·${columns.length} ${state.view==='subcategories'?'subcategories':'task types'}`;
$('#reset-filters').hidden=!(state.query||state.group!=='all'||state.input!=='all');
$$('#group-filters button').forEach(el=>el.setAttribute('aria-pressed',el.dataset.group===state.group));
renderTray();
}
function renderTray() {
$('#compare-tray').hidden=state.selected.length===0;
$('#compare-count').textContent=state.selected.length+' / 3 models selected';
$('#compare-chips').innerHTML=state.selected.map(id=>`${esc(modelById(id).name)}`).join('');
$('#compare-open').disabled=state.selected.length<2;
updateTraySpacing();
}
function updateTraySpacing() {
document.body.style.paddingBottom=state.selected.length?($('#compare-tray').offsetHeight+48)+'px':'';
}
if ('ResizeObserver' in window) new ResizeObserver(updateTraySpacing).observe($('#compare-tray'));
window.addEventListener('resize',updateTraySpacing);
function toggleSelected(id,checked) {
if(checked&&state.selected.length>=3){toast('Compare up to 3 models at a time. Remove one to add another.');renderTable();return;}
state.selected=checked?[...state.selected,id]:state.selected.filter(x=>x!==id);renderTable();
}
function resetFilters() {
state.query='';state.group='all';state.input='all';
$('#model-search').value='';$('#input-filter').value='all';renderTable();
}
function switchBenchmark(id) {
state.benchmark=id;state.selected=[];state.group='all';state.input='all';state.view='accuracy';state.sort='total';state.direction='desc';
state.query='';$('#model-search').value='';$('#input-filter').value='all';
state.visible=availableColumns().map(c=>c.key);
$$('[data-benchmark]').forEach(btn=>btn.setAttribute('aria-pressed',btn.dataset.benchmark===id));
$$('[data-view]').forEach(btn=>btn.setAttribute('aria-pressed',btn.dataset.view===state.view));
renderResources();renderTable();saveState();
}
function switchLanguage(language) {
if(!benchmarkMetadata().results_by_language[language] || language===languageCode())return;
state.language=language;state.selected=[];
renderResources();renderTable();saveState();
}
function switchView(view) {
if(!isPlus()&&['subcategories','task_types'].includes(view))return;
state.view=view;
if(view==='task_types')state.subcategory=benchmark().subcategory_columns.find(c=>state.dimension==='all'||c.dimension===state.dimension).key;
state.sort=isBreakdown()?availableColumns()[0].key:view==='accuracy'?'total':'instruction_reasoning';state.direction='desc';state.visible=availableColumns().map(c=>c.key);
$$('[data-view]').forEach(btn=>btn.setAttribute('aria-pressed',btn.dataset.view===view));
renderBreakdownControls();renderColumns();renderTable();$('.table-scroll').scrollLeft=0;
}
function download(name,content,type) {
const url=URL.createObjectURL(new Blob([content],{type}));const a=document.createElement('a');a.href=url;a.download=name;document.body.appendChild(a);a.click();a.remove();setTimeout(()=>URL.revokeObjectURL(url),1000);
}
function exportCSV(models,name='filtered') {
const b=benchmark(), mainFields=[...(isPlus()?orderedAccuracyColumns():b.accuracy_columns).map(c=>({type:'accuracy',...c})),...b.metrics.map(c=>({type:'metrics',...c}))];
const fields=isBreakdown()?[...(name==='comparison'?mainFields:[]),...availableColumns().filter(c=>state.visible.includes(c.key)).map(c=>({type:state.view,...c,label:qualifiedLabel(c)}))]:mainFields;
const safeCSV=value=>{let t=String(value??'');if(/^\s*[=+@-]|^[\t\r\n]/.test(t))t="'"+t;return '"'+t.replace(/"/g,'""')+'"';};
const rows=[['Benchmark','Result language','Model','Category','Multi-image support',...fields.map(c=>c.label+(c.type==='metrics'?' (/100)':' (%)'))],...models.map(m=>[b.name,b.language,m.name,groupLabels[m.group]||'Not specified',m.multi_image===null?'Not specified':m.multi_image?'Yes':'No',...fields.map(c=>m[c.type][c.key]===null?'':fmt(m[c.type][c.key]))])];
download(`${b.id}-${languageCode()}-${state.view}-${name}-results.csv`,'\uFEFF'+rows.map(r=>r.map(safeCSV).join(',')).join('\r\n'),'text/csv;charset=utf-8');toast(`Exported ${models.length} models · ${resultLanguageLabel()} results.`);
}
function radar(models) {
const dims=benchmark().dimensions, cx=205,cy=156,r=98,n=dims.length;
const point=(i,radius)=>{const a=-Math.PI/2+i*2*Math.PI/n;return [cx+Math.cos(a)*radius,cy+Math.sin(a)*radius];};
const polygon=radius=>dims.map((_,i)=>point(i,radius).map(x=>x.toFixed(1)).join(',')).join(' ');
const rings=[20,40,60,80,100].map(v=>`${v}`).join('');
const lines=dims.map((d,i)=>{const p=point(i,r),l=point(i,r+24);return `${esc(d.label)}`;}).join('');
const plots=models.map((m,j)=>{const points=dims.map((d,i)=>point(i,r*(m.accuracy[d.key]??0)/100));return ``+points.map(p=>``).join('');}).join('');
return ``;
}
function showDialog(eyebrow,html) {
$('#dialog-eyebrow').textContent=eyebrow;$('#dialog-content').innerHTML=html;
$('#detail-dialog').showModal();$('#detail-dialog').scrollTop=0;
}
function showModels(ids) {
const b=benchmark(), models=ids.map(modelById).filter(Boolean);if(!models.length)return;
const mixed=isPlus()&&new Set(models.map(m=>m.multi_image)).size>1;
const tableSection=(title,fields,type)=>`| ${title} |
`+fields.map(c=>{const vals=models.map(m=>m[type]?.[c.key] ?? null);const max=Math.max(...vals.filter(v=>v!==null));return `| ${esc(c.label)} | ${vals.map(v=>`${fmt(v)}${v!==null&&type!=='metrics'?'%':''} | `).join('')}
`;}).join('');
showDialog(b.name+' · '+resultLanguageLabel().toUpperCase()+' RESULTS',`${models.length===1?esc(models[0].name):'Model comparison'}
${mixed?'These models use different evaluation subsets: single-image only and full benchmark.
':''}${radar(models)}
${models.map((m,i)=>`
${esc(m.name)}
${esc(groupLabels[m.group]||b.name)}${isPlus()?' · '+(m.multi_image?'Multi-image supported':'Single-image only'):''}
${isPlus()?'Total':'Overall'}`).join('')}${models.some(m=>m.accuracy.total===null)?'
Single-image-only models have no Total score.
':''}
| Metric | ${models.map((m,i)=>`${esc(m.name)} | `).join('')}
|---|
${tableSection('REASONING ACCURACY (%)',isPlus()?orderedAccuracyColumns():b.accuracy_columns,'accuracy')}${tableSection('EVALUATION SCORES (/100)',b.metrics,'metrics')}${isBreakdown()?tableSection(viewLabel().toUpperCase()+' (%)',availableColumns().filter(c=>state.visible.includes(c.key)).map(c=>({...c,label:qualifiedLabel(c)})),state.view):''}
Cite ${esc(benchmark().name)}
${esc(citation)}
`);
$('#copy-citation').addEventListener('click',e=>copyCitation(citation,e.currentTarget));
$('#save-citation').addEventListener('click',()=>download(state.benchmark+'.bib',citation+'\n','text/plain;charset=utf-8'));
}
document.addEventListener('click',event=>{
const button=event.target.closest('button');if(!button)return;
if(button.dataset.benchmark)switchBenchmark(button.dataset.benchmark);
if(button.dataset.view)switchView(button.dataset.view);
if(button.dataset.language)switchLanguage(button.dataset.language);
if(button.dataset.dimension){state.dimension=button.dataset.dimension;state.subcategory='all';resetScoreColumns();}
if(button.dataset.group){state.group=button.dataset.group;renderTable();}
if(button.dataset.sort){state.direction=state.sort===button.dataset.sort&&state.direction==='desc'?'asc':'desc';state.sort=button.dataset.sort;renderTable();}
if(button.dataset.model)showModels([button.dataset.model]);
if(button.dataset.remove)toggleSelected(button.dataset.remove,false);
});
document.addEventListener('change',event=>{
const input=event.target;
if(input.dataset.select)toggleSelected(input.dataset.select,input.checked);
if(input.dataset.column){
if(!input.checked&&state.visible.length===1){input.checked=true;toast('Keep at least one score column visible.');return;}
state.visible=input.checked?[...state.visible,input.dataset.column]:state.visible.filter(c=>c!==input.dataset.column);
if(!state.visible.includes(state.sort)){state.sort=availableColumns().find(c=>state.visible.includes(c.key)).key;state.direction='desc';}
renderTable();
}
});
$('#model-search').addEventListener('input',e=>{state.query=e.target.value;renderTable();});
$('#input-filter').addEventListener('change',e=>{state.input=e.target.value;renderTable();});
$('#subcategory-select').addEventListener('change',e=>{state.subcategory=e.target.value;resetScoreColumns();});
$('#page-copy-citation').addEventListener('click',e=>copyCitation(CITATIONS[state.benchmark],e.currentTarget));
$('#page-save-citation').addEventListener('click',()=>download(state.benchmark+'.bib',CITATIONS[state.benchmark]+'\n','text/plain;charset=utf-8'));
$('#reset-filters').addEventListener('click',resetFilters);$('#empty-reset').addEventListener('click',resetFilters);
$('#download-csv').addEventListener('click',()=>exportCSV(filteredModels()));
$('#compare-open').addEventListener('click',()=>showModels(state.selected));
$('#clear-compare').addEventListener('click',()=>{state.selected=[];renderTable();});
$('.dialog-close').addEventListener('click',()=>$('#detail-dialog').close());
$('#detail-dialog').addEventListener('click',e=>{if(e.target===$('#detail-dialog')){const r=e.target.getBoundingClientRect();if(e.clientXr.right||e.clientYr.bottom)e.target.close();}});
$('#citation-button').addEventListener('click',showCitation);
document.addEventListener('click',e=>{if(!$('#column-menu').contains(e.target))$('#column-menu').open=false;});
document.addEventListener('keydown',e=>{
if(e.key==='/'&&!['INPUT','TEXTAREA','SELECT'].includes(document.activeElement.tagName)&&!$('#detail-dialog').open){e.preventDefault();$('#model-search').focus();}
const tab=e.target.closest('[data-benchmark], [data-view], [data-language]');
if(tab&&['ArrowLeft','ArrowRight','Home','End'].includes(e.key)){
e.preventDefault();const siblings=[...tab.parentElement.querySelectorAll('[data-benchmark], [data-view], [data-language]')].filter(el=>!el.hidden),i=siblings.indexOf(tab);
const n=e.key==='Home'?0:e.key==='End'?siblings.length-1:(i+(e.key==='ArrowRight'?1:-1)+siblings.length)%siblings.length;
siblings[n].click();siblings[n].focus();
}
});
switchBenchmark(state.benchmark);
})();