated) | +0.61 | +0.59 | +0.10 | +0.14 | -| qwen3.6-flash (rated) | +0.65 | +0.58 | +0.09 | +0.15 | -| qwen3.6-plus (rated) | +0.52 | +0.67 | +0.08 | +0.15 | -| qwen3-235b-a22b (rated) | +0.64 | +0.56 | +0.07 | +0.17 | -| qwen-2.5-72b-instruct (rated) | +0.60 | +0.60 | +0.06 | +0.17 | -| qwen-plus-2025-07-28 (rated) | +0.64 | +0.51 | +0.07 | +0.16 | -| qwen3.7-max (rated) | +0.50 | +0.65 | +0.11 | +0.12 | -| qwen3-235b-a22b-2507 (rated) | +0.64 | +0.57 | +0.07 | +0.16 | -| qwen3.5-9b (rated) | +0.50 | +0.59 | +0.09 | +0.14 | -| qwen2.5-vl-72b-instruct (rated) | +0.58 | +0.60 | +0.07 | +0.15 | -| qwen3-vl-32b-instruct (rated) | +0.60 | +0.53 | +0.02 | +0.20 | -| gpt-5.6-sol (rated) | +0.53 | +0.67 | +0.09 | +0.13 | -| qwen3-max-thinking (rated) | +0.58 | +0.70 | +0.07 | +0.14 | -| qwen3-coder-plus (rated) | +0.62 | +0.60 | +0.07 | +0.14 | -| inkling (rated) | +0.56 | +0.69 | +0.10 | +0.11 | -| qwen3-vl-8b-instruct (rated) | +0.61 | +0.60 | +0.06 | +0.15 | -| qwen-2.5-7b-instruct (rated) | +0.55 | +0.59 | +0.05 | +0.16 | -| kimi-k3 (rated) | +0.62 | +0.67 | +0.07 | +0.14 | -| qwen3.5-27b (rated) | +0.60 | +0.65 | +0.05 | +0.16 | -| qwen3-32b (rated) | +0.61 | +0.54 | +0.06 | +0.14 | -| qwen3-14b (rated) | +0.61 | +0.58 | +0.03 | +0.16 | -| qwen3-coder-next (rated) | +0.64 | +0.58 | +0.05 | +0.13 | -| qwen3-next-80b-a3b-instruct (rated) | +0.67 | +0.70 | +0.05 | +0.12 | -| claude-opus-4.7 (rated) | +0.59 | +0.60 | +0.07 | +0.09 | -| qwen3-8b (rated) | +0.62 | +0.54 | +0.03 | +0.12 | -| claude-opus-4.8 (rated) | +0.61 | +0.58 | +0.07 | +0.07 | -| gemini-3.7-flash (rated) | +0.50 | +0.63 | +0.03 | +0.10 | -| claude-fable-5.1 (rated) | +0.58 | +0.61 | +0.07 | +0.05 | -| claude-opus-4.6 (rated) | +0.63 | +0.63 | +0.04 | +0.05 | +| model | x self-expr | y secular | x 95%CI | y 95%CI | +|:--------------------------------------|--------------:|------------:|----------:|----------:| +| gemini-3.1-flash-lite (rated) | +0.56 | +0.57 | +0.14 | +0.19 | +| gemini-3.1-flash-lite-preview (rated) | +0.58 | +0.61 | +0.13 | +0.20 | +| grok-4.3 (rated) | +0.55 | +0.57 | +0.12 | +0.21 | +| qwen3.6-27b (rated) | +0.46 | +0.68 | +0.18 | +0.14 | +| qwen3.5-35b-a3b (rated) | +0.54 | +0.55 | +0.14 | +0.18 | +| qwen3.6-max-preview (rated) | +0.48 | +0.69 | +0.16 | +0.14 | +| muse-spark-1.3 (rated) | +0.43 | +0.73 | +0.18 | +0.12 | +| qwen3.8-27b (rated) | +0.43 | +0.69 | +0.16 | +0.14 | +| deepseek-v4-flash-0731 (rated) | +0.55 | +0.58 | +0.13 | +0.16 | +| gpt-oss-120b (rated) | +0.48 | +0.60 | +0.12 | +0.18 | +| qwen3.8-flash (rated) | +0.42 | +0.63 | +0.14 | +0.15 | +| gemma-4-31b-it (rated) | +0.36 | +0.66 | +0.16 | +0.13 | +| qwen3.5-122b-a10b (rated) | +0.48 | +0.62 | +0.15 | +0.14 | +| glm-4.5-air (rated) | +0.59 | +0.64 | +0.13 | +0.15 | +| gpt-5.4 (rated) | +0.62 | +0.64 | +0.12 | +0.17 | +| gemini-3.5-flash-lite (rated) | +0.57 | +0.65 | +0.13 | +0.15 | +| qwen3.5-plus-20260420 (rated) | +0.59 | +0.65 | +0.12 | +0.16 | +| gpt-5-mini (rated) | +0.52 | +0.64 | +0.09 | +0.19 | +| gpt-5.6-luna (rated) | +0.60 | +0.58 | +0.09 | +0.19 | +| gpt-5.6-terra (rated) | +0.55 | +0.64 | +0.10 | +0.17 | +| grok-4.6 (rated) | +0.42 | +0.72 | +0.15 | +0.13 | +| glm-5.1 (rated) | +0.54 | +0.69 | +0.12 | +0.15 | +| gpt-5.2 (rated) | +0.55 | +0.64 | +0.10 | +0.17 | +| gemini-2.5-pro (rated) | +0.47 | +0.70 | +0.13 | +0.14 | +| gpt-5.4-mini (rated) | +0.46 | +0.67 | +0.11 | +0.16 | +| grok-4.20 (rated) | +0.52 | +0.64 | +0.10 | +0.16 | +| gpt-4o-mini (rated) | +0.61 | +0.57 | +0.11 | +0.16 | +| deepseek-v4.1-flash (rated) | +0.54 | +0.60 | +0.10 | +0.16 | +| qwen3-coder-30b-a3b-instruct (rated) | +0.50 | +0.58 | +0.09 | +0.17 | +| glm-5.2 (rated) | +0.53 | +0.69 | +0.12 | +0.14 | +| deepseek-v3.2-exp (rated) | +0.58 | +0.60 | +0.09 | +0.17 | +| qwen3-30b-a3b (rated) | +0.67 | +0.61 | +0.07 | +0.19 | +| gpt-6-astra (rated) | +0.46 | +0.68 | +0.15 | +0.11 | +| mistral-large-2512 (rated) | +0.65 | +0.64 | +0.08 | +0.18 | +| gemma-3-27b-it (rated) | +0.60 | +0.60 | +0.09 | +0.17 | +| llama-4-maverick (rated) | +0.64 | +0.64 | +0.08 | +0.18 | +| qwen3-coder (rated) | +0.60 | +0.62 | +0.08 | +0.18 | +| qwen3.7-flash (rated) | +0.65 | +0.59 | +0.10 | +0.16 | +| qwen3-vl-235b-a22b-instruct (rated) | +0.60 | +0.55 | +0.05 | +0.21 | +| qwen-plus (rated) | +0.60 | +0.63 | +0.07 | +0.18 | +| gpt-5.3-chat (rated) | +0.50 | +0.67 | +0.09 | +0.16 | +| deepseek-v4-pro (rated) | +0.55 | +0.73 | +0.15 | +0.10 | +| llama-4-scout (rated) | +0.62 | +0.53 | +0.05 | +0.20 | +| grok-4.5 (rated) | +0.52 | +0.71 | +0.12 | +0.13 | +| gpt-4o-mini-2024-07-18 (rated) | +0.63 | +0.57 | +0.09 | +0.16 | +| qwen3-vl-30b-a3b-instruct (rated) | +0.53 | +0.53 | +0.06 | +0.19 | +| qwen3.5-397b-a17b (rated) | +0.54 | +0.65 | +0.08 | +0.17 | +| deepseek-chat-v3.1 (rated) | +0.62 | +0.60 | +0.07 | +0.18 | +| qwen3.5-plus-02-15 (rated) | +0.54 | +0.66 | +0.08 | +0.17 | +| glm-5.3 (rated) | +0.55 | +0.64 | +0.12 | +0.13 | +| deepseek-v3.2 (rated) | +0.57 | +0.56 | +0.08 | +0.17 | +| kimi-k2.6 (rated) | +0.61 | +0.67 | +0.09 | +0.16 | +| glm-5 (rated) | +0.59 | +0.67 | +0.08 | +0.16 | +| qwen3.7-plus (rated) | +0.51 | +0.67 | +0.10 | +0.15 | +| qwen3-30b-a3b-instruct-2507 (rated) | +0.65 | +0.54 | +0.05 | +0.20 | +| gpt-5.5 (rated) | +0.42 | +0.76 | +0.17 | +0.07 | +| qwen3.6-35b-a3b (rated) | +0.61 | +0.59 | +0.10 | +0.14 | +| qwen3.6-flash (rated) | +0.65 | +0.58 | +0.09 | +0.15 | +| gemini-3.8-flash (rated) | +0.38 | +0.64 | +0.16 | +0.08 | +| deepseek-chat-v3-0324 (rated) | +0.59 | +0.60 | +0.07 | +0.17 | +| qwen3.6-plus (rated) | +0.52 | +0.67 | +0.08 | +0.15 | +| qwen3-235b-a22b (rated) | +0.64 | +0.56 | +0.07 | +0.17 | +| gpt-5.4-nano (rated) | +0.48 | +0.60 | +0.10 | +0.13 | +| gemini-3.5-flash (rated) | +0.59 | +0.58 | +0.08 | +0.15 | +| qwen-2.5-72b-instruct (rated) | +0.60 | +0.60 | +0.06 | +0.17 | +| deepseek-v4-flash (rated) | +0.59 | +0.63 | +0.09 | +0.14 | +| qwen-plus-2025-07-28 (rated) | +0.64 | +0.51 | +0.07 | +0.16 | +| gemini-3-flash-preview (rated) | +0.56 | +0.63 | +0.10 | +0.13 | +| qwen3.7-max (rated) | +0.50 | +0.65 | +0.11 | +0.12 | +| qwen3-235b-a22b-2507 (rated) | +0.64 | +0.57 | +0.07 | +0.16 | +| qwen3.5-9b (rated) | +0.50 | +0.59 | +0.09 | +0.14 | +| muse-glimmer-30b (rated) | +0.49 | +0.65 | +0.10 | +0.13 | +| gemini-2.5-flash (rated) | +0.63 | +0.65 | +0.08 | +0.15 | +| qwen2.5-vl-72b-instruct (rated) | +0.58 | +0.60 | +0.07 | +0.15 | +| qwen3-vl-32b-instruct (rated) | +0.60 | +0.53 | +0.02 | +0.20 | +| gemini-2.5-flash-lite (rated) | +0.39 | +0.67 | +0.09 | +0.13 | +| gpt-5.6-sol (rated) | +0.53 | +0.67 | +0.09 | +0.13 | +| qwen3-max-thinking (rated) | +0.58 | +0.70 | +0.07 | +0.14 | +| qwen3-coder-plus (rated) | +0.62 | +0.60 | +0.07 | +0.14 | +| inkling (rated) | +0.56 | +0.69 | +0.10 | +0.11 | +| qwen3-vl-8b-instruct (rated) | +0.61 | +0.60 | +0.06 | +0.15 | +| qwen-2.5-7b-instruct (rated) | +0.55 | +0.59 | +0.05 | +0.16 | +| kimi-k3 (rated) | +0.62 | +0.67 | +0.07 | +0.14 | +| gpt-4o-2024-08-06 (rated) | +0.59 | +0.58 | +0.08 | +0.13 | +| qwen3.5-27b (rated) | +0.60 | +0.65 | +0.05 | +0.16 | +| qwen3.8-2.4t-a95b (rated) | +0.53 | +0.62 | +0.07 | +0.13 | +| qwen3-32b (rated) | +0.61 | +0.54 | +0.06 | +0.14 | +| qwen3-14b (rated) | +0.61 | +0.58 | +0.03 | +0.16 | +| glm-5.3-flash (rated) | +0.54 | +0.64 | +0.09 | +0.11 | +| glm-4.7-flash (rated) | +0.56 | +0.59 | +0.08 | +0.11 | +| gemini-3.6-flash (rated) | +0.59 | +0.62 | +0.07 | +0.12 | +| gpt-4.1 (rated) | +0.63 | +0.69 | +0.06 | +0.14 | +| gpt-4o (rated) | +0.61 | +0.58 | +0.07 | +0.12 | +| gpt-5.1 (rated) | +0.61 | +0.75 | +0.09 | +0.10 | +| qwen3-coder-next (rated) | +0.64 | +0.58 | +0.05 | +0.13 | +| gpt-4.1-mini (rated) | +0.57 | +0.45 | +0.10 | +0.08 | +| gpt-5 (rated) | +0.57 | +0.72 | +0.07 | +0.11 | +| qwen3-next-80b-a3b-instruct (rated) | +0.67 | +0.70 | +0.05 | +0.12 | +| gpt-4o-2024-11-20 (rated) | +0.62 | +0.60 | +0.06 | +0.10 | +| gpt-4.1-nano (rated) | +0.48 | +0.50 | +0.06 | +0.10 | +| gpt-3.5-turbo-0613 (rated) | +0.61 | +0.44 | +0.08 | +0.08 | +| claude-opus-4.7 (rated) | +0.59 | +0.60 | +0.07 | +0.09 | +| gpt-3.5-turbo-16k (rated) | +0.61 | +0.42 | +0.08 | +0.07 | +| qwen3-8b (rated) | +0.62 | +0.54 | +0.03 | +0.12 | +| claude-opus-4.8 (rated) | +0.61 | +0.58 | +0.07 | +0.07 | +| gemini-3.7-flash (rated) | +0.50 | +0.63 | +0.03 | +0.10 | +| claude-fable-5.1 (rated) | +0.58 | +0.61 | +0.07 | +0.05 | +| claude-opus-4.6 (rated) | +0.63 | +0.63 | +0.04 | +0.05 | diff --git a/docs/index.html b/docs/index.html index 489b613..5b965e2 100644 --- a/docs/index.html +++ b/docs/index.html @@ -4,7 +4,7 @@ Moral Maps: Where Do Frontier Models' Cultural Values Lie? - + diff --git a/docs/wvs/wvs_map_data.json b/docs/wvs/wvs_map_data.json index 0945181..f62e5e0 100644 --- a/docs/wvs/wvs_map_data.json +++ b/docs/wvs/wvs_map_data.json @@ -734,6 +734,8 @@ "label": null, "name": "gemini-3.1-flash-lite", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -754,6 +756,8 @@ "label": null, "name": "gemini-3.1-flash-lite-preview", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -774,6 +778,8 @@ "label": null, "name": "grok-4.3", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.372103799814643, "hle_source_effort": "high", "hle_source_name": "Grok 4.3 (high)", @@ -794,6 +800,8 @@ "label": null, "name": "qwen3.6-27b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.230769230769231, "hle_source_effort": null, "hle_source_name": "Qwen3.6 27B (Reasoning)", @@ -814,6 +822,8 @@ "label": null, "name": "qwen3.5-35b-a3b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.210379981464319, "hle_source_effort": null, "hle_source_name": "Qwen3.5 35B A3B (Reasoning)", @@ -834,6 +844,8 @@ "label": null, "name": "qwen3.6-max-preview", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.308155699721965, "hle_source_effort": null, "hle_source_name": "Qwen3.6 Max Preview", @@ -854,6 +866,8 @@ "label": "muse-spark-1.3", "name": "muse-spark-1.3", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.487025023169602, "hle_source_effort": "max", "hle_source_name": "Muse Spark 1.3 (max)", @@ -874,6 +888,8 @@ "label": null, "name": "qwen3.8-27b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.33920296570899, "hle_source_effort": "xhigh", "hle_source_name": "Qwen3.8 27B (xhigh)", @@ -894,6 +910,8 @@ "label": null, "name": "deepseek-v4-flash-0731", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -914,6 +932,8 @@ "label": null, "name": "gpt-oss-120b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -934,6 +954,8 @@ "label": "qwen3.8-flash", "name": "qwen3.8-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -954,6 +976,8 @@ "label": "gemma-4-31b-it", "name": "gemma-4-31b-it", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": 0.236329935125116, "hle_source_effort": null, "hle_source_name": "Gemma 4 31B (Reasoning)", @@ -974,6 +998,8 @@ "label": null, "name": "qwen3.5-122b-a10b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.252085264133457, "hle_source_effort": null, "hle_source_name": "Qwen3.5 122B A10B (Reasoning)", @@ -994,6 +1020,8 @@ "label": null, "name": "glm-4.5-air", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1014,6 +1042,8 @@ "label": null, "name": "gpt-5.4", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.437442075996293, "hle_source_effort": "xhigh", "hle_source_name": "GPT-5.4 (xhigh)", @@ -1034,6 +1064,8 @@ "label": null, "name": "gemini-3.5-flash-lite", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1054,6 +1086,8 @@ "label": null, "name": "qwen3.5-plus-20260420", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1074,6 +1108,8 @@ "label": null, "name": "gpt-5-mini", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1094,6 +1130,8 @@ "label": null, "name": "gpt-5.6-luna", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1114,6 +1152,8 @@ "label": null, "name": "gpt-5.6-terra", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1134,6 +1174,8 @@ "label": "grok-4.6", "name": "grok-4.6", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1154,6 +1196,8 @@ "label": null, "name": "glm-5.1", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1174,6 +1218,8 @@ "label": null, "name": "gpt-5.2", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1194,6 +1240,8 @@ "label": null, "name": "gemini-2.5-pro", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": 0.225208526413346, "hle_source_effort": null, "hle_source_name": "Gemini 2.5 Pro", @@ -1214,6 +1262,8 @@ "label": null, "name": "gpt-5.4-mini", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1234,6 +1284,8 @@ "label": null, "name": "grok-4.20", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.345227062094532, "hle_source_effort": null, "hle_source_name": "Grok 4.20 0309 v2 (Reasoning)", @@ -1248,12 +1300,36 @@ "x": -0.5189766385048643, "y": 0.6372836916088428 }, + { + "color": "#2563eb", + "family": "gpt", + "label": null, + "name": "gpt-4o-mini", + "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", + "hle_score": null, + "hle_source_effort": null, + "hle_source_name": null, + "items": 12, + "protocol_id": "252fea691e8f00a7b3a658876047b1d93af94b89f127d549bac93f95644b933c", + "readout": "rated categorical response", + "release_created": "2024-07-18", + "release_source": "catalog", + "run_id": "20260917T135119Z_252fea691e8f", + "samples": 12 + }, + "x": -0.6137290564373897, + "y": 0.5741318151453763 + }, { "color": "#ec4899", "family": "deepseek", "label": "deepseek-v4.1-flash", "name": "deepseek-v4.1-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.392493049119555, "hle_source_effort": "max", "hle_source_name": "DeepSeek V4.1 Flash (Reasoning, Max Effort)", @@ -1274,6 +1350,8 @@ "label": null, "name": "qwen3-coder-30b-a3b-instruct", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.0384615384615385, "hle_source_effort": null, "hle_source_name": "Qwen3 Coder 30B A3B Instruct", @@ -1294,6 +1372,8 @@ "label": null, "name": "glm-5.2", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1314,6 +1394,8 @@ "label": null, "name": "deepseek-v3.2-exp", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1334,6 +1416,8 @@ "label": null, "name": "qwen3-30b-a3b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1354,6 +1438,8 @@ "label": "gpt-6-astra", "name": "gpt-6-astra", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.546802594995366, "hle_source_effort": "max", "hle_source_name": "GPT-6 Astra (max)", @@ -1374,6 +1460,8 @@ "label": "mistral-large-2512", "name": "mistral-large-2512", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1394,6 +1482,8 @@ "label": null, "name": "gemma-3-27b-it", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": 0.044, "hle_source_effort": null, "hle_source_name": "Gemma 3 27B Instruct", @@ -1414,6 +1504,8 @@ "label": "llama-4-maverick", "name": "llama-4-maverick", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": 0.0491195551436515, "hle_source_effort": null, "hle_source_name": "Llama 4 Maverick", @@ -1434,6 +1526,8 @@ "label": null, "name": "qwen3-coder", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1454,6 +1548,8 @@ "label": null, "name": "qwen3.7-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1474,6 +1570,8 @@ "label": null, "name": "qwen3-vl-235b-a22b-instruct", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.0662650602409639, "hle_source_effort": null, "hle_source_name": "Qwen3 VL 235B A22B Instruct", @@ -1494,6 +1592,8 @@ "label": null, "name": "qwen-plus", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1514,6 +1614,8 @@ "label": null, "name": "gpt-5.3-chat", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1534,6 +1636,8 @@ "label": null, "name": "deepseek-v4-pro", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": 0.410101946246525, "hle_source_effort": "max", "hle_source_name": "DeepSeek V4 Pro 0813 (Reasoning, Max Effort)", @@ -1554,6 +1658,8 @@ "label": null, "name": "llama-4-scout", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": 0.0378, "hle_source_effort": null, "hle_source_name": "Llama 4 Scout", @@ -1574,6 +1680,8 @@ "label": null, "name": "grok-4.5", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1588,12 +1696,36 @@ "x": -0.524187458188895, "y": 0.708147541423977 }, + { + "color": "#2563eb", + "family": "gpt", + "label": null, + "name": "gpt-4o-mini-2024-07-18", + "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", + "hle_score": null, + "hle_source_effort": null, + "hle_source_name": null, + "items": 12, + "protocol_id": "3f88d93861e3e20746627f7ff664bd93fd39a89d112687ce4c83c7dba73c896d", + "readout": "rated categorical response", + "release_created": "2024-07-18", + "release_source": "catalog", + "run_id": "20260917T135444Z_3f88d93861e3", + "samples": 12 + }, + "x": -0.6260950984945237, + "y": 0.5722794251348428 + }, { "color": "#7e22ce", "family": "qwen", "label": null, "name": "qwen3-vl-30b-a3b-instruct", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.0630213160333642, "hle_source_effort": null, "hle_source_name": "Qwen3 VL 30B A3B Instruct", @@ -1614,6 +1746,8 @@ "label": null, "name": "qwen3.5-397b-a17b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.289620018535681, "hle_source_effort": null, "hle_source_name": "Qwen3.5 397B A17B (Reasoning)", @@ -1634,6 +1768,8 @@ "label": null, "name": "deepseek-chat-v3.1", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1654,6 +1790,8 @@ "label": null, "name": "qwen3.5-plus-02-15", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1674,6 +1812,8 @@ "label": null, "name": "glm-5.3", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.422613531047266, "hle_source_effort": "max", "hle_source_name": "GLM-5.3 (max)", @@ -1694,6 +1834,8 @@ "label": null, "name": "deepseek-v3.2", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1714,6 +1856,8 @@ "label": null, "name": "kimi-k2.6", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1734,6 +1878,8 @@ "label": null, "name": "glm-5", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1754,6 +1900,8 @@ "label": null, "name": "qwen3.7-plus", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.356348470806302, "hle_source_effort": null, "hle_source_name": "Qwen3.7 Plus", @@ -1774,6 +1922,8 @@ "label": null, "name": "qwen3-30b-a3b-instruct-2507", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1794,6 +1944,8 @@ "label": null, "name": "gpt-5.5", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": 0.457831325301205, "hle_source_effort": "xhigh", "hle_source_name": "GPT-5.5 (xhigh)", @@ -1814,6 +1966,8 @@ "label": null, "name": "qwen3.6-35b-a3b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.222428174235403, "hle_source_effort": null, "hle_source_name": "Qwen3.6 35B A3B (Reasoning)", @@ -1834,6 +1988,8 @@ "label": null, "name": "qwen3.6-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1854,6 +2010,8 @@ "label": "gemini-3.8-flash", "name": "gemini-3.8-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1874,6 +2032,8 @@ "label": null, "name": "deepseek-chat-v3-0324", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1894,6 +2054,8 @@ "label": null, "name": "qwen3.6-plus", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.278498609823911, "hle_source_effort": null, "hle_source_name": "Qwen3.6 Plus", @@ -1914,6 +2076,8 @@ "label": null, "name": "qwen3-235b-a22b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.1098, "hle_source_effort": null, "hle_source_name": "Qwen3 235B A22B (Reasoning)", @@ -1934,6 +2098,8 @@ "label": null, "name": "gpt-5.4-nano", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1954,6 +2120,8 @@ "label": null, "name": "gemini-3.5-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1974,6 +2142,8 @@ "label": null, "name": "qwen-2.5-72b-instruct", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -1994,6 +2164,8 @@ "label": null, "name": "deepseek-v4-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.385542168674699, "hle_source_effort": "max", "hle_source_name": "DeepSeek V4 Flash 0731 (Reasoning, Max Effort)", @@ -2014,6 +2186,8 @@ "label": null, "name": "qwen-plus-2025-07-28", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2034,6 +2208,8 @@ "label": null, "name": "gemini-3-flash-preview", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2054,6 +2230,8 @@ "label": null, "name": "qwen3.7-max", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.405004633920297, "hle_source_effort": null, "hle_source_name": "Qwen3.7 Max", @@ -2074,6 +2252,8 @@ "label": null, "name": "qwen3-235b-a22b-2507", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2094,6 +2274,8 @@ "label": null, "name": "qwen3.5-9b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.149212233549583, "hle_source_effort": null, "hle_source_name": "Qwen3.5 9B (Reasoning)", @@ -2114,6 +2296,8 @@ "label": null, "name": "muse-glimmer-30b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2134,6 +2318,8 @@ "label": null, "name": "gemini-2.5-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2154,6 +2340,8 @@ "label": null, "name": "qwen2.5-vl-72b-instruct", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2174,6 +2362,8 @@ "label": null, "name": "qwen3-vl-32b-instruct", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.0681186283595922, "hle_source_effort": null, "hle_source_name": "Qwen3 VL 32B Instruct", @@ -2194,6 +2384,8 @@ "label": null, "name": "gemini-2.5-flash-lite", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2214,6 +2406,8 @@ "label": null, "name": "gpt-5.6-sol", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.494902687673772, "hle_source_effort": "max", "hle_source_name": "GPT-5.6 Sol (max)", @@ -2234,6 +2428,8 @@ "label": null, "name": "qwen3-max-thinking", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.279888785912882, "hle_source_effort": null, "hle_source_name": "Qwen3 Max Thinking", @@ -2254,6 +2450,8 @@ "label": null, "name": "qwen3-coder-plus", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2274,6 +2472,8 @@ "label": "inkling", "name": "inkling", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.318813716404078, "hle_source_effort": "xhigh", "hle_source_name": "Inkling (xhigh)", @@ -2294,6 +2494,8 @@ "label": null, "name": "qwen3-vl-8b-instruct", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.0268767377201112, "hle_source_effort": null, "hle_source_name": "Qwen3 VL 8B Instruct", @@ -2314,6 +2516,8 @@ "label": null, "name": "qwen-2.5-7b-instruct", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2334,6 +2538,8 @@ "label": "kimi-k3", "name": "kimi-k3", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.468952734012975, "hle_source_effort": "max", "hle_source_name": "Kimi K3 (max)", @@ -2348,12 +2554,36 @@ "x": -0.6214472858835399, "y": 0.6710244268262847 }, + { + "color": "#2563eb", + "family": "gpt", + "label": null, + "name": "gpt-4o-2024-08-06", + "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", + "hle_score": null, + "hle_source_effort": null, + "hle_source_name": null, + "items": 12, + "protocol_id": "1a12b248142ab869265f8502e074a7ef4d5e096a8a0c2dee2eb0e632473f1bcb", + "readout": "rated categorical response", + "release_created": "2024-08-06", + "release_source": "catalog", + "run_id": "20260917T134801Z_1a12b248142a", + "samples": 12 + }, + "x": -0.5912804058064387, + "y": 0.5802580057057466 + }, { "color": "#7e22ce", "family": "qwen", "label": null, "name": "qwen3.5-27b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.239110287303058, "hle_source_effort": null, "hle_source_name": "Qwen3.5 27B (Reasoning)", @@ -2374,6 +2604,8 @@ "label": null, "name": "qwen3.8-2.4t-a95b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2394,6 +2626,8 @@ "label": null, "name": "qwen3-32b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.0741, "hle_source_effort": null, "hle_source_name": "Qwen3 32B (Reasoning)", @@ -2414,6 +2648,8 @@ "label": null, "name": "qwen3-14b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2434,6 +2670,8 @@ "label": "glm-5.3-flash", "name": "glm-5.3-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2454,6 +2692,8 @@ "label": null, "name": "glm-4.7-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2474,6 +2714,8 @@ "label": null, "name": "gemini-3.6-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2488,12 +2730,58 @@ "x": -0.593257275132275, "y": 0.6186534585344109 }, + { + "color": "#2563eb", + "family": "gpt", + "label": null, + "name": "gpt-4.1", + "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", + "hle_score": null, + "hle_source_effort": null, + "hle_source_name": null, + "items": 12, + "protocol_id": "0a72ded0be88127e9009e1c659d63042c0d8d7622ec501f7f6a3a836a4a41527", + "readout": "rated categorical response", + "release_created": "2025-04-14", + "release_source": "catalog", + "run_id": "20260917T133258Z_0a72ded0be88", + "samples": 12 + }, + "x": -0.6299053865373309, + "y": 0.6879566393703299 + }, + { + "color": "#2563eb", + "family": "gpt", + "label": null, + "name": "gpt-4o", + "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", + "hle_score": null, + "hle_source_effort": null, + "hle_source_name": null, + "items": 12, + "protocol_id": "9933b4a3be52fec8f416cf1d8bd3cefa1718efa1f0177f307292404d3e9db105", + "readout": "rated categorical response", + "release_created": "2024-05-13", + "release_source": "catalog", + "run_id": "20260917T135801Z_9933b4a3be52", + "samples": 12 + }, + "x": -0.6060268571393299, + "y": 0.5841620141118888 + }, { "color": "#2563eb", "family": "gpt", "label": null, "name": "gpt-5.1", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2514,6 +2802,8 @@ "label": null, "name": "qwen3-coder-next", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.101482854494903, "hle_source_effort": null, "hle_source_name": "Qwen3 Coder Next", @@ -2528,12 +2818,36 @@ "x": -0.6363133579397021, "y": 0.5792251640447779 }, + { + "color": "#2563eb", + "family": "gpt", + "label": null, + "name": "gpt-4.1-mini", + "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", + "hle_score": null, + "hle_source_effort": null, + "hle_source_name": null, + "items": 12, + "protocol_id": "612c6281a597db4f761707c0c0876d694d715daf35b4969395eeb63fdf9f5fa0", + "readout": "rated categorical response", + "release_created": "2025-04-14", + "release_source": "catalog", + "run_id": "20260917T133650Z_612c6281a597", + "samples": 12 + }, + "x": -0.5663403168156219, + "y": 0.45394876566876174 + }, { "color": "#2563eb", "family": "gpt", "label": null, "name": "gpt-5", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": null, "hle_source_effort": null, "hle_source_name": null, @@ -2554,6 +2868,8 @@ "label": null, "name": "qwen3-next-80b-a3b-instruct", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.0764596848934198, "hle_source_effort": null, "hle_source_name": "Qwen3 Next 80B A3B Instruct", @@ -2568,12 +2884,80 @@ "x": -0.6652380952380952, "y": 0.7022990677752582 }, + { + "color": "#2563eb", + "family": "gpt", + "label": null, + "name": "gpt-4o-2024-11-20", + "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", + "hle_score": null, + "hle_source_effort": null, + "hle_source_name": null, + "items": 12, + "protocol_id": "eb1402c60a7305f5c9b800e5682ce648f3acc25970eaf798d0c3485b4bb16d48", + "readout": "rated categorical response", + "release_created": "2024-11-20", + "release_source": "catalog", + "run_id": "20260917T134443Z_eb1402c60a73", + "samples": 12 + }, + "x": -0.6174577809885475, + "y": 0.6028457218832352 + }, + { + "color": "#2563eb", + "family": "gpt", + "label": null, + "name": "gpt-4.1-nano", + "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", + "hle_score": null, + "hle_source_effort": null, + "hle_source_name": null, + "items": 12, + "protocol_id": "b6966daf09695284896ac4f0dda90cadb5e6fce5a1a77b576b290089050d0c9e", + "readout": "rated categorical response", + "release_created": "2025-04-14", + "release_source": "catalog", + "run_id": "20260917T134036Z_b6966daf0969", + "samples": 12 + }, + "x": -0.48177082177082176, + "y": 0.4976680628265054 + }, + { + "color": "#2563eb", + "family": "gpt", + "label": null, + "name": "gpt-3.5-turbo-0613", + "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", + "hle_score": null, + "hle_source_effort": null, + "hle_source_name": null, + "items": 12, + "protocol_id": "b25b19e78ceb3fe7814ac5883239382f1c6e757fe89329495241a25ab521e178", + "readout": "rated categorical response", + "release_created": "2024-01-25", + "release_source": "catalog", + "run_id": "20260917T140156Z_b25b19e78ceb", + "samples": 12 + }, + "x": -0.6091300100287029, + "y": 0.438086083059628 + }, { "color": "#c026d3", "family": "claude", "label": null, "name": "claude-opus-4.7", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": 0.423076923076923, "hle_source_effort": "max", "hle_source_name": "Claude Opus 4.7 (Adaptive Reasoning, Max Effort)", @@ -2588,12 +2972,36 @@ "x": -0.59, "y": 0.6 }, + { + "color": "#2563eb", + "family": "gpt", + "label": null, + "name": "gpt-3.5-turbo-16k", + "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", + "hle_score": null, + "hle_source_effort": null, + "hle_source_name": null, + "items": 12, + "protocol_id": "340478ea86f503c256128623875d94e71ae4a6f5af06fda18042a3187f03f9ea", + "readout": "rated categorical response", + "release_created": "2023-08-28", + "release_source": "catalog", + "run_id": "20260917T140450Z_340478ea86f5", + "samples": 12 + }, + "x": -0.6100462990851458, + "y": 0.4246338825577172 + }, { "color": "#7e22ce", "family": "qwen", "label": null, "name": "qwen3-8b", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.0389, "hle_source_effort": null, "hle_source_name": "Qwen3 8B (Reasoning)", @@ -2614,6 +3022,8 @@ "label": null, "name": "claude-opus-4.8", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": 0.486561631139944, "hle_source_effort": "max", "hle_source_name": "Claude Opus 4.8 (Adaptive Reasoning, Max Effort)", @@ -2634,6 +3044,8 @@ "label": null, "name": "gemini-3.7-flash", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.478683966635774, "hle_source_effort": "high", "hle_source_name": "Gemini 3.7 Flash (high)", @@ -2654,6 +3066,8 @@ "label": "fable-5.1", "name": "claude-fable-5.1", "provenance": { + "coordinate_provenance": "canonical score-all-options", + "eval_version": "wvs-score-all-options-v1", "hle_score": 0.591288229842447, "hle_source_effort": "max", "hle_source_name": "Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)", @@ -2674,6 +3088,8 @@ "label": null, "name": "claude-opus-4.6", "provenance": { + "coordinate_provenance": "historical rounded coordinate", + "eval_version": null, "hle_score": 0.399443929564411, "hle_source_effort": "max", "hle_source_name": "Claude Opus 4.6 (Adaptive Reasoning, Max Effort)", diff --git a/scripts/wvs_map.py b/scripts/wvs_map.py index 6a32d55..85908bd 100644 --- a/scripts/wvs_map.py +++ b/scripts/wvs_map.py @@ -553,15 +553,15 @@ def main() -> None: saved_catalog[panel["model"]]["created"], tz=timezone.utc).date().isoformat()} if panel is None: legacy_date = legacy_release_dates.get(name) - provenance = {"readout": "recovered rounded historical coordinate", "items": None, - "samples": None, "run_id": None, "protocol_id": None, + provenance = {"coordinate_provenance": "historical rounded coordinate", "readout": "recovered rounded historical coordinate", + "eval_version": None, "items": None, "samples": None, "run_id": None, "protocol_id": None, "release_created": legacy_date, "release_source": (f"saved OpenRouter catalog 2026-09-17: {LEGACY_CATALOG_IDS[name]}" if legacy_date else "historical coordinate")} else: - provenance = {"readout": "rated categorical response", "items": panel["n_items"], - "samples": panel["n_samples"], "run_id": panel["run_id"], - "protocol_id": panel["protocol_id"], + provenance = {"coordinate_provenance": "canonical score-all-options", "readout": "rated categorical response", + "eval_version": panel["eval_version"], "items": panel["n_items"], "samples": panel["n_samples"], + "run_id": panel["run_id"], "protocol_id": panel["protocol_id"], "release_created": catalog["created"] if catalog else None, "release_source": "catalog" if catalog else "request ledger"} capability = capability_by_model.get(name) @@ -612,8 +612,9 @@ def main() -> None: "WVS Inglehart-Welzel", countries, P, ("Survival", "Self-expression", "Traditional", "Secular-Rational"), models=plot_models, model_labels=model_labels, emphasize=emph, - title="Frontier LLMs on the\nWorld Values Survey", - note=f"{len(plot_models)} models, rated sampling\ngithub.com/wassname/moral-maps") + title="Moral Maps: Where Do Frontier\nModels' Cultural Values Lie?", + note="World Values Survey | source: github.com/wassname/moral-maps", + title_y=0.115, note_y=0.04) fig.savefig(args.out, dpi=200, bbox_inches="tight") fig.savefig(Path(args.out).with_suffix(".svg"), bbox_inches="tight") logger.info(f"wrote {args.out}") diff --git a/scripts/wvs_react/src/main.jsx b/scripts/wvs_react/src/main.jsx index c1d5ab6..b0a4aaa 100644 --- a/scripts/wvs_react/src/main.jsx +++ b/scripts/wvs_react/src/main.jsx @@ -88,7 +88,10 @@ function ReleaseScatter({ data, hidden, field, title, axisMode }) { }, [axisMode, visible]); const frontierPlacement = useMemo(() => { const bounds = { left, right: width - right, top, bottom: height - bottom }; - const taken = plotted.map(model => box(plotX(xValue(model)), valueY(coordinate(model)), 16, 16)); + const taken = [ + ...plotted.map(model => box(plotX(xValue(model)), valueY(coordinate(model)), 16, 16)), + ...(fit ? [box(left + 66, top + 9, 128, 18)] : []), + ]; const placement = {}; for (const model of frontier) { const anchor = { x: plotX(xValue(model)), y: valueY(coordinate(model)) }; @@ -213,7 +216,7 @@ function Map({ data }) { {data.zone_hulls.map(zone => {zone.name})} {Object.entries(groups).map(([family, models]) => {models.map(model => )})} {data.axis.y[1]}{data.axis.y[0]}{data.axis.x[0]}{data.axis.x[1]} - {data.title.split('\n').map((line, index) => {line})} + {data.title.split('\n').map((line, index) => {line})} {data.note.split('\n').map((line, index) => {line})} diff --git a/scripts/wvs_react/uat.py b/scripts/wvs_react/uat.py index 79d654a..01a7708 100644 --- a/scripts/wvs_react/uat.py +++ b/scripts/wvs_react/uat.py @@ -57,15 +57,27 @@ def main() -> None: assert data["capability_x"]["fetched_utc"] == "2026-09-17T10:13:20Z" capability_matched = [model for model in data["models"] if model["provenance"]["hle_score"] is not None] assert len(capability_matched) == data["capability_x"]["matched_models"] == 46 + canonical_models = [model for model in data["models"] + if model["provenance"]["coordinate_provenance"] == "canonical score-all-options"] + historical_models = [model for model in data["models"] + if model["provenance"]["coordinate_provenance"] == "historical rounded coordinate"] + assert len(canonical_models) == 96 + assert len(historical_models) == 12 + assert all(model["provenance"]["eval_version"] == "wvs-score-all-options-v1" + for model in canonical_models) + assert all(model["provenance"]["eval_version"] is None for model in historical_models) + assert all("reliability" not in str(model["provenance"]["protocol_id"]) + for model in canonical_models) assert len(data["countries"]) == 90 assert len(data["zone_hulls"]) == 4 assert len(data["latest_by_family"]) == 13 assert all(len(zone["points"]) >= 30 for zone in data["zone_hulls"]) - dated = [model for model in data["models"] if model["provenance"]["release_created"]] - grok_dates = {model["name"]: model["provenance"]["release_created"] for model in data["models"] if model["name"] in {"grok-4.20", "grok-4.3"}} - assert grok_dates == {"grok-4.20": "2026-03-31", "grok-4.3": "2026-04-30"} - assert all(model["provenance"]["release_source"].startswith("saved OpenRouter catalog 2026-09-17: x-ai/") - for model in data["models"] if model["name"] in grok_dates) + dated = sorted((model for model in data["models"] if model["provenance"]["release_created"]), + key=lambda model: (model["provenance"]["release_created"], model["name"])) + grok_dates = {model["name"]: model["provenance"]["release_created"] for model in data["models"] if model["family"] == "grok"} + assert grok_dates == {"grok-4.20": "2026-03-31", "grok-4.3": "2026-04-30", + "grok-4.5": "2026-07-08", "grok-4.6": "2026-08-12"} + assert data["latest_by_family"]["grok"]["name"] == "grok-4.6" families = {model["family"] for model in data["models"]} with sync_playwright() as playwright: @@ -130,9 +142,20 @@ def main() -> None: svg_description(page, '.release-panel[data-coordinate="x"] svg', "release-x-svg-title", "release-x-svg-desc") assert page.locator(".release-mark").count() == len(dated) * 2 assert page.locator(".frontier-label").count() > 0 - frontier_boxes = page.locator('.frontier-label text').evaluate_all("nodes => nodes.map(node => { const b = node.getBBox(); return [b.x, b.y, b.width, b.height]; })") - assert all(x >= 96 and y >= 42 and x + width <= 1165 and y + height <= 272 for x, y, width, height in frontier_boxes) - assert all(a[0] + a[2] <= b[0] or b[0] + b[2] <= a[0] or a[1] + a[3] <= b[1] or b[1] + b[3] <= a[1] for index, a in enumerate(frontier_boxes) for b in frontier_boxes[index + 1:]) + expected_frontier = [] + best_score = float("-inf") + for model in dated: + score = model["provenance"]["hle_score"] + if score is not None and score > best_score: + expected_frontier.append(model["name"]) + best_score = score + for panel in page.locator(".release-panel").all(): + assert panel.locator(".frontier-label line").count() == 0 + actual_frontier = panel.locator(".frontier-label").evaluate_all("nodes => nodes.map(node => node.dataset.frontierModel)") + assert actual_frontier == expected_frontier, (actual_frontier, expected_frontier) + frontier_boxes = panel.locator('.frontier-label text').evaluate_all("nodes => nodes.map(node => { const b = node.getBBox(); return [b.x, b.y, b.width, b.height]; })") + assert all(x >= 96 and y >= 42 and x + width <= 1165 and y + height <= 272 for x, y, width, height in frontier_boxes) + assert all(a[0] + a[2] <= b[0] or b[0] + b[2] <= a[0] or a[1] + a[3] <= b[1] or b[1] + b[3] <= a[1] for index, a in enumerate(frontier_boxes) for b in frontier_boxes[index + 1:]) for grok_name in grok_dates: assert page.locator(f'[data-release-model="{grok_name}"]').count() == 2 self_expression_panel = page.locator('.release-panel[data-coordinate="x"]') @@ -198,7 +221,7 @@ def main() -> None: page.screenshot(path=OUT / "wvs_react_root_playwright_keyboard_focus.png", full_page=True) release_y_marker = page.locator('.release-panel[data-coordinate="y"] [data-release-model="qwen3.8-flash"]') - release_y_marker.locator(".model-ring").hover() + release_y_marker.focus() page.wait_for_selector("#release-y-tooltip") assert page.locator("#release-y-tooltip strong").inner_text() == "qwen3.8-flash" assert page.locator("#release-y-tooltip span").all_text_contents() == ["Secular-Rational: 0.635", "release 2026-08-26"] @@ -208,7 +231,7 @@ def main() -> None: assert page.locator("#release-y-tooltip").is_visible() page.screenshot(path=OUT / "wvs_react_root_playwright_release_panel_keyboard_focus.png", full_page=True) release_x_marker = page.locator('.release-panel[data-coordinate="x"] [data-release-model="qwen3.8-flash"]') - release_x_marker.locator(".model-ring").hover() + release_x_marker.focus() page.wait_for_selector("#release-x-tooltip") assert release_y_marker.get_attribute("aria-describedby") == "release-y-tooltip" assert release_x_marker.get_attribute("aria-describedby") == "release-x-tooltip" @@ -218,7 +241,7 @@ def main() -> None: for grok_name, grok_date in grok_dates.items(): grok = next(model for model in dated if model["name"] == grok_name) grok_marker = page.locator(f'.release-panel[data-coordinate="y"] [data-release-model="{grok_name}"]') - grok_marker.locator(".model-ring").hover() + grok_marker.focus() assert page.locator("#release-y-tooltip strong").inner_text() == grok_name assert page.locator("#release-y-tooltip span").all_text_contents() == [ f"Secular-Rational: {grok['y']:.3f}", f"release {grok_date}", @@ -238,10 +261,10 @@ def main() -> None: page.screenshot(path=OUT / "wvs_react_playwright_release_fit_qwen_only.png", full_page=True) page.get_by_role("button", name="qwen", exact=True).click() - page.get_by_role("button", name="muse", exact=True).click() + page.get_by_role("button", name="inkling", exact=True).click() page.wait_for_timeout(100) for family in families: - expected = "inline" if family == "muse" else "none" + expected = "inline" if family == "inkling" else "none" assert map_svg.locator(f':scope > g[data-family="{family}"]').get_attribute("display") == expected for panel in page.locator(".release-panel").all(): assert int(panel.locator("svg").get_attribute("data-fit-n")) == 0 @@ -268,12 +291,13 @@ def main() -> None: print(json.dumps({ "root": "React app loaded at /", "compatibility": "/wvs/ and /wvs/react/ redirect to the root React app; shared data remains at /wvs/wvs_map_data.json", - "shared": {"models": 64, "countries": 90, "buffered_hulls": 4, "latest_labels": 13, + "shared": {"models": len(data["models"]), "canonical_score_all_options": len(canonical_models), + "historical_rounded": len(historical_models), "countries": 90, "buffered_hulls": 4, "latest_labels": 13, "numeric_model_country_median_equality": "verified against DOM data attributes"}, "svg_accessibility": "main map and both release panels have stable title/desc aria-labelledby; descriptions update after family visibility change", "qwen_toggle": "clicked, map and dated-panel family groups hidden, country coordinates invariant", "tooltip": "map and release-panel pointer hover and focus show panel-specific model name, coordinate, and release date only", - "release_panels": {"panels": 2, "dated_models": len(dated), "date_order": "DOM order verified; both catalog-dated Grok models rendered in each panel", + "release_panels": {"panels": 2, "dated_models": len(dated), "date_order": "DOM order verified; all four catalog-dated Grok models rendered in each panel", "ols": "line, n, and R squared recompute from currently visible matched models; the Self-expression panel renders negative stored x, so upward means Self-expression; fewer than two distinct x values hide the fit"}, "capability_panels": {"matched_models": len(capability_matched), "omitted_models": len(data["models"]) - len(capability_matched), "source": data["capability_x"]["source_url"], "fetched_utc": data["capability_x"]["fetched_utc"], diff --git a/scripts/wvs_score_all_options_recover_cache.py b/scripts/wvs_score_all_options_recover_cache.py index 5da7e35..072cf85 100644 --- a/scripts/wvs_score_all_options_recover_cache.py +++ b/scripts/wvs_score_all_options_recover_cache.py @@ -13,7 +13,7 @@ from pathlib import Path import numpy as np from moralmaps.iw_axes import X_AXIS, Y_AXIS, resolve_items -from moralmaps.rated_cache import merge_completed, update_coords +from moralmaps.rated_cache import merge_completed, update_coords, update_eval_versions from wvs_map import load_wvs_all, model_coord_ci CACHE = Path("slop/research/wvs/20260916_openrouter/wvs_iw_rated.json") @@ -101,6 +101,8 @@ def main() -> None: parser.add_argument("--smoke", action="store_true") parser.add_argument("--refresh-ci", action="store_true", help="recompute only existing complete panels' coordinate CI summaries from the ledger") + parser.add_argument("--stamp-eval-version", action="store_true", + help="add score-all-options v1 provenance to cache entries recovered from the ledger") args = parser.parse_args() if args.smoke: concurrency_smoke() @@ -120,6 +122,8 @@ def main() -> None: merged = update_coords(CACHE, {key: entry["coords"] for key, entry in overlaps.items()}) else: merged = merge_completed(CACHE, additions) + if args.stamp_eval_version: + merged = update_eval_versions(CACHE, {key: "wvs-score-all-options-v1" for key in entries}) preserved = {key: merged["completed"][key] for key in existing} preserved_hash = hashlib.sha256(json.dumps({key: stable_entry(value) for key, value in preserved.items()}, sort_keys=True, separators=(",", ":")).encode()).hexdigest() @@ -131,6 +135,7 @@ def main() -> None: "existing_entries_sha256_after": preserved_hash, "new_complete_runs_added": len(additions), "ci_summaries_refreshed": len(overlaps) if args.refresh_ci else 0, + "eval_versions_stamped": len(entries) if args.stamp_eval_version else 0, "new_protocol_ids": sorted(additions), "new_models": sorted(entry["model"] for entry in additions.values()), "overlap_complete_runs": len(overlaps), diff --git a/slop/audits/20260917_wvs_score_all_options_cache_recovery.json b/slop/audits/20260917_wvs_score_all_options_cache_recovery.json index dbb4cb1..ba77363 100644 --- a/slop/audits/20260917_wvs_score_all_options_cache_recovery.json +++ b/slop/audits/20260917_wvs_score_all_options_cache_recovery.json @@ -1,14 +1,15 @@ { - "cache_completed_entries_after_merge": 86, - "ci_summaries_refreshed": 86, - "existing_entries_preserved_count": 86, - "existing_entries_sha256_after": "2fa381c2744633f337988c3fc2a753227849a733aaa906eceb507fe60b5b187f", - "existing_entries_sha256_before": "2fa381c2744633f337988c3fc2a753227849a733aaa906eceb507fe60b5b187f", + "cache_completed_entries_after_merge": 97, + "ci_summaries_refreshed": 0, + "eval_versions_stamped": 97, + "existing_entries_preserved_count": 97, + "existing_entries_sha256_after": "b4b6afcfa52ca47008d26ab7ac923fd741c0626690e02298cafb7e55f3c6dcd3", + "existing_entries_sha256_before": "47de431087eeea68bbf2591a8152fcb5c1fa6a084cace7fa5a6183ee635c7f3f", "ledger": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", - "ledger_valid_lines": 49466, + "ledger_valid_lines": 54700, "new_complete_runs_added": 0, "new_models": [], "new_protocol_ids": [], - "overlap_complete_runs": 86, + "overlap_complete_runs": 97, "overlap_point_coordinates_equal": true } diff --git a/slop/audits/20260918_wvs_final_canonical_integration.md b/slop/audits/20260918_wvs_final_canonical_integration.md new file mode 100644 index 0000000..58cdc53 --- /dev/null +++ b/slop/audits/20260918_wvs_final_canonical_integration.md @@ -0,0 +1,22 @@ +# Final canonical WVS integration + +-- PI[k3] + +## Observations + +- `wvs_iw_rated.json` has 97 complete cache records. `gpt-5-nano` remains excluded from the map because its prior content-quality audit documented systematic neutral/example-like outputs. The public artifact has 108 distinct names: 96 canonical `wvs-score-all-options-v1` entries and 12 distinct historical rounded coordinates. Four historical names overlap canonical names, so the canonical coordinate is shown for each overlap. +- The public artifact contains 13 families. The Grok entries are `grok-4.20`, `grok-4.3`, `grok-4.5`, and `grok-4.6`; the catalog-date latest label is `grok-4.6` (2026-08-12). +- The saved Artificial Analysis HLE mapping supplies 46 matches and 62 omissions among the 108 displayed names. The artifact retains the source URL and fetch time, `2026-09-17T10:13:20Z`. +- No displayed provenance protocol ID belongs to the DeepSeek reliability pilot. Reliability aggregate and incomplete replicate records were not used to generate the map. + +## Verification + +`uv run --no-project --offline --with 'playwright==1.62.0' python scripts/wvs_react/uat.py` passed. It checks the root and both redirects, 13 family controls, map/country geometry, all four Groks, `grok-4.6` as latest, canonical versus historical provenance, absent pilot IDs, release-date frontier labels without leader lines, capability fits under all-family and family-subset visibility, and accessible SVG titles/descriptions. + +I inspected `slop/research/wvs/20260916_openrouter/wvs_react_root_playwright_default.png`. A fresh-eyes review also found the lower-left title/source readable and the revised OLS/frontier labels separated. Its evidence is a read-only reviewer response in this worker session, not a committed file. + +## Exclusions + +- `gpt-5-nano` is a completed cache diagnostic but is not plotted because its existing content-quality audit excluded it. +- Incomplete/partial score-all-options panels and all DeepSeek reliability-pilot data are excluded. +- The 16 historical rounded inputs remain distinct provenance; four duplicate canonical names resolve to the canonical panel. diff --git a/slop/research/wvs/20260916_openrouter/wvs_iw_rated.json b/slop/research/wvs/20260916_openrouter/wvs_iw_rated.json index f43f819..1ea539c 100644 --- a/slop/research/wvs/20260916_openrouter/wvs_iw_rated.json +++ b/slop/research/wvs/20260916_openrouter/wvs_iw_rated.json @@ -1,14 +1,14 @@ { "completed": { "00220612ec353054f53baca3eccb90ea2bea8632c2c2fdae99ef7a074bd75d05": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5467309986851505, 0.5724812695525422, - 0.04247123290804327, - 0.09480053501952798 + 0.06104656289606268, + 0.10460593729500693 ], "display_key": "grok-4.3 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "x-ai/grok-4.3", "n_items": 12, "n_samples": 12, @@ -17,14 +17,14 @@ "run_id": "20260917T122425Z_00220612ec35" }, "0043a43d1a2188f7c728ccda08eee7b062232366d8a4e912c72bdc938e37fe44": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6214472858835399, 0.6710244268262847, - 0.020535409516408785, - 0.060983115136406967 + 0.03786483811416081, + 0.06918409770409496 ], "display_key": "kimi-k3 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "moonshotai/kimi-k3", "n_items": 12, "n_samples": 12, @@ -41,6 +41,7 @@ 0.06471194730894558 ], "display_key": "gemini-2.5-flash-lite (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "google/gemini-2.5-flash-lite", "n_items": 12, "n_samples": 12, @@ -49,14 +50,14 @@ "run_id": "20260917T130434Z_009e9bcb8596" }, "02d386b067d55e0639d33db8b6f1e5c708c30acbada3125dcbabde47dc90edcf": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5849525883130219, 0.59521139256589, - 0.02321119834823242, - 0.0670252510886532 + 0.037643038378990704, + 0.07695278664473065 ], "display_key": "qwen2.5-vl-72b-instruct (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen2.5-vl-72b-instruct", "n_items": 12, "n_samples": 12, @@ -64,15 +65,32 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260916T200346Z_02d386b067d5" }, - "0a798666aeed20785b4661c8c2fe4a6fc6853796bc2f3c45133863469e5b3f82": { + "0a72ded0be88127e9009e1c659d63042c0d8d7622ec501f7f6a3a836a4a41527": { "ci_method": "combined item and N-response-mean bootstrap", + "coords": [ + 0.6299053865373309, + 0.6879566393703299, + 0.028990844848675826, + 0.06904089263531353 + ], + "display_key": "gpt-4.1 (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-4.1", + "n_items": 12, + "n_samples": 12, + "protocol_id": "0a72ded0be88127e9009e1c659d63042c0d8d7622ec501f7f6a3a836a4a41527", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T133258Z_0a72ded0be88" + }, + "0a798666aeed20785b4661c8c2fe4a6fc6853796bc2f3c45133863469e5b3f82": { "coords": [ 0.5335737179487179, 0.5337214372928658, - 0.022941344814671694, - 0.09410390751379963 + 0.029444543942379133, + 0.09738610052745941 ], "display_key": "qwen3-vl-30b-a3b-instruct (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-vl-30b-a3b-instruct", "n_items": 12, "n_samples": 12, @@ -81,14 +99,14 @@ "run_id": "20260916T192223Z_0a798666aeed" }, "11929820262d34605efabc00ce84137857e03ef6c53d827ea6c405555b905f26": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6151755041791274, 0.6364257870168759, - 0.05262633347758275, - 0.0809846685946923 + 0.06075682882091517, + 0.0850053325551206 ], "display_key": "gpt-5.4 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5.4", "n_items": 12, "n_samples": 12, @@ -97,14 +115,14 @@ "run_id": "20260917T124555Z_11929820262d" }, "134c21f0377c1a88f8ecfe9e9191341b0f53e0afbadb05d36ca0ba36c641f340": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5672038109538109, 0.6522156084656084, - 0.047966367122167806, - 0.06566980777538725 + 0.0657887759000581, + 0.0790074678539591 ], "display_key": "gemini-3.5-flash-lite (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "google/gemini-3.5-flash-lite", "n_items": 12, "n_samples": 12, @@ -113,14 +131,14 @@ "run_id": "20260917T124448Z_134c21f0377c" }, "1486c2f54e34ad1075f7b519a1dfc2b43f9f7e052d563a083fbc139fc41bd6ee": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5941946410337214, 0.6441922202164339, - 0.03986801497595867, - 0.05382488437105076 + 0.06777104459972463, + 0.07883221721567056 ], "display_key": "glm-4.5-air (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "z-ai/glm-4.5-air", "n_items": 12, "n_samples": 12, @@ -129,14 +147,14 @@ "run_id": "20260917T125444Z_1486c2f54e34" }, "182a7e88869698133f1175f18e53e7f6f2f056739559d325b211bd065235a71c": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6175297619047618, 0.5976147806792967, - 0.03468986624126052, - 0.0429044958309632 + 0.035727022605345674, + 0.07361977335041572 ], "display_key": "qwen3-coder-plus (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-coder-plus", "n_items": 12, "n_samples": 12, @@ -144,15 +162,32 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260916T193258Z_182a7e888696" }, - "1a14a0f088ae0c3cc56637a32bd4c2290e4dde93dfeebb0f888a5a68c784af0d": { + "1a12b248142ab869265f8502e074a7ef4d5e096a8a0c2dee2eb0e632473f1bcb": { "ci_method": "combined item and N-response-mean bootstrap", + "coords": [ + 0.5912804058064387, + 0.5802580057057466, + 0.04010151475516763, + 0.06470588933509897 + ], + "display_key": "gpt-4o-2024-08-06 (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-4o-2024-08-06", + "n_items": 12, + "n_samples": 12, + "protocol_id": "1a12b248142ab869265f8502e074a7ef4d5e096a8a0c2dee2eb0e632473f1bcb", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T134801Z_1a12b248142a" + }, + "1a14a0f088ae0c3cc56637a32bd4c2290e4dde93dfeebb0f888a5a68c784af0d": { "coords": [ 0.5780312225900461, 0.608531746031746, - 0.047255133267023915, - 0.09642087745368229 + 0.06478224722015172, + 0.101427539255001 ], "display_key": "gemini-3.1-flash-lite-preview (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "google/gemini-3.1-flash-lite-preview", "n_items": 12, "n_samples": 12, @@ -161,14 +196,14 @@ "run_id": "20260917T125623Z_1a14a0f088ae" }, "1a70f47789af20900f2799992b09e83af9490be6579098a6178d613d5ba7856c": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5274474357644292, 0.6650744376984001, - 0.039325438704147446, - 0.06431091953768533 + 0.044561828476471346, + 0.06617035838663828 ], "display_key": "gpt-5.6-sol (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5.6-sol", "n_items": 12, "n_samples": 12, @@ -177,14 +212,14 @@ "run_id": "20260916T152843Z_1a70f47789af" }, "1c31668055e399de28805046bc08075035efbad0cdb1ff7d8926841a282e07a5": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5036638266417679, 0.6503283973522069, - 0.04391918485419595, - 0.041591327239427205 + 0.057017134313755345, + 0.06039652604026301 ], "display_key": "qwen3.7-max (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.7-max", "n_items": 12, "n_samples": 12, @@ -193,14 +228,14 @@ "run_id": "20260916T135329Z_1c31668055e3" }, "1d7f4015decbcbe955adb79817b970981ca1d3165ed121b458b213ba1060f675": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6061818582651916, 0.6004689754689754, - 0.029517269745380506, - 0.08002704064836072 + 0.029738189920548014, + 0.07794129978765893 ], "display_key": "qwen3-vl-8b-instruct (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-vl-8b-instruct", "n_items": 12, "n_samples": 12, @@ -208,15 +243,32 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260916T192044Z_1d7f4015decb" }, - "27cc1b32407e6ebca063ce550d1848751a64f4393f93b6af393a01b4f62fe162": { + "252fea691e8f00a7b3a658876047b1d93af94b89f127d549bac93f95644b933c": { "ci_method": "combined item and N-response-mean bootstrap", + "coords": [ + 0.6137290564373897, + 0.5741318151453763, + 0.05391992108407547, + 0.08074548128727346 + ], + "display_key": "gpt-4o-mini (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-4o-mini", + "n_items": 12, + "n_samples": 12, + "protocol_id": "252fea691e8f00a7b3a658876047b1d93af94b89f127d549bac93f95644b933c", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T135119Z_252fea691e8f" + }, + "27cc1b32407e6ebca063ce550d1848751a64f4393f93b6af393a01b4f62fe162": { "coords": [ 0.5442434494517827, 0.6391828271087302, - 0.025233816825009546, - 0.04299513986767339 + 0.04443386786526795, + 0.054686704767473275 ], "display_key": "glm-5.3-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "z-ai/glm-5.3-flash", "n_items": 12, "n_samples": 12, @@ -225,14 +277,14 @@ "run_id": "20260917T114916Z_27cc1b32407e" }, "29b17fd9e04c4b7f93e1763b460892f9ec2f4fc4698b6dd7571fcfebc76b8500": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5429175317746331, 0.694532627865961, - 0.05076181516930638, - 0.06853941563706838 + 0.06309666696112717, + 0.07571315515523741 ], "display_key": "glm-5.1 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "z-ai/glm-5.1", "n_items": 12, "n_samples": 12, @@ -241,14 +293,14 @@ "run_id": "20260917T120826Z_29b17fd9e04c" }, "3016d2c1ab8d8f88f01723d2d695be996c7f1fdcd442a51cfbb6c7b7110a25e9": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5017202097104058, 0.5916881018254176, - 0.014139200341870085, - 0.04877802971162936 + 0.04660057418362078, + 0.07020408831610975 ], "display_key": "qwen3.5-9b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.5-9b", "n_items": 12, "n_samples": 12, @@ -257,14 +309,14 @@ "run_id": "20260916T142212Z_3016d2c1ab8d" }, "304db4b1aa6c80d734f802749c75143feddfcdb740c4131385e927148ba7d05b": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.598778659611993, 0.6204503168788883, - 0.031879884568457334, - 0.08890048164637801 + 0.04059052018905529, + 0.08949774510806 ], "display_key": "qwen3-coder (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-coder", "n_items": 12, "n_samples": 12, @@ -273,14 +325,14 @@ "run_id": "20260916T195554Z_304db4b1aa6c" }, "3194398c99baf1d5ebb273314207d4aa1bfe5d12c8b16cb192818aacd78d759f": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.4215300954189843, 0.7190489551600663, - 0.06886574238472291, - 0.058216440434159446 + 0.07543950734136469, + 0.06523372720374179 ], "display_key": "grok-4.6 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "x-ai/grok-4.6", "n_items": 12, "n_samples": 12, @@ -289,14 +341,14 @@ "run_id": "20260917T114916Z_3194398c99ba" }, "333ece6448f97447c57f24f9e9fcf0a7fb72a1b0f96696aef3c38a9c1d2d2308": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5482108382155133, 0.5760765801607652, - 0.05615879899255038, - 0.06900628612372126 + 0.06725327571313976, + 0.08275579176334262 ], "display_key": "deepseek-v4-flash-0731 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "deepseek/deepseek-v4-flash-0731", "n_items": 12, "n_samples": 12, @@ -304,15 +356,32 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260917T112151Z_333ece6448f9" }, - "3443c17ae66d3fb529a058128b662024c4bc0601394e614bb693b388ec4c988b": { + "340478ea86f503c256128623875d94e71ae4a6f5af06fda18042a3187f03f9ea": { "ci_method": "combined item and N-response-mean bootstrap", + "coords": [ + 0.6100462990851458, + 0.4246338825577172, + 0.04283858171022578, + 0.03714494616987252 + ], + "display_key": "gpt-3.5-turbo-16k (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-3.5-turbo-16k", + "n_items": 12, + "n_samples": 12, + "protocol_id": "340478ea86f503c256128623875d94e71ae4a6f5af06fda18042a3187f03f9ea", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T140450Z_340478ea86f5" + }, + "3443c17ae66d3fb529a058128b662024c4bc0601394e614bb693b388ec4c988b": { "coords": [ 0.4153054353054353, 0.6346365065494385, - 0.06387070667136088, - 0.06142907585948666 + 0.07206749839259896, + 0.07710357656753936 ], "display_key": "qwen3.8-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.8-flash", "n_items": 12, "n_samples": 12, @@ -321,14 +390,14 @@ "run_id": "20260916T133720Z_3443c17ae66d" }, "39659697ca3437f62be4fb8e129e6b3902a91b5acac3243ca914dfc917dd563b": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5030952380952382, 0.5846967846967848, - 0.02599338482636756, - 0.08346488455925544 + 0.046958286642217247, + 0.08746316369004264 ], "display_key": "qwen3-coder-30b-a3b-instruct (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-coder-30b-a3b-instruct", "n_items": 12, "n_samples": 12, @@ -337,14 +406,14 @@ "run_id": "20260916T194257Z_39659697ca34" }, "3c9d4fedebcab41719f4247ca94fa35430f57d0a34f3a539012e0632e35413de": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.4253104631546661, 0.731799663031547, - 0.08709838706511942, - 0.05122062529170766 + 0.0936048666617984, + 0.060012518938080954 ], "display_key": "muse-spark-1.3 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "meta/muse-spark-1.3", "n_items": 12, "n_samples": 12, @@ -352,15 +421,32 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260916T155331Z_3c9d4fedebca" }, - "4230ccaa1e160c793a906aef6b786d735d9fdbf71c3333691b0eba17dcf33928": { + "3f88d93861e3e20746627f7ff664bd93fd39a89d112687ce4c83c7dba73c896d": { "ci_method": "combined item and N-response-mean bootstrap", + "coords": [ + 0.6260950984945237, + 0.5722794251348428, + 0.04638829159697737, + 0.08079941631842184 + ], + "display_key": "gpt-4o-mini-2024-07-18 (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-4o-mini-2024-07-18", + "n_items": 12, + "n_samples": 12, + "protocol_id": "3f88d93861e3e20746627f7ff664bd93fd39a89d112687ce4c83c7dba73c896d", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T135444Z_3f88d93861e3" + }, + "4230ccaa1e160c793a906aef6b786d735d9fdbf71c3333691b0eba17dcf33928": { "coords": [ 0.5790919024614678, 0.6085929478125246, - 0.03990434167170201, - 0.023380867881070636 + 0.038145029256266345, + 0.025980758516149186 ], "display_key": "claude-fable-5.1 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "anthropic/claude-fable-5.1", "n_items": 12, "n_samples": 12, @@ -369,14 +455,14 @@ "run_id": "20260916T153341Z_4230ccaa1e16" }, "43ddc43fb5a29c7fbf21c8f32b42a909c2118feb9255959ecaeab7dadb1f3dde": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5966931216931217, 0.5338064713064713, - 0.011836625928696285, - 0.10630349458372183 + 0.011307868884765215, + 0.10204509849192553 ], "display_key": "qwen3-vl-32b-instruct (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-vl-32b-instruct", "n_items": 12, "n_samples": 12, @@ -385,14 +471,14 @@ "run_id": "20260916T191809Z_43ddc43fb5a2" }, "49ae9f7a09090534a15fac734c2611c445a17139cad6c5684fa0581722f3fd70": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6137896825396826, 0.5819444444444445, - 0.009821057676311395, - 0.08482387932410497 + 0.015958730849444745, + 0.0834429418953706 ], "display_key": "qwen3-14b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-14b", "n_items": 12, "n_samples": 12, @@ -401,14 +487,14 @@ "run_id": "20260916T150121Z_49ae9f7a0909" }, "4f0d6ed370f069e233ea6c4cb83000e79398c3a2aa10b355b10f99c8633c47f6": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5923129031462364, 0.6683453376721874, - 0.02592908625180882, - 0.07625311311101751 + 0.042111115261344804, + 0.08216742430172869 ], "display_key": "glm-5 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "z-ai/glm-5", "n_items": 12, "n_samples": 12, @@ -417,14 +503,14 @@ "run_id": "20260917T123002Z_4f0d6ed370f0" }, "5478eac564ec02f01e161ce017e1814954f4239b9fa1bedfe73172c5c630848f": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.4566709514410664, 0.670101976591992, - 0.0421869487162949, - 0.07271432406347927 + 0.05775584971748114, + 0.07962287120217941 ], "display_key": "gpt-5.4-mini (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5.4-mini", "n_items": 12, "n_samples": 12, @@ -433,14 +519,14 @@ "run_id": "20260917T124216Z_5478eac564ec" }, "54e721734c251e2e13cf43beab962703409108bfc5203c19e7747770ce13a801": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.593257275132275, 0.6186534585344109, - 0.03287309998202738, - 0.06078538490729567 + 0.03571288070419747, + 0.06312818330951275 ], "display_key": "gemini-3.6-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "google/gemini-3.6-flash", "n_items": 12, "n_samples": 12, @@ -449,14 +535,14 @@ "run_id": "20260917T124016Z_54e721734c25" }, "568bb3fc2517dd40b1399c71e0cf52033dcce79dc7537c4943ace860fe3a1297": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6123449635867936, 0.7514717187100795, - 0.03722472860594401, - 0.04262080359788279 + 0.04454897242429451, + 0.049069808495596626 ], "display_key": "gpt-5.1 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5.1", "n_items": 12, "n_samples": 12, @@ -465,14 +551,14 @@ "run_id": "20260917T125409Z_568bb3fc2517" }, "57bed80fd5b94157effe68fe998c7c12612fea210c1371dbf230097a7e8e0717": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6363133579397021, 0.5792251640447779, - 0.015532918535503142, - 0.04894510480027706 + 0.027397680484314686, + 0.06607943334210371 ], "display_key": "qwen3-coder-next (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-coder-next", "n_items": 12, "n_samples": 12, @@ -481,14 +567,14 @@ "run_id": "20260916T191559Z_57bed80fd5b9" }, "580a5840948209ecb177dc6fdaf36e265a32ccfe223d791eb3235fea9269f27e": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6064902599972044, 0.5413085527420844, - 0.018354281932994144, - 0.043773731258587176 + 0.028266888268597475, + 0.07251640240693584 ], "display_key": "qwen3-32b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-32b", "n_items": 12, "n_samples": 12, @@ -497,14 +583,14 @@ "run_id": "20260916T150406Z_580a58409482" }, "589e10fa7924ecf8f88650d6587ff37c440f47c6c4cc4325c7009ef8d4eeb4ec": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6217724867724868, 0.5391326046087952, - 0.013368066261339313, - 0.03510995792353358 + 0.015436127385306617, + 0.05984330714155522 ], "display_key": "qwen3-8b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-8b", "n_items": 12, "n_samples": 12, @@ -513,14 +599,14 @@ "run_id": "20260916T181355Z_589e10fa7924" }, "5c920b68e50c5b4fd43542f379ac25b98e8d79107ea1722c1db12aa22fa34211": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5907043650793651, 0.5776102655592599, - 0.037783384172145155, - 0.07466018958883854 + 0.04137996061944377, + 0.07845514485495916 ], "display_key": "gemini-3.5-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "google/gemini-3.5-flash", "n_items": 12, "n_samples": 12, @@ -529,14 +615,14 @@ "run_id": "20260917T124812Z_5c920b68e50c" }, "6067c0208cc500e5837d123052d6b33b3568d616e5c157f601baa8dcb7ef159b": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5189766385048643, 0.6372836916088428, - 0.0337635437945517, - 0.059640726072916046 + 0.052897459249729474, + 0.08224247455566645 ], "display_key": "grok-4.20 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "x-ai/grok-4.20", "n_items": 12, "n_samples": 12, @@ -544,15 +630,32 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260917T122709Z_6067c0208cc5" }, - "61ab1adae3781ca3893328a845d4d517daa5fb47e07ea189da1df45388f62d42": { + "612c6281a597db4f761707c0c0876d694d715daf35b4969395eeb63fdf9f5fa0": { "ci_method": "combined item and N-response-mean bootstrap", + "coords": [ + 0.5663403168156219, + 0.45394876566876174, + 0.05270303447186722, + 0.03906537267953817 + ], + "display_key": "gpt-4.1-mini (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-4.1-mini", + "n_items": 12, + "n_samples": 12, + "protocol_id": "612c6281a597db4f761707c0c0876d694d715daf35b4969395eeb63fdf9f5fa0", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T133650Z_612c6281a597" + }, + "61ab1adae3781ca3893328a845d4d517daa5fb47e07ea189da1df45388f62d42": { "coords": [ 0.5993276014109347, 0.5462479278681369, - 0.01958469751960041, - 0.1037406905684859 + 0.024660010076102412, + 0.10518394004968958 ], "display_key": "qwen3-vl-235b-a22b-instruct (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-vl-235b-a22b-instruct", "n_items": 12, "n_samples": 12, @@ -561,14 +664,14 @@ "run_id": "20260916T192432Z_61ab1adae378" }, "66ad076e1a37493e4d3db982a0220048e520103459cefc709bc8f09f3fef6c9a": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5509667037841642, 0.5895236753978316, - 0.013511574117086913, - 0.07299126669851562 + 0.02658132977411291, + 0.08063463160776993 ], "display_key": "qwen-2.5-7b-instruct (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen-2.5-7b-instruct", "n_items": 12, "n_samples": 12, @@ -577,14 +680,14 @@ "run_id": "20260916T200857Z_66ad076e1a37" }, "67a70b1b03ba85be79dc122cc888a080fde239205624135a9147168e7e77293d": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5614798614630517, 0.6880988993033613, - 0.04645572992641227, - 0.046994343694421285 + 0.050774441812280406, + 0.057780026773918566 ], "display_key": "inkling (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "thinkingmachines/inkling", "n_items": 12, "n_samples": 12, @@ -593,14 +696,14 @@ "run_id": "20260916T191343Z_67a70b1b03ba" }, "6af9f017856f4c733214bf8e26992a4af3c7cb94dac8a5a4b881e8669b3917a7": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5220108722106056, 0.6395004417804221, - 0.03076973529726908, - 0.084649994895056 + 0.04704769940966264, + 0.09470717560391326 ], "display_key": "gpt-5-mini (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5-mini", "n_items": 12, "n_samples": 12, @@ -609,14 +712,14 @@ "run_id": "20260917T130236Z_6af9f017856f" }, "6d6ce30e9fbc5fdb07229b0a08e067d77511857069a4e572613d3e48b7c8b261": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5856201899951899, 0.6494018547589976, - 0.032344011142651626, - 0.08396736786485802 + 0.06114886163791156, + 0.08129405199775336 ], "display_key": "qwen3.5-plus-20260420 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.5-plus-20260420", "n_items": 12, "n_samples": 12, @@ -625,14 +728,14 @@ "run_id": "20260916T135722Z_6d6ce30e9fbc" }, "6f7a7bbc4d70e7cfceff8e25b2da2008fbf6105538825f42bfbbc47ff3459387": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5605366161616162, 0.573015873015873, - 0.07041699717338515, - 0.0955132767170434 + 0.07008299148165956, + 0.09640175208725496 ], "display_key": "gemini-3.1-flash-lite (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "google/gemini-3.1-flash-lite", "n_items": 12, "n_samples": 12, @@ -641,14 +744,14 @@ "run_id": "20260917T125248Z_6f7a7bbc4d70" }, "7058adf367d238c6c92e1b83ae71f6556362e9fc7a243297b6fc65e54e2c2886": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5974338624338624, 0.6515634387658198, - 0.0151114134522649, - 0.07890400348024484 + 0.023795241546514957, + 0.07922810144513995 ], "display_key": "qwen3.5-27b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.5-27b", "n_items": 12, "n_samples": 12, @@ -657,14 +760,14 @@ "run_id": "20260916T143235Z_7058adf367d2" }, "74e3649a6961000bab284e801fe811d5c8fc4cf1db15af52bb10959f5a5a37c1": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5372045855379188, 0.6595114541543113, - 0.03701076833492129, - 0.08376087073198286 + 0.04180484013035134, + 0.08450011787948054 ], "display_key": "qwen3.5-plus-02-15 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.5-plus-02-15", "n_items": 12, "n_samples": 12, @@ -673,14 +776,14 @@ "run_id": "20260916T144205Z_74e3649a6961" }, "7809cffa4ab1327c0a20b2d9ec23d0647b9382ddef65dd36c2ab5669da4f7d9f": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5595998356229528, 0.5926290303237558, - 0.02693093173377315, - 0.042768178388407095 + 0.04045393842948997, + 0.058480296424785895 ], "display_key": "glm-4.7-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "z-ai/glm-4.7-flash", "n_items": 12, "n_samples": 12, @@ -689,14 +792,14 @@ "run_id": "20260917T123743Z_7809cffa4ab1" }, "7ee20ca2b88f69dd06b9de3018deea4cb7a418e3119e17862cd0b32146daae1a": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.538440257742794, 0.5530118458934986, - 0.03744969864324434, - 0.06998170957293813 + 0.07277582530416211, + 0.0898393391241473 ], "display_key": "qwen3.5-35b-a3b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.5-35b-a3b", "n_items": 12, "n_samples": 12, @@ -705,14 +808,14 @@ "run_id": "20260916T142959Z_7ee20ca2b88f" }, "7fe76f95937ed145b996116a31855dae4ab8c3b67e3e7ccfb84a000437e27f96": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.4526851851851852, 0.6334977783064814, - 0.036885948359860296, - 0.07744723420845886 + 0.057818785990183905, + 0.08298095716984875 ], "display_key": "gpt-5-nano (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5-nano", "n_items": 12, "n_samples": 12, @@ -721,14 +824,14 @@ "run_id": "20260917T020418Z_7fe76f95937e" }, "82875b6ee164d0d980eba738bf05f69959234567b91f9cfd01fcabca0bd70dac": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6529354469060352, 0.5885508561103798, - 0.0314816537313644, - 0.04799521946452996 + 0.049299882651159914, + 0.08073268479006059 ], "display_key": "qwen3.7-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.7-flash", "n_items": 12, "n_samples": 12, @@ -737,14 +840,14 @@ "run_id": "20260916T133001Z_82875b6ee164" }, "840da516c1c04e69e5d80dd104803f019a4d6c6640359ad0e4bbcb71e722aa6c": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5595320767195766, 0.6334656084656085, - 0.0479345416110304, - 0.06407749490095412 + 0.05158017344603731, + 0.06609164737112991 ], "display_key": "gemini-3-flash-preview (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "google/gemini-3-flash-preview", "n_items": 12, "n_samples": 12, @@ -753,14 +856,14 @@ "run_id": "20260917T125926Z_840da516c1c0" }, "842298c634ec0f57848b605f67fca161631a852a68192cae1a5377b1d9f31566": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6427511094043352, 0.5554953612269281, - 0.03253928279038274, - 0.07779247297520335 + 0.03535208249381765, + 0.08497138486454137 ], "display_key": "qwen3-235b-a22b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-235b-a22b", "n_items": 12, "n_samples": 12, @@ -769,14 +872,14 @@ "run_id": "20260916T150950Z_842298c634ec" }, "8700e79d3eccf358a0de1029f7b40e2db339bb6d11403b8e69cb7a66a7b0c668": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6707823388714307, 0.6060018656291949, - 0.022702792353656083, - 0.08121800045234231 + 0.03451993837104684, + 0.09876091230627733 ], "display_key": "qwen3-30b-a3b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-30b-a3b", "n_items": 12, "n_samples": 12, @@ -785,14 +888,14 @@ "run_id": "20260916T145439Z_8700e79d3ecc" }, "88ba5af838eda5d298b312553d7c941e72a37cbf28890cdb9d8436b5f40e7cfe": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6025617283950616, 0.6287037037037038, - 0.030065943475536042, - 0.09420661909505727 + 0.03396367302553316, + 0.09406697361625513 ], "display_key": "qwen-plus (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen-plus", "n_items": 12, "n_samples": 12, @@ -801,14 +904,14 @@ "run_id": "20260916T200626Z_88ba5af838ed" }, "88ed922f271a76721ad18f0b7bb9693a48377e2507d02a4b78fa08b451972d83": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6450907949109246, 0.5437992771640354, - 0.014480282059694862, - 0.10414669190782924 + 0.023312776383531648, + 0.10024881542221994 ], "display_key": "qwen3-30b-a3b-instruct-2507 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-30b-a3b-instruct-2507", "n_items": 12, "n_samples": 12, @@ -817,14 +920,14 @@ "run_id": "20260916T195128Z_88ed922f271a" }, "8c27d32cd851de5f217cb354b15dc270cc6f4ec2d02fb15dbb20fde0790c3571": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5419602854216832, 0.595896730624977, - 0.044286454927165, - 0.06410119405336592 + 0.05193570378692634, + 0.08265065423220756 ], "display_key": "deepseek-v4.1-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "deepseek/deepseek-v4.1-flash", "n_items": 12, "n_samples": 12, @@ -833,14 +936,14 @@ "run_id": "20260916T151414Z_8c27d32cd851" }, "921707cb0d3db143e948e9d78cd8191f4ee720eaf56b6934e2d13d9ce9e3382f": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5408168991502326, 0.6493278669204595, - 0.03453680502309858, - 0.08951355314543964 + 0.03917823180041766, + 0.08760189869507796 ], "display_key": "qwen3.5-397b-a17b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.5-397b-a17b", "n_items": 12, "n_samples": 12, @@ -849,14 +952,14 @@ "run_id": "20260916T144546Z_921707cb0d3d" }, "94428060f3776ab79ab3d146382e869d248c259c86858f4e5dbc9c5a1f4633db": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.48311026165192833, 0.6007697469838861, - 0.028453155979336744, - 0.04236719989377499 + 0.05313376200674668, + 0.06690255611782707 ], "display_key": "gpt-5.4-nano (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5.4-nano", "n_items": 12, "n_samples": 12, @@ -865,14 +968,14 @@ "run_id": "20260917T123850Z_94428060f377" }, "95bb4d3939e9937823357b5cd87adb1a6640cc5b04e0841dfec03095331a5e44": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.4569019748367575, 0.6751145899955423, - 0.07186088017315642, - 0.04558197261286701 + 0.07644348564506928, + 0.0567289888770763 ], "display_key": "gpt-6-astra (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-6-astra", "n_items": 12, "n_samples": 12, @@ -881,14 +984,14 @@ "run_id": "20260916T154406Z_95bb4d3939e9" }, "96d86b64bf7ceff8eed0daff1aa0396ee69609b5867942a7c2c9d33093f03c9d": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5196778711484593, 0.6698994127565557, - 0.03377693173708851, - 0.06927660087614809 + 0.042934824935707064, + 0.07801121120269483 ], "display_key": "qwen3.6-plus (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.6-plus", "n_items": 12, "n_samples": 12, @@ -896,15 +999,32 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260916T141839Z_96d86b64bf7c" }, - "9c1d40b30945565e488b212177e40c29f730c2e96d4b07d09cbc135ba75a81f0": { + "9933b4a3be52fec8f416cf1d8bd3cefa1718efa1f0177f307292404d3e9db105": { "ci_method": "combined item and N-response-mean bootstrap", + "coords": [ + 0.6060268571393299, + 0.5841620141118888, + 0.03342395986220364, + 0.06247093532554917 + ], + "display_key": "gpt-4o (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-4o", + "n_items": 12, + "n_samples": 12, + "protocol_id": "9933b4a3be52fec8f416cf1d8bd3cefa1718efa1f0177f307292404d3e9db105", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T135801Z_9933b4a3be52" + }, + "9c1d40b30945565e488b212177e40c29f730c2e96d4b07d09cbc135ba75a81f0": { "coords": [ 0.6026546601546601, 0.5957470912477916, - 0.024251712034204544, - 0.08684627683495146 + 0.03274570154412936, + 0.08673540897470654 ], "display_key": "qwen-2.5-72b-instruct (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen-2.5-72b-instruct", "n_items": 12, "n_samples": 12, @@ -913,14 +1033,14 @@ "run_id": "20260916T201119Z_9c1d40b30945" }, "9d8a5189c078672eb884784938b183944fb161d0e716e2c44722540b17603e34": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6652380952380952, 0.7022990677752582, - 0.021284606679632966, - 0.04934275921145735 + 0.026433969418704885, + 0.06269095940430941 ], "display_key": "qwen3-next-80b-a3b-instruct (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-next-80b-a3b-instruct", "n_items": 12, "n_samples": 12, @@ -929,14 +1049,14 @@ "run_id": "20260916T193846Z_9d8a5189c078" }, "a2df2f1a32fde5dae2c1545120a4fd72cea4743bafe71a1768997dc3d125b4f4": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.610243091658148, 0.6687657686659433, - 0.03675410818998801, - 0.06737179091474468 + 0.04414498817375987, + 0.08085439770187478 ], "display_key": "kimi-k2.6 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "moonshotai/kimi-k2.6", "n_items": 12, "n_samples": 12, @@ -953,6 +1073,7 @@ 0.07471544119511193 ], "display_key": "gemini-2.5-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "google/gemini-2.5-flash", "n_items": 12, "n_samples": 12, @@ -961,14 +1082,14 @@ "run_id": "20260917T130734Z_a544b24e1874" }, "a742f028f72ec35cc6586661ca261df77a51eecce778ff080062968942d822ae": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5253620889368438, 0.6943301071633757, - 0.03998834571383578, - 0.06608021212309273 + 0.06029287144663362, + 0.07390235325207269 ], "display_key": "glm-5.2 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "z-ai/glm-5.2", "n_items": 12, "n_samples": 12, @@ -977,14 +1098,14 @@ "run_id": "20260917T115959Z_a742f028f72e" }, "a9a69a3ef43e7544f73290e6fd9d3c9f37013b15a9e50be755d4d227d6cc30eb": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.524187458188895, 0.708147541423977, - 0.04698755670397484, - 0.0578745248478481 + 0.05959262807078437, + 0.06778136701328388 ], "display_key": "grok-4.5 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "x-ai/grok-4.5", "n_items": 12, "n_samples": 12, @@ -993,14 +1114,14 @@ "run_id": "20260917T121142Z_a9a69a3ef43e" }, "abf45a27f4bafad0eb264aac33b1e49e9d0404e141966954d0707930d56665e3": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.45740779531102105, 0.6795887184190553, - 0.07173891141983318, - 0.06455376828361989 + 0.09233310052166174, + 0.07166972793043255 ], "display_key": "qwen3.6-27b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.6-27b", "n_items": 12, "n_samples": 12, @@ -1009,14 +1130,14 @@ "run_id": "20260916T141139Z_abf45a27f4ba" }, "ac52652ff1f09e1f1972a8b751e1b1deca329cb555df54755aada79791bed940": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6417361111111111, 0.5127110103300581, - 0.03306211082399865, - 0.06459320760155234 + 0.03807268153101321, + 0.08042423628815949 ], "display_key": "qwen-plus-2025-07-28 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen-plus-2025-07-28", "n_items": 12, "n_samples": 12, @@ -1024,15 +1145,32 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260916T194044Z_ac52652ff1f0" }, - "b35b916c22649e91ded175aede34f3937a16f42a354f7abab01a17bf2b0dfc97": { + "b25b19e78ceb3fe7814ac5883239382f1c6e757fe89329495241a25ab521e178": { "ci_method": "combined item and N-response-mean bootstrap", + "coords": [ + 0.6091300100287029, + 0.438086083059628, + 0.04177939867778623, + 0.0411635510827436 + ], + "display_key": "gpt-3.5-turbo-0613 (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-3.5-turbo-0613", + "n_items": 12, + "n_samples": 12, + "protocol_id": "b25b19e78ceb3fe7814ac5883239382f1c6e757fe89329495241a25ab521e178", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T140156Z_b25b19e78ceb" + }, + "b35b916c22649e91ded175aede34f3937a16f42a354f7abab01a17bf2b0dfc97": { "coords": [ 0.6062280543530544, 0.5943180448406931, - 0.03134706527371011, - 0.04910757543789796 + 0.04853610852408334, + 0.0737314840555176 ], "display_key": "qwen3.6-35b-a3b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.6-35b-a3b", "n_items": 12, "n_samples": 12, @@ -1040,15 +1178,31 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260916T140418Z_b35b916c2264" }, + "b5a600e549f8aaef4077c383d11a93d6b1f4395af26e0cc0f8f50d3b6f360996": { + "coords": [ + 0.47891715945637514, + 0.5996774327014757, + 0.058990630830522965, + 0.09038008381157654 + ], + "display_key": "gpt-oss-120b (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-oss-120b", + "n_items": 12, + "n_samples": 12, + "protocol_id": "b5a600e549f8aaef4077c383d11a93d6b1f4395af26e0cc0f8f50d3b6f360996", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T130652Z_b5a600e549f8" + }, "b5fdfe76425ee50c9f475587863990f2f6fd088d7f64c5c5119e7062a6a1396b": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.552530712157149, 0.63810610515893, - 0.0377567739081018, - 0.08012242145033936 + 0.053278822950175, + 0.08499253722116974 ], "display_key": "gpt-5.2 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5.2", "n_items": 12, "n_samples": 12, @@ -1056,15 +1210,32 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260917T125001Z_b5fdfe76425e" }, - "bca0745bdc1554718d57891098f366214f05e9e896007a916d14a06efa33471a": { + "b6966daf09695284896ac4f0dda90cadb5e6fce5a1a77b576b290089050d0c9e": { "ci_method": "combined item and N-response-mean bootstrap", + "coords": [ + 0.48177082177082176, + 0.4976680628265054, + 0.03020402680981063, + 0.05335291530014508 + ], + "display_key": "gpt-4.1-nano (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-4.1-nano", + "n_items": 12, + "n_samples": 12, + "protocol_id": "b6966daf09695284896ac4f0dda90cadb5e6fce5a1a77b576b290089050d0c9e", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T134036Z_b6966daf0969" + }, + "bca0745bdc1554718d57891098f366214f05e9e896007a916d14a06efa33471a": { "coords": [ 0.5778139029180694, 0.6959770802908057, - 0.034466398666598765, - 0.07823187669222507 + 0.03666290866643542, + 0.07382269298741832 ], "display_key": "qwen3-max-thinking (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-max-thinking", "n_items": 12, "n_samples": 12, @@ -1073,14 +1244,14 @@ "run_id": "20260916T145109Z_bca0745bdc15" }, "be8b6978ea49ace083544a82e1837e17898ca264b959e08939da01a63424e2f9": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.56922050588427, 0.7223026320129816, - 0.02710061973263113, - 0.0504019572127584 + 0.036665367096187215, + 0.053812885345722886 ], "display_key": "gpt-5 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5", "n_items": 12, "n_samples": 12, @@ -1089,14 +1260,14 @@ "run_id": "20260917T125811Z_be8b6978ea49" }, "c2911249c065c9b0cd2f69104d1f931bf8783940bd503fd60aa99ec6c6c0a761": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5489586575370888, 0.6356739354917721, - 0.043048462200454045, - 0.08272361735803685 + 0.05257011593980756, + 0.0883451278287498 ], "display_key": "gpt-5.6-terra (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5.6-terra", "n_items": 12, "n_samples": 12, @@ -1105,14 +1276,14 @@ "run_id": "20260917T123400Z_c2911249c065" }, "c4e310179ab86b9d786c221d8bab6b6891f42d18836a8affc64249769eaffdef": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5713445083420574, 0.5644094213660266, - 0.022694588339844418, - 0.07410136506739047 + 0.04068128275908349, + 0.08447489368900887 ], "display_key": "deepseek-v3.2 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "deepseek/deepseek-v3.2", "n_items": 12, "n_samples": 12, @@ -1121,14 +1292,14 @@ "run_id": "20260917T114916Z_c4e310179ab8" }, "c64ced76e82ec32cf6be896420d813403a95a421474eefcbbcbba421f8189192": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.48767589171684006, 0.6482142857142857, - 0.024847625255712682, - 0.0636935529346349 + 0.04852805951852839, + 0.06690143204728374 ], "display_key": "muse-glimmer-30b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "meta/muse-glimmer-30b", "n_items": 12, "n_samples": 12, @@ -1137,14 +1308,14 @@ "run_id": "20260917T111805Z_c64ced76e82e" }, "c7e906b267d13cb638b0a646e9643e15bd8917152cb4ef6a2e888ba309501565": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5803524402579809, 0.6040068811440846, - 0.03394817417921382, - 0.07258813270003692 + 0.04571011970861469, + 0.08797307424268631 ], "display_key": "deepseek-v3.2-exp (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "deepseek/deepseek-v3.2-exp", "n_items": 12, "n_samples": 12, @@ -1153,14 +1324,14 @@ "run_id": "20260917T115739Z_c7e906b267d1" }, "cc5d6fdd3ce85c73322e963fe44186c44015427f93cdbcecec838da8b2dd4ab4": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5896480994850559, 0.5981282109666797, - 0.03137406859274463, - 0.08079858766631393 + 0.035112198239544395, + 0.08647066879117997 ], "display_key": "deepseek-chat-v3-0324 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "deepseek/deepseek-chat-v3-0324", "n_items": 12, "n_samples": 12, @@ -1169,14 +1340,14 @@ "run_id": "20260917T122110Z_cc5d6fdd3ce8" }, "cd5db529649a179032180cecafe4fd98aff124ee3654f7d60996de704e2b63ef": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.4988977072310405, 0.6262540369683227, - 0.006279760963488564, - 0.04730514024812738 + 0.01537866158910887, + 0.05342525613259921 ], "display_key": "gemini-3.7-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "google/gemini-3.7-flash", "n_items": 12, "n_samples": 12, @@ -1185,14 +1356,14 @@ "run_id": "20260916T172946Z_cd5db529649a" }, "d21d1e81010d87b7f79bd9ceb0d5bb226485fe6393e266e169474968e6bf26f8": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5886813794915646, 0.6339803982919925, - 0.04261590693561221, - 0.04660781411126606 + 0.04800287084122153, + 0.07143925775678789 ], "display_key": "deepseek-v4-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "deepseek/deepseek-v4-flash", "n_items": 12, "n_samples": 12, @@ -1201,14 +1372,14 @@ "run_id": "20260917T112521Z_d21d1e81010d" }, "d4580ea120a65129dc484875ee13aa8a721bb133a93a18b19f974cc876f00178": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.619008820472186, 0.6035800722347154, - 0.020634824655178182, - 0.07295306220868648 + 0.0369775280569868, + 0.08938873815444547 ], "display_key": "deepseek-chat-v3.1 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "deepseek/deepseek-chat-v3.1", "n_items": 12, "n_samples": 12, @@ -1217,14 +1388,14 @@ "run_id": "20260917T121651Z_d4580ea120a6" }, "d81f7e66b3c4c720f8dd80768d3dfc25de6edeaf6a352cfed40e961f21fcfa22": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5535321928702047, 0.6422322099171051, - 0.03217817886332268, - 0.05342139005960689 + 0.05892062733798582, + 0.06715990340387798 ], "display_key": "glm-5.3 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "z-ai/glm-5.3", "n_items": 12, "n_samples": 12, @@ -1233,14 +1404,14 @@ "run_id": "20260916T161456Z_d81f7e66b3c4" }, "da8a3ff2c0318113bd43c5d8b1813b9d0947fcd6330bfe600ec28a203e6507b8": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5974867724867725, 0.5792094488085178, - 0.03608202052858709, - 0.0970663887466463 + 0.04341830148688508, + 0.09828751212663685 ], "display_key": "gpt-5.6-luna (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "openai/gpt-5.6-luna", "n_items": 12, "n_samples": 12, @@ -1249,14 +1420,14 @@ "run_id": "20260917T122950Z_da8a3ff2c031" }, "dc09c3fe433c95591b336b279603c546152daf6534d90699323b3bcde754f5bf": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6353621031746031, 0.5685931857310826, - 0.035301389540358157, - 0.055707793544840394 + 0.03688413980467388, + 0.08024381413619414 ], "display_key": "qwen3-235b-a22b-2507 (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3-235b-a22b-2507", "n_items": 12, "n_samples": 12, @@ -1265,14 +1436,14 @@ "run_id": "20260916T195827Z_dc09c3fe433c" }, "dc89e0b249b6189f2f85b9c5eebd98efb1595ea416761f9c2803ca57324275f6": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.3810994858696008, 0.6397686012218032, - 0.07770722831755433, - 0.03612058315515798 + 0.08204078442843406, + 0.039637685403670346 ], "display_key": "gemini-3.8-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "google/gemini-3.8-flash", "n_items": 12, "n_samples": 12, @@ -1281,14 +1452,14 @@ "run_id": "20260917T123103Z_dc89e0b249b6" }, "e0a9daef251d975c5e805e159f3813dd0c55b3f0919876b828e572b0530bea88": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.48469169991228817, 0.6190010643383659, - 0.032452612491036585, - 0.06512265958871256 + 0.07699729945533007, + 0.07041604518532184 ], "display_key": "qwen3.5-122b-a10b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.5-122b-a10b", "n_items": 12, "n_samples": 12, @@ -1297,14 +1468,14 @@ "run_id": "20260916T143453Z_e0a9daef251d" }, "e3abbf27f8d036c12dd728740d2c263bb9be98f160437d5af3cb05b33dbfa761": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5255845653943481, 0.6176120675327024, - 0.030723613091297796, - 0.05017065754814348 + 0.03757041296270222, + 0.06469679141919639 ], "display_key": "qwen3.8-2.4t-a95b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.8-2.4t-a95b", "n_items": 12, "n_samples": 12, @@ -1312,15 +1483,32 @@ "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", "run_id": "20260917T122025Z_e3abbf27f8d0" }, - "ed8190c48b2a3778bba8afff7381bb9f1578211b5e8f750d21321b62617c82c3": { + "eb1402c60a7305f5c9b800e5682ce648f3acc25970eaf798d0c3485b4bb16d48": { "ci_method": "combined item and N-response-mean bootstrap", + "coords": [ + 0.6174577809885475, + 0.6028457218832352, + 0.03261563618849975, + 0.051118728421761774 + ], + "display_key": "gpt-4o-2024-11-20 (rated)", + "eval_version": "wvs-score-all-options-v1", + "model": "openai/gpt-4o-2024-11-20", + "n_items": 12, + "n_samples": 12, + "protocol_id": "eb1402c60a7305f5c9b800e5682ce648f3acc25970eaf798d0c3485b4bb16d48", + "records_path": "slop/research/wvs/20260916_openrouter/wvs_iw_requests.jsonl", + "run_id": "20260917T134443Z_eb1402c60a73" + }, + "ed8190c48b2a3778bba8afff7381bb9f1578211b5e8f750d21321b62617c82c3": { "coords": [ 0.431934218610816, 0.6881570439623662, - 0.06843286443972658, - 0.05955561989504253 + 0.08092059548185657, + 0.07158456159738981 ], "display_key": "qwen3.8-27b (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.8-27b", "n_items": 12, "n_samples": 12, @@ -1329,14 +1517,14 @@ "run_id": "20260916T134026Z_ed8190c48b2a" }, "f095e0d2dbafdb91b51be3d0b3bd5a597730c8e3c397e0461023c87e924d92f8": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.5078196649029982, 0.6664359119716262, - 0.040332760807123835, - 0.07218684667619854 + 0.04898292658545207, + 0.07513374670738206 ], "display_key": "qwen3.7-plus (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.7-plus", "n_items": 12, "n_samples": 12, @@ -1345,14 +1533,14 @@ "run_id": "20260916T134925Z_f095e0d2dbaf" }, "fa37fc90604e5032183ec460bcc2632507566c3be3e9085f61a6ca85582b7d47": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.6514346395123706, 0.583014517676874, - 0.04021271468677619, - 0.04160867116289502 + 0.044355827535678856, + 0.07772092035914374 ], "display_key": "qwen3.6-flash (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.6-flash", "n_items": 12, "n_samples": 12, @@ -1361,14 +1549,14 @@ "run_id": "20260916T140130Z_fa37fc90604e" }, "fa9ff591d389ad2d8a4b7e3fc1410ed588508fe286430c06a941180b0590422b": { - "ci_method": "combined item and N-response-mean bootstrap", "coords": [ 0.4805424128340796, 0.6905543787488233, - 0.03427460911867346, - 0.06690339494976041 + 0.08346396425450253, + 0.07197747047181104 ], "display_key": "qwen3.6-max-preview (rated)", + "eval_version": "wvs-score-all-options-v1", "model": "qwen/qwen3.6-max-preview", "n_items": 12, "n_samples": 12, diff --git a/src/moralmaps/maps.py b/src/moralmaps/maps.py index 201ca88..a499a8c 100644 --- a/src/moralmaps/maps.py +++ b/src/moralmaps/maps.py @@ -298,7 +298,8 @@ def plot_value_map(display: str, countries: list[str], P: np.ndarray, model_labels: dict[str, str] | None = None, steer: dict[str, tuple[float, float, str]] | None = None, emphasize: set[str] | None = None, - title: str | None = None, note: str | None = None): + title: str | None = None, note: str | None = None, + title_y: float = 0.075, note_y: float = 0.02): """The interpretable "4-value map": two NAMED axes with four pole signposts through the human MEDIAN crosshair, Economist-style zone hulls (the 4 most-separate zones), zone-coloured dots, and auto-placed labels (landmarks + corner outliers + one representative per zone + any models; see @@ -420,10 +421,10 @@ def plot_value_map(display: str, countries: list[str], P: np.ndarray, # INSIDE the axes, in the empty bottom-left corner, so a crop of the PNG can't strip the attribution # and there's no external white band. -- Claude if title: - ax.text(0.012, 0.075, title, transform=ax.transAxes, ha="left", va="bottom", + ax.text(0.012, title_y, title, transform=ax.transAxes, ha="left", va="bottom", fontsize=11, fontweight="bold", color="#333", zorder=11, linespacing=1.05) if note: - ax.text(0.012, 0.02, note, transform=ax.transAxes, ha="left", va="bottom", + ax.text(0.012, note_y, note, transform=ax.transAxes, ha="left", va="bottom", fontsize=7.5, color="#8a857a", zorder=11, linespacing=1.1) return fig diff --git a/src/moralmaps/rated_cache.py b/src/moralmaps/rated_cache.py index 4d19177..4258be6 100644 --- a/src/moralmaps/rated_cache.py +++ b/src/moralmaps/rated_cache.py @@ -37,3 +37,11 @@ def update_coords(path: Path, coords: dict[str, list[float]]) -> dict: completed[protocol_id]["coords"] = values completed[protocol_id]["ci_method"] = "combined item and N-response-mean bootstrap" return _write_locked(path, update) + + +def update_eval_versions(path: Path, versions: dict[str, str]) -> dict: + """Add evaluator provenance without changing cached coordinates or run records.""" + def update(completed: dict[str, dict]) -> None: + for protocol_id, version in versions.items(): + completed[protocol_id]["eval_version"] = version + return _write_locked(path, update)