Docs: Update benchmark results

results.json

@@ -1,42 +1,46 @@
 {
   "openai/gpt-5.1-codex": {
-    "1_dijkstra": null,
+    "1_dijkstra": 15.241279893,
-    "2_convex_hull": null,
+    "2_convex_hull": 13.933983282,
-    "3_lis": 4.134811772999999,
+    "3_lis": 3.980165064000001,
-    "4_determinant": 3.128456531999999,
+    "4_determinant": 6.332550646999996,
-    "5_markdown_parser": 2.4388916820000004,
+    "5_markdown_parser": 3.2380716380000014,
-    "6_csv_processor": 43.478935838999995,
+    "6_csv_processor": 26.09038364200001,
-    "7_scheduler": 84.317196593,
+    "7_scheduler": 130.18603798900003,
-    "8_json_validator": 6.765057070999989
+    "8_json_validator": 8.38715121099999,
+    "9_stream_visualizer": 12.031535295999987
   },
   "openai/gpt-5.1-chat": {
-    "1_dijkstra": null,
+    "1_dijkstra": 2.65463601799999,
-    "2_convex_hull": 4.03333652099999,
+    "2_convex_hull": 3.1764218759999787,
-    "3_lis": 3.2548832980000006,
+    "3_lis": 2.5808361240000233,
-    "4_determinant": 3.480435989999998,
+    "4_determinant": 1.857067280999996,
-    "5_markdown_parser": 12.157845766999992,
+    "5_markdown_parser": 2.1500550130000047,
-    "6_csv_processor": null,
+    "6_csv_processor": 2.8134082990000024,
-    "7_scheduler": 3.6146848039999897,
+    "7_scheduler": 6.116577822999999,
-    "8_json_validator": 3.9886151490000192
+    "8_json_validator": 1.477971265,
+    "9_stream_visualizer": 1.7076946140000073
   },
   "google/gemini-2.5-pro": {
-    "1_dijkstra": 44.864160204000015,
+    "1_dijkstra": 33.989065073000006,
-    "2_convex_hull": 33.892027145,
+    "2_convex_hull": 56.566219838000016,
-    "3_lis": 36.906535906,
+    "3_lis": 52.14045403000002,
-    "4_determinant": 11.497210719000025,
+    "4_determinant": 14.913826469999972,
-    "5_markdown_parser": 13.62311761299998,
+    "5_markdown_parser": 26.86344819299999,
-    "6_csv_processor": 42.61058959099999,
+    "6_csv_processor": 49.397067434999975,
-    "7_scheduler": 12.12732526100002,
+    "7_scheduler": 52.55558026300004,
-    "8_json_validator": 50.54006011399999
+    "8_json_validator": 19.59284627099999,
+    "9_stream_visualizer": 36.41860035799991
   },
   "anthropic/claude-sonnet-4.5 TEMP:0.7": {
-    "1_dijkstra": 4.8069697199999935,
+    "1_dijkstra": 4.49742717099993,
-    "2_convex_hull": 4.475060508999974,
+    "2_convex_hull": 4.570548665000009,
-    "3_lis": 2.835424330000009,
+    "3_lis": 2.375867489000084,
-    "4_determinant": 2.3172859399999726,
+    "4_determinant": 1.5574152070001,
-    "5_markdown_parser": 2.080127800999966,
+    "5_markdown_parser": 1.8310976730000694,
-    "6_csv_processor": 5.218456623000034,
+    "6_csv_processor": 4.079893573999987,
-    "7_scheduler": 11.33699588100001,
+    "7_scheduler": 12.446870330999955,
-    "8_json_validator": 3.298284380999976
+    "8_json_validator": 2.8953664760000537,
+    "9_stream_visualizer": 5.022439357000054
   }
 }

tests/1_dijkstra/outputs/anthropic_claude-sonnet-4.5 TEMP_0.7.js

@@ -1,15 +1,15 @@
 async function findShortestPath(graph, start, end) {
-  const { default: PQ } = await import('https://cdn.jsdelivr.net/npm/js-priority-queue@0.1.5/+esm');
+  const { default: PriorityQueue } = await import('https://cdn.jsdelivr.net/npm/js-priority-queue@0.1.5/+esm');
   
   const distances = Object.keys(graph).reduce((acc, node) => ({ ...acc, [node]: Infinity }), {});
   distances[start] = 0;
   
-  const pq = new PQ({ comparator: (a, b) => a.dist - b.dist });
+  const pq = new PriorityQueue({ comparator: (a, b) => a.dist - b.dist });
   pq.queue({ node: start, dist: 0 });
   
   const visited = new Set();
   
-  while (pq.length) {
+  while (pq.length > 0) {
     const { node, dist } = pq.dequeue();
     
     if (visited.has(node)) continue;

tests/1_dijkstra/outputs/google_gemini-2.5-pro.js

@@ -1,33 +1,36 @@
-async function findShortestPath(graph, start, end) {
+const findShortestPath = async (graph, startNode, endNode) => {
   const { default: PriorityQueue } = await import('https://cdn.jsdelivr.net/npm/js-priority-queue@0.1.5/priority-queue.min.js');
 
   const distances = Object.fromEntries(
-    Object.keys(graph).map(node => [node, Infinity])
+    Object.keys(graph).map(node => [node, 1 / 0])
   );
-  distances[start] = 0;
+  distances[startNode] = 0;
 
   const pq = new PriorityQueue({
-    comparator: (a, b) => distances[b] - distances[a]
+    comparator: (a, b) => a.priority - b.priority
   });
-  pq.queue(start);
+  pq.queue({ node: startNode, priority: 0 });
 
   while (pq.length) {
-    const currentNode = pq.dequeue();
+    const { node: u, priority: dist } = pq.dequeue();
 
-    if (currentNode === end) break;
+    if (dist > distances[u]) {
+      continue;
+    }
 
-    if (!graph[currentNode]) continue;
+    if (u === endNode) {
+      break;
+    }
 
-    for (const [neighbor, weight] of Object.entries(graph[currentNode])) {
+    for (const [v, weight] of Object.entries(graph[u] ?? {})) {
-      const newDist = distances[currentNode] + weight;
+      const newDist = distances[u] + weight;
-
+      if (newDist < distances[v]) {
-      if (newDist < distances[neighbor]) {
+        distances[v] = newDist;
-        distances[neighbor] = newDist;
+        pq.queue({ node: v, priority: newDist });
-        pq.queue(neighbor);
       }
     }
   }
 
-  return distances[end];
+  return distances[endNode];
-}
+};
 export default findShortestPath;

tests/1_dijkstra/outputs/openai_gpt-5.1-chat.js

@@ -0,0 +1,23 @@
+async function findShortestPath(g,s,e){
+  const {default:Q}=await import('https://cdn.jsdelivr.net/npm/js-priority-queue@latest/dist/priority-queue.min.js')
+  const d={},v={}
+  for(const k in g)d[k]=Infinity
+  d[s]=0
+  const q=new Q({comparator:(a,b)=>a[1]-b[1]})
+  q.queue([s,0])
+  while(q.length){
+    const [n,w]=q.dequeue()
+    if(v[n])continue
+    v[n]=1
+    if(n===e)return w
+    for(const k in g[n]){
+      const nw=w+g[n][k]
+      if(nw<d[k]){
+        d[k]=nw
+        q.queue([k,nw])
+      }
+    }
+  }
+  return Infinity
+}
+export default findShortestPath;

tests/1_dijkstra/outputs/openai_gpt-5.1-codex.js

@@ -0,0 +1,23 @@
+const findShortestPath=async(g,s,t)=>{
+  if(s===t)return 0
+  const{default:PriorityQueue}=await import('https://cdn.skypack.dev/js-priority-queue')
+  if(!g[s]||!g[t])return Infinity
+  const seen=new Map([[s,0]])
+  const pq=new PriorityQueue({comparator:(a,b)=>a[1]-b[1]})
+  pq.queue([s,0])
+  while(pq.length){
+    const[n,d]=pq.dequeue()
+    if(d>(seen.get(n)??Infinity))continue
+    if(n===t)return d
+    const nbrs=g[n]||{}
+    for(const k in nbrs){
+      const nd=d+nbrs[k]
+      if(nd<(seen.get(k)??Infinity)){
+        seen.set(k,nd)
+        pq.queue([k,nd])
+      }
+    }
+  }
+  return Infinity
+}
+export default findShortestPath;

tests/2_convex_hull/outputs/anthropic_claude-sonnet-4.5 TEMP_0.7.js

@@ -21,6 +21,9 @@ async function findConvexHull(points) {
     return hull;
   };
   
-  return [...buildHull(sorted), ...buildHull([...sorted].reverse())];
+  const lower = buildHull(sorted);
+  const upper = buildHull([...sorted].reverse());
+  
+  return [...lower, ...upper];
 }
 export default findConvexHull;

tests/2_convex_hull/outputs/google_gemini-2.5-pro.js

@@ -1,28 +1,29 @@
 const findConvexHull = async (points) => {
-  const _ = await import('https://cdn.jsdelivr.net/npm/lodash-es@4.17.21/lodash.min.js');
+  const { sortBy, uniqWith, isEqual } =
+    await import('https://cdn.jsdelivr.net/npm/lodash-es@4.17.21/lodash.js');
 
-  const sortedPoints = _.sortBy(_.uniqWith(points, _.isEqual), ['x', 'y']);
+  const cross = (o, a, b) => (a.x - o.x) * (b.y - o.y) - (a.y - o.y) * (b.x - o.x);
 
-  if (sortedPoints.length < 3) {
+  const pts = sortBy(uniqWith(points, isEqual), ['x', 'y']);
-    return sortedPoints;
+
+  if (pts.length < 3) return pts;
+
+  const lower = [];
+  for (const p of pts) {
+    while (lower.length >= 2 && cross(lower.at(-2), lower.at(-1), p) <= 0) {
+      lower.pop();
+    }
+    lower.push(p);
   }
 
-  const crossProduct = (p1, p2, p3) =>
+  const upper = [];
-    (p2.x - p1.x) * (p3.y - p1.y) - (p2.y - p1.y) * (p3.x - p1.x);
+  for (let i = pts.length - 1; i >= 0; i--) {
-
+    const p = pts[i];
-  const buildHalfHull = (pointSet) => {
+    while (upper.length >= 2 && cross(upper.at(-2), upper.at(-1), p) <= 0) {
-    const hull = [];
+      upper.pop();
-    for (const p of pointSet) {
-      while (hull.length >= 2 && crossProduct(hull.at(-2), hull.at(-1), p) <= 0) {
-        hull.pop();
     }
-      hull.push(p);
+    upper.push(p);
   }
-    return hull;
-  };
-
-  const lower = buildHalfHull(sortedPoints);
-  const upper = buildHalfHull([...sortedPoints].reverse());
 
   return [...lower.slice(0, -1), ...upper.slice(0, -1)];
 };

tests/2_convex_hull/outputs/openai_gpt-5.1-chat.js

@@ -1,20 +1,20 @@
-export async function findConvexHull(a){
+export async function findConvexHull(p){
   const _=(await import('https://cdn.jsdelivr.net/npm/lodash-es@4.17.21/lodash.min.js')).default
-  a = _.uniqWith(a,_.isEqual)
+  p=_.uniqWith(p,_.isEqual)
-  a = _.sortBy(a,['x','y'])
+  p=_.sortBy(p,['x','y'])
-  const c=(o,p,q)=>((p.x-o.x)*(q.y-o.y)-(p.y-o.y)*(q.x-o.x))
+  const c=(a,b,d)=>(b.x-a.x)*(d.y-a.y)-(b.y-a.y)*(d.x-a.x)
-  const b=[]
+  const h=[]
-  for(const p of a){
+  for(const v of p){
-    while(b.length>1 && c(b[b.length-2],b[b.length-1],p)<=0) b.pop()
+    while(h.length>1&&c(h[h.length-2],h[h.length-1],v)<=0)h.pop()
-    b.push(p)
+    h.push(v)
   }
   const u=[]
-  for(const p of [...a].reverse()){
+  for(const v of [...p].reverse()){
-    while(u.length>1 && c(u[u.length-2],u[u.length-1],p)<=0) u.pop()
+    while(u.length>1&&c(u[u.length-2],u[u.length-1],v)<=0)u.pop()
-    u.push(p)
+    u.push(v)
   }
+  h.pop()
   u.pop()
-  b.pop()
+  return [...h,...u]
-  return [...b,...u]
 }
 export default findConvexHull;

tests/2_convex_hull/outputs/openai_gpt-5.1-codex.js

@@ -0,0 +1,18 @@
+async function findConvexHull(points) {
+  const {sortBy, uniqWith} = await import('https://cdn.jsdelivr.net/npm/lodash-es@4.17.21/lodash.js')
+  const ps = uniqWith(sortBy(points.map(({x, y}) => ({x: +x, y: +y})), ['x', 'y']), (a, b) => a.x === b.x && a.y === b.y)
+  if (ps.length < 2) return ps
+  const c = (o, a, b) => (a.x - o.x) * (b.y - o.y) - (a.y - o.y) * (b.x - o.x)
+  const lower = []
+  for (const p of ps) {
+    while (lower.length > 1 && c(lower.at(-2), lower.at(-1), p) <= 0) lower.pop()
+    lower.push(p)
+  }
+  const upper = []
+  for (const p of [...ps].reverse()) {
+    while (upper.length > 1 && c(upper.at(-2), upper.at(-1), p) <= 0) upper.pop()
+    upper.push(p)
+  }
+  return lower.slice(0, -1).concat(upper.slice(0, -1))
+}
+export default findConvexHull;

tests/3_lis/outputs/anthropic_claude-sonnet-4.5 TEMP_0.7.js

@@ -1,13 +1,12 @@
 const findLISLength = async (arr) => {
   if (!arr?.length) return 0;
   
-  const { bisectLeft } = await import('https://cdn.jsdelivr.net/npm/d3-array@3/+esm');
+  const { bisectLeft } = await import('https://cdn.skypack.dev/d3-array');
-  
   const tails = [];
   
   for (const num of arr) {
-    const pos = bisectLeft(tails, num);
+    const idx = bisectLeft(tails, num);
-    tails[pos] = num;
+    tails[idx] = num;
   }
   
   return tails.length;

tests/3_lis/outputs/google_gemini-2.5-pro.js

@@ -1,12 +1,16 @@
-async function findLISLength(nums) {
+const findLISLength = async (nums) => {
-  const { bisectLeft } = await import('https://cdn.jsdelivr.net/npm/d3-array@3/+esm');
+  if (!nums?.length) {
+    return 0;
+  }
+
+  const { bisectLeft } = await import('https://esm.sh/d3-array');
 
   const tails = [];
 
-  for (const num of nums || []) {
+  for (const num of nums) {
     tails[bisectLeft(tails, num)] = num;
   }
 
   return tails.length;
-}
+};
 export default findLISLength;

tests/3_lis/outputs/openai_gpt-5.1-chat.js

@@ -1,10 +1,10 @@
 async function findLISLength(a) {
-  const {bisectLeft:b} = await import('https://cdn.jsdelivr.net/npm/d3-array@3/+esm')
+  const {bisectLeft} = await import("https://cdn.jsdelivr.net/npm/d3-array@3/+esm");
-  const t = []
+  const t = [];
   for (const x of a) {
-    const i = b(t, x)
+    const i = bisectLeft(t, x);
-    t[i] = x
+    t[i] = x;
   }
-  return t.length
+  return t.length;
 }
 export default findLISLength;

tests/3_lis/outputs/openai_gpt-5.1-codex.js

@@ -1,10 +1,7 @@
-async function findLISLength(nums){
+async function findLISLength(a){
-  const {bisectLeft}=await import('https://cdn.jsdelivr.net/npm/d3-array@3/+esm')
+  const {bisectLeft:b}=await import('https://cdn.jsdelivr.net/npm/d3-array@3/+esm')
-  const tails=[]
+  const t=[]
-  for(const n of nums){
+  for(const n of a)t[b(t,n)]=n
-    const i=bisectLeft(tails,n)
+  return t.length
-    tails[i]=n
-  }
-  return tails.length
 }
 export default findLISLength;

tests/4_determinant/outputs/anthropic_claude-sonnet-4.5 TEMP_0.7.js

@@ -1,5 +1,5 @@
-async function calculateDeterminant(matrix) {
+const calculateDeterminant = async (matrix) => {
   const { det } = await import('https://cdn.jsdelivr.net/npm/mathjs@11.11.0/+esm');
   return det(matrix);
-}
+};
 export default calculateDeterminant;

tests/4_determinant/outputs/google_gemini-2.5-pro.js

@@ -1,5 +1,5 @@
-const calculateDeterminant = async matrix => {
+const calculateDeterminant = async (matrix) => {
-  const { det } = await import('https://cdn.jsdelivr.net/npm/mathjs@12.4.2/es/index.js');
+  const { det } = await import('https://cdn.jsdelivr.net/npm/mathjs@13.0.0/+esm');
   return det(matrix);
 };
 export default calculateDeterminant;

tests/4_determinant/outputs/openai_gpt-5.1-chat.js

@@ -1,5 +1,5 @@
 async function calculateDeterminant(m){
-  const {det}=await import('https://cdn.jsdelivr.net/npm/mathjs@11.11.0/+esm')
+  const {det}=await import('https://cdn.jsdelivr.net/npm/mathjs@14.0.0/+esm')
   return det(m)
 }
 export default calculateDeterminant;

tests/4_determinant/outputs/openai_gpt-5.1-codex.js

@@ -1,5 +1,5 @@
-async function calculateDeterminant(m){
+const calculateDeterminant=async m=>{
-  const {det}=await import('https://cdn.jsdelivr.net/npm/mathjs@11.11.0/+esm')
+  const {det}=await import('https://cdn.skypack.dev/mathjs')
   return det(m)
 }
 export default calculateDeterminant;

tests/5_markdown_parser/outputs/anthropic_claude-sonnet-4.5 TEMP_0.7.js

@@ -1,5 +1,5 @@
-async function parseMarkdown(md) {
+const parseMarkdown = async (md) => {
-  const { marked } = await import('https://cdn.jsdelivr.net/npm/marked@11/+esm');
+  const { marked } = await import('https://cdn.jsdelivr.net/npm/marked@11.1.1/+esm');
   return marked.parse(md);
-}
+};
 export default parseMarkdown;

tests/5_markdown_parser/outputs/google_gemini-2.5-pro.js

@@ -1,6 +1,6 @@
-const parseMarkdown = async (md) => {
+const parseMarkdown = async (markdown) => {
-  const cdn = 'https://cdn.jsdelivr.net/npm/marked@13.0.0/lib/marked.esm.js';
+  const cdnUrl = 'https://cdn.jsdelivr.net/npm/marked@13.0.0/lib/marked.esm.js';
-  const { marked } = await import(cdn);
+  const { marked } = await import(cdnUrl);
-  return marked.parse(md);
+  return marked.parse(markdown ?? '');
 };
 export default parseMarkdown;

tests/5_markdown_parser/outputs/openai_gpt-5.1-chat.js

@@ -1,2 +1,5 @@
-
+export async function parseMarkdown(m){
+  const {marked}=await import('https://cdn.jsdelivr.net/npm/marked/lib/marked.esm.js')
+  return marked.parse(m)
+}
 export default parseMarkdown;

tests/5_markdown_parser/outputs/openai_gpt-5.1-codex.js

@@ -1,5 +1,5 @@
-const parseMarkdown = async m => {
+const parseMarkdown=async(md='')=>{
-  const { marked } = await import('https://cdn.jsdelivr.net/npm/marked/lib/marked.esm.js')
+  const{marked}=await import('https://cdn.jsdelivr.net/npm/marked/lib/marked.esm.js');
-  return marked.parse(m)
+  return marked.parse(md);
-}
+};
 export default parseMarkdown;

tests/6_csv_processor/outputs/anthropic_claude-sonnet-4.5 TEMP_0.7.js

@@ -1,32 +1,24 @@
-async function processCSV(csvString, config) {
+const processCSV = async (csv, { filterColumn, filterValue, groupBy, aggregateColumn, operation }) => {
-  const { parse } = await import('https://cdn.skypack.dev/papaparse@5.4.1');
+  const { parse } = await import('https://cdn.skypack.dev/csv-parse/sync');
   
-  const { data } = parse(csvString, { header: true, skipEmptyLines: true });
+  const records = parse(csv, { columns: true, skip_empty_lines: true });
   
-  const filtered = data.filter(row => 
+  const filtered = records.filter(r => r[filterColumn] == filterValue);
-    String(row[config.filterColumn]) === String(config.filterValue)
-  );
   
-  const grouped = filtered.reduce((acc, row) => {
+  const groups = filtered.reduce((acc, row) => {
-    const key = row[config.groupBy];
+    const key = row[groupBy];
-    if (!acc[key]) acc[key] = [];
+    (acc[key] ??= []).push(row);
-    acc[key].push(row);
     return acc;
   }, {});
   
-  return Object.entries(grouped).map(([key, rows]) => {
+  return Object.entries(groups).map(([key, rows]) => {
-    let result;
+    const vals = rows.map(r => Number(r[aggregateColumn]) || 0);
-    const vals = rows.map(r => parseFloat(r[config.aggregateColumn]) || 0);
+    const sum = vals.reduce((a, b) => a + b, 0);
     
-    if (config.operation === 'sum') {
+    return {
-      result = vals.reduce((a, b) => a + b, 0);
+      [groupBy]: key,
-    } else if (config.operation === 'avg') {
+      result: operation === 'sum' ? sum : operation === 'avg' ? sum / vals.length : vals.length
-      result = vals.reduce((a, b) => a + b, 0) / vals.length;
+    };
-    } else {
-      result = rows.length;
-    }
-    
-    return { [config.groupBy]: key, result };
   });
-}
+};
 export default processCSV;

tests/6_csv_processor/outputs/google_gemini-2.5-pro.js

@@ -1,44 +1,39 @@
-async function processCSV(csvString, config) {
+async function processCSV(csv, {
-  const {
   filterColumn,
   filterValue,
   groupBy,
   aggregateColumn,
   operation,
-  } = config;
+}) {
+  const { default: Papa } = await import(
+    'https://cdn.jsdelivr.net/npm/papaparse@5.4.1/papaparse.min.js'
+  );
 
-  const { default: Papa } = await import('https://cdnjs.cloudflare.com/ajax/libs/PapaParse/5.4.1/papaparse.min.js');
+  const { data } = Papa.parse(csv, { header: true, skipEmptyLines: true });
 
-  const { data } = Papa.parse(csvString, {
+  const groups = data.reduce((acc, row) => {
-    header: true,
+    if (row[filterColumn] != filterValue) {
-    dynamicTyping: true,
+      return acc;
-    skipEmptyLines: true,
+    }
-  });
+
+    const key = row[groupBy];
+    const stats = acc.get(key) || { sum: 0, count: 0 };
+
+    stats.sum += Number(row[aggregateColumn]) || 0;
+    stats.count++;
+
+    return acc.set(key, stats);
+  }, new Map());
 
   const aggregators = {
-    sum: (acc, val) => (acc || 0) + val,
+    sum: ({ sum }) => sum,
-    count: (acc, _val) => (acc || 0) + 1,
+    avg: ({ sum, count }) => (count ? sum / count : 0),
-    avg: (acc, val) => {
+    count: ({ count }) => count,
-      const state = acc || { sum: 0, count: 0 };
-      state.sum += val;
-      state.count += 1;
-      return state;
-    },
   };
 
-  const groups = data
+  return Array.from(groups, ([key, stats]) => ({
-    .filter(row => row[filterColumn] == filterValue)
-    .reduce((acc, row) => {
-      const key = row[groupBy];
-      acc[key] = aggregators[operation](acc[key], row[aggregateColumn]);
-      return acc;
-    }, {});
-
-  return Object.entries(groups).map(([key, value]) => ({
     [groupBy]: key,
-    result: operation === 'avg'
+    result: aggregators[operation](stats),
-      ? (value.count ? value.sum / value.count : 0)
-      : value,
   }));
 }
 export default processCSV;

tests/6_csv_processor/outputs/openai_gpt-5.1-chat.js

@@ -0,0 +1,22 @@
+async function processCSV(csv, cfg) {
+  const p = await import("https://cdn.jsdelivr.net/npm/papaparse@5.4.1/papaparse.min.js");
+  const f = p.parse(csv, { header: true }).data.filter(r => r[cfg.filterColumn] == cfg.filterValue);
+  const g = {};
+  for (const r of f) {
+    const k = r[cfg.groupBy];
+    if (!g[k]) g[k] = [];
+    g[k].push(r);
+  }
+  const out = [];
+  for (const k in g) {
+    const rows = g[k];
+    let nums = rows.map(r => +r[cfg.aggregateColumn] || 0);
+    let res = 0;
+    if (cfg.operation === "sum") res = nums.reduce((a, b) => a + b, 0);
+    else if (cfg.operation === "avg") res = nums.reduce((a, b) => a + b, 0) / nums.length;
+    else if (cfg.operation === "count") res = rows.length;
+    out.push({ [cfg.groupBy]: k, result: res });
+  }
+  return out;
+}
+export default processCSV;

tests/6_csv_processor/outputs/openai_gpt-5.1-codex.js

@@ -1,21 +1,16 @@
-const libs=(()=>{let c;return()=>c||(c=Promise.all([
+async function processCSV(csv,cfg){
-import('https://cdn.jsdelivr.net/npm/d3-dsv@3/+esm'),
+ const [{Papa},{groupBy:gb}]=await Promise.all([
-import('https://cdn.jsdelivr.net/npm/lodash-es@4.17.21/groupBy.js')
+  import('https://cdn.jsdelivr.net/npm/papaparse@5.4.1/papaparse.min.mjs'),
-]).then(([dsv,gb])=>({parse:dsv.csvParse,groupBy:gb.default||gb})));})();
+  import('https://cdn.jsdelivr.net/npm/lodash-es@4.17.21/lodash.js')
-
+ ])
-const toNum=v=>{const n=+v;return Number.isFinite(n)?n:0};
+ const {filterColumn:fc,filterValue:fv,groupBy:gbk,aggregateColumn:ac,operation:op}=cfg
-
+ const rows=Papa.parse(csv,{header:!0,skipEmptyLines:!0}).data.filter(r=>r&&r[fc]==fv)
-async function processCSV(csv,opts){
+ const grouped=gb(rows,r=>r?.[gbk]??'')
-  const {parse,groupBy}=await libs();
+ return Object.entries(grouped).map(([k,arr])=>{
-  const rows=parse(csv);
+  const nums=arr.map(r=>{const n=+r[ac];return Number.isFinite(n)?n:0})
-  const {filterColumn:fc,filterValue:fv,groupBy:gc,aggregateColumn:ac,operation:op}=opts;
+  const sum=nums.reduce((a,b)=>a+b,0)
-  const pool=groupBy(rows.filter(r=>r[fc]==fv),r=>r[gc]??'');
+  const result=op==='sum'?sum:op==='avg'?sum/nums.length:arr.length
-  const calc={
+  return {[gbk]:k,result}
-    sum:x=>x.reduce((t,r)=>t+toNum(r[ac]),0),
+ })
-    avg:x=>x.length?x.reduce((t,r)=>t+toNum(r[ac]),0)/x.length:0,
-    count:x=>x.length
-  }[op];
-  if(!calc) throw new Error('Unsupported operation');
-  return Object.entries(pool).map(([k,x])=>({[gc]:k,result:calc(x)}));
 }
 export default processCSV;

tests/7_scheduler/outputs/anthropic_claude-sonnet-4.5 TEMP_0.7.js

@@ -1,71 +1,81 @@
-const findAvailableSlots = async (cal1, cal2, constraints) => {
+async function findAvailableSlots(cal1, cal2, constraints) {
-  const { default: dayjs } = await import('https://cdn.jsdelivr.net/npm/dayjs@1.11.10/+esm');
+  const { parse, format, addMinutes, isWithinInterval, isBefore, isAfter, startOfDay, setHours, setMinutes } = await import('https://cdn.jsdelivr.net/npm/date-fns@3.0.0/+esm');
-  const [{ default: utc }, { default: isBetween }] = await Promise.all([
-    import('https://cdn.jsdelivr.net/npm/dayjs@1.11.10/plugin/utc.js/+esm'),
-    import('https://cdn.jsdelivr.net/npm/dayjs@1.11.10/plugin/isBetween.js/+esm')
-  ]);
-  
-  dayjs.extend(utc);
-  dayjs.extend(isBetween);
   
   const { durationMinutes, searchRange, workHours } = constraints;
-  const [whStart, whEnd] = [workHours.start.split(':'), workHours.end.split(':')];
+  const duration = durationMinutes;
   
-  const merged = [...cal1, ...cal2]
+  const parseTime = (dateStr) => parse(dateStr, "yyyy-MM-dd'T'HH:mm:ss.SSSxxx", new Date());
-    .map(s => ({ start: dayjs(s.start), end: dayjs(s.end) }))
+  const toISO = (date) => format(date, "yyyy-MM-dd'T'HH:mm:ss.SSSxxx");
-    .sort((a, b) => a.start - b.start);
   
-  const busy = merged.reduce((acc, curr) => {
+  const rangeStart = parseTime(searchRange.start);
-    if (!acc.length || curr.start > acc[acc.length - 1].end) {
+  const rangeEnd = parseTime(searchRange.end);
-      acc.push(curr);
+  
+  const [whStart, whEnd] = workHours.start.split(':').map(Number);
+  const [whEndH, whEndM] = workHours.end.split(':').map(Number);
+  
+  const allBusy = [...cal1, ...cal2].map(slot => ({
+    start: parseTime(slot.start),
+    end: parseTime(slot.end)
+  })).sort((a, b) => a.start - b.start);
+  
+  const merged = [];
+  for (const slot of allBusy) {
+    if (!merged.length || isBefore(merged[merged.length - 1].end, slot.start)) {
+      merged.push({ ...slot });
     } else {
-      acc[acc.length - 1].end = dayjs.max(acc[acc.length - 1].end, curr.end);
+      merged[merged.length - 1].end = isAfter(slot.end, merged[merged.length - 1].end) 
+        ? slot.end 
+        : merged[merged.length - 1].end;
     }
-    return acc;
+  }
-  }, []);
+  
+  const freePeriods = [];
+  let current = rangeStart;
+  
+  for (const busy of merged) {
+    if (isBefore(current, busy.start)) {
+      freePeriods.push({ start: current, end: busy.start });
+    }
+    current = isAfter(busy.end, current) ? busy.end : current;
+  }
+  
+  if (isBefore(current, rangeEnd)) {
+    freePeriods.push({ start: current, end: rangeEnd });
+  }
+  
+  const isInWorkHours = (date) => {
+    const day = startOfDay(date);
+    const workStart = setMinutes(setHours(day, whStart), whEnd > 0 ? 0 : 0);
+    const workEnd = setMinutes(setHours(day, whEndH), whEndM);
+    return isWithinInterval(date, { start: workStart, end: workEnd });
+  };
   
   const slots = [];
-  let current = dayjs(searchRange.start);
-  const rangeEnd = dayjs(searchRange.end);
   
-  while (current < rangeEnd) {
+  for (const period of freePeriods) {
-    const dayStart = current.hour(+whStart[0]).minute(+whStart[1]).second(0);
+    let slotStart = period.start;
-    const dayEnd = current.hour(+whEnd[0]).minute(+whEnd[1]).second(0);
     
-    let slotStart = dayStart < current ? current : dayStart;
+    while (true) {
+      const slotEnd = addMinutes(slotStart, duration);
       
-    for (const b of busy) {
+      if (isAfter(slotEnd, period.end)) break;
-      if (b.end <= slotStart || b.start >= dayEnd) continue;
       
-      if (b.start > slotStart) {
+      if (isInWorkHours(slotStart) && isInWorkHours(slotEnd)) {
-        const slotEnd = dayjs.min(b.start, dayEnd);
+        const day1 = startOfDay(slotStart);
-        let probe = slotStart;
+        const day2 = startOfDay(slotEnd);
         
-        while (probe.add(durationMinutes, 'minute') <= slotEnd) {
+        if (day1.getTime() === day2.getTime()) {
           slots.push({
-            start: probe.toISOString(),
+            start: toISO(slotStart),
-            end: probe.add(durationMinutes, 'minute').toISOString()
+            end: toISO(slotEnd)
           });
-          probe = probe.add(durationMinutes, 'minute');
-        }
-      }
-      slotStart = dayjs.max(slotStart, b.end);
-    }
-    
-    if (slotStart < dayEnd) {
-      let probe = slotStart;
-      while (probe.add(durationMinutes, 'minute') <= dayEnd) {
-        slots.push({
-          start: probe.toISOString(),
-          end: probe.add(durationMinutes, 'minute').toISOString()
-        });
-        probe = probe.add(durationMinutes, 'minute');
         }
       }
       
-    current = current.add(1, 'day').startOf('day');
+      slotStart = slotEnd;
+    }
   }
   
   return slots;
-};
+}
 export default findAvailableSlots;

tests/7_scheduler/outputs/google_gemini-2.5-pro.js

@@ -1,2 +1,72 @@
+const findAvailableSlots = async (calendar1, calendar2, constraints) => {
+  const [{ default: dayjs }, { default: utc }] = await Promise.all([
+    import('https://cdn.skypack.dev/dayjs'),
+    import('https://cdn.skypack.dev/dayjs/plugin/utc.js'),
+  ]);
+  dayjs.extend(utc);
 
+  const { durationMinutes, searchRange, workHours } = constraints;
+  
+  const toDayjs = (t) => dayjs.utc(t);
+  const toDayjsRange = ({ start, end }) => ({ start: toDayjs(start), end: toDayjs(end) });
+
+  const search = toDayjsRange(searchRange);
+  const allBusy = [...calendar1, ...calendar2]
+    .map(toDayjsRange)
+    .sort((a, b) => a.start.valueOf() - b.start.valueOf());
+
+  const mergedBusy = allBusy.reduce((acc, current) => {
+    const last = acc[acc.length - 1];
+    if (last && current.start.valueOf() < last.end.valueOf()) {
+      if (current.end.isAfter(last.end)) {
+        last.end = current.end;
+      }
+    } else {
+      acc.push({ ...current });
+    }
+    return acc;
+  }, []);
+
+  const boundaryPoints = [
+    search.start,
+    ...mergedBusy.flatMap(b => [b.start, b.end]),
+    search.end,
+  ];
+
+  const freeGaps = [];
+  for (let i = 0; i < boundaryPoints.length - 1; i += 2) {
+    const start = boundaryPoints[i];
+    const end = boundaryPoints[i + 1];
+    if (end.isAfter(start)) {
+      freeGaps.push({ start, end });
+    }
+  }
+
+  const availableSlots = [];
+  const [workStartH, workStartM] = workHours.start.split(':').map(Number);
+  const [workEndH, workEndM] = workHours.end.split(':').map(Number);
+
+  for (const gap of freeGaps) {
+    let cursor = gap.start.startOf('day');
+    while (cursor.isBefore(gap.end)) {
+      const workWindowStart = cursor.hour(workStartH).minute(workStartM);
+      const workWindowEnd = cursor.hour(workEndH).minute(workEndM);
+
+      let effectiveStart = dayjs.max(gap.start, workWindowStart);
+      const effectiveEnd = dayjs.min(gap.end, workWindowEnd);
+
+      while (effectiveStart.add(durationMinutes, 'minute').valueOf() <= effectiveEnd.valueOf()) {
+        const slotEnd = effectiveStart.add(durationMinutes, 'minute');
+        availableSlots.push({ start: effectiveStart, end: slotEnd });
+        effectiveStart = slotEnd;
+      }
+      cursor = cursor.add(1, 'day');
+    }
+  }
+
+  return availableSlots.map(({ start, end }) => ({
+    start: start.toISOString(),
+    end: end.toISOString(),
+  }));
+};
 export default findAvailableSlots;

tests/7_scheduler/outputs/openai_gpt-5.1-chat.js

@@ -1,48 +1,36 @@
 async function findAvailableSlots(a,b,c){
-  const d=await import('https://esm.sh/date-fns@3.6.0')
+  const f=await import('https://esm.sh/date-fns')
-  const {parseISO,formatISO,max,min,addMinutes,isBefore}=d
+  const p=f.parseISO,d=f.addMinutes,x=f.max,y=f.min,i=f.isBefore,j=f.isAfter
-  const p=t=>parseISO(t)
+  const {durationMinutes:r,searchRange:s,workHours:w}=c
-  const wh=(dt,s)=>{
+  const h=t=>t.split(':').map(n=>+n)
-    const [H,M]=s.split(':')
+  const [ws,we]=[h(w.start),h(w.end)]
-    const x=new Date(dt)
+  const z=(d0,[H,M])=>{d0=new Date(d0);d0.setHours(H,M,0,0);return d0.toISOString()}
-    x.setHours(H,M,0,0)
+  const S=p(s.start),E=p(s.end)
-    return x
+  let u=[...a,...b].map(o=>({start:p(o.start),end:p(o.end)}))
+  u=u.sort((A,B)=>A.start-B.start)
+  let m=[]
+  for(let k of u){
+    if(!m.length||k.start>m[m.length-1].end)m.push({start:k.start,end:k.end})
+    else m[m.length-1].end=new Date(Math.max(m[m.length-1].end,k.end))
   }
-  const rng=[p(c.searchRange.start),p(c.searchRange.end)]
+  let q=[]
-  const dur=c.durationMinutes
+  let st=S
-  const norm=x=>x.map(v=>[p(v.start),p(v.end)])
+  for(let k of m){
-  const A=norm(a),B=norm(b)
+    if(i(st,k.start)){
-  const merge=x=>{
+      let fs=x(st,S),fe=y(k.start,E)
-    let r=x.sort((x,y)=>x[0]-y[0]),o=[r[0]]
+      if(i(fs,fe))q.push({start:fs,end:fe})
-    for(let i=1;i<r.length;i++){
-      let [s,e]=r[i],l=o[o.length-1]
-      if(s<=l[1]) l[1]=new Date(Math.max(e,l[1]))
-      else o.push([s,e])
     }
-    return o
+    st=k.end
   }
-  const busy=merge(A.concat(B))
+  if(i(st,E))q.push({start:x(st,S),end:E})
-  const free=[]
+  let out=[]
-  let cur=rng[0]
+  for(let v of q){
-  for(let i=0;i<=busy.length;i++){
+    let cs=z(v.start,ws),ce=z(v.start,we)
-    let s=i<busy.length?max(cur,busy[i][1]):cur
+    cs=p(cs);ce=p(ce)
-    let e=i<busy.length?max(cur,busy[i][0]):rng[1]
+    let ss=x(v.start,cs),ee=y(v.end,ce)
-    if(i<busy.length){
+    for(let t=ss;i(d(t,r),ee)||+d(t,r)==+ee;t=d(t,r)){
-      if(busy[i][0]>cur) free.push([cur,busy[i][0]])
+      let e=d(t,r)
-      cur=max(cur,busy[i][1])
+      if(!j(t,ss)||i(e,ee))out.push({start:t.toISOString(),end:e.toISOString()})
-    }else free.push([s,e])
-  }
-  const out=[]
-  for(let [s,e]of free){
-    s=max(s,rng[0])
-    e=min(e,rng[1])
-    let ws=wh(s,c.workHours.start)
-    let we=wh(s,c.workHours.end)
-    if(ws>we){let t=ws;ws=we;we=t}
-    let ss=max(s,ws),ee=min(e,we)
-    while(isBefore(addMinutes(ss,dur),ee)){
-      out.push({start:formatISO(ss),end:formatISO(addMinutes(ss,dur))})
-      ss=addMinutes(ss,dur)
     }
   }
   return out

tests/7_scheduler/outputs/openai_gpt-5.1-codex.js

@@ -1,55 +1,52 @@
-const luxon=import('https://cdn.skypack.dev/luxon');
+let luxon$
 
-const findAvailableSlots=async(c1,c2,k)=>{
+const findAvailableSlots = async (calA, calB, cfg) => {
-  const {DateTime}=await luxon;
+  const {DateTime, Interval} = await (luxon$ ||= import('https://cdn.skypack.dev/luxon'))
-  const {durationMinutes:d,searchRange:r,workHours:w}=k;
+  const {durationMinutes: d, searchRange: r, workHours: w} = cfg
-  const zone=DateTime.fromISO(r.start).zoneName;
+  const s = DateTime.fromISO(r.start)
-  const iso=v=>DateTime.fromISO(v,{zone});
+  const e = DateTime.fromISO(r.end)
-  const rangeStart=iso(r.start);
+  const range = Interval.fromDateTimes(s, e)
-  const rangeEnd=iso(r.end);
+  const [sh, sm] = w.start.split(':').map(Number)
-  if(rangeEnd<=rangeStart)return[];
+  const [eh, em] = w.end.split(':').map(Number)
-  const [hs,ms]=w.start.split(':').map(Number);
+  const busy = [...calA, ...calB]
-  const [he,me]=w.end.split(':').map(Number);
+    .map(({start, end}) => ({start: DateTime.fromISO(start), end: DateTime.fromISO(end)}))
-  const daysEnd=rangeEnd.startOf('day');
+    .filter(v => v.end > s && v.start < e)
-  const windows=[];
+    .map(v => ({start: v.start < s ? s : v.start, end: v.end > e ? e : v.end}))
-  for(let day=rangeStart.startOf('day');day<=daysEnd;day=day.plus({days:1})){
+    .sort((a, b) => a.start.valueOf() - b.start.valueOf())
-    let s=day.set({hour:hs,minute:ms,second:0,millisecond:0});
+  const merged = []
-    let e=day.set({hour:he,minute:me,second:0,millisecond:0});
-    if(e<=s||e<=rangeStart||s>=rangeEnd)continue;
-    if(s<rangeStart)s=rangeStart;
-    if(e>rangeEnd)e=rangeEnd;
-    windows.push({start:s,end:e});
-  }
-  if(!windows.length)return[];
-  const busy=[...c1,...c2].map(v=>{
-    let s=iso(v.start),e=iso(v.end);
-    if(e<=s||e<=rangeStart||s>=rangeEnd)return null;
-    if(s<rangeStart)s=rangeStart;
-    if(e>rangeEnd)e=rangeEnd;
-    return{start:s,end:e};
-  }).filter(Boolean).sort((a,b)=>a.start.valueOf()-b.start.valueOf());
-  const merged=[];
   for (const slot of busy) {
-    const last=merged[merged.length-1];
+    const last = merged.at(-1)
-    if(last&&slot.start<=last.end){
+    if (!last || slot.start > last.end) merged.push({...slot})
-      if(slot.end>last.end)last.end=slot.end;
+    else if (slot.end > last.end) last.end = slot.end
-    }else merged.push({start:slot.start,end:slot.end});
   }
-  const out=[];
+  const out = []
-  const push=(s,e)=>e.diff(s,'minutes').minutes>=d&&out.push({start:s.toISO(),end:e.toISO()});
+  const emit = (from, to) => {
-  for(const wSlot of windows){
+    if (!(to > from)) return
-    let cur=wSlot.start;
+    for (let st = from, en = st.plus({minutes: d}); en <= to; st = en, en = st.plus({minutes: d}))
-    for(const b of merged){
+      out.push({start: st.toISO(), end: en.toISO()})
-      if(b.start>=wSlot.end)break;
-      if(b.end<=wSlot.start)continue;
-      const bs=b.start>wSlot.start?b.start:wSlot.start;
-      const be=b.end<wSlot.end?b.end:wSlot.end;
-      if(bs>cur)push(cur,bs);
-      if(be>cur)cur=be;
-      if(cur>=wSlot.end)break;
   }
-    if(cur<wSlot.end)push(cur,wSlot.end);
+  let i = 0
+  for (let day = s.startOf('day'); day < e; day = day.plus({days: 1})) {
+    const ws = day.set({hour: sh, minute: sm, second: 0, millisecond: 0})
+    const we = day.set({hour: eh, minute: em, second: 0, millisecond: 0})
+    const block = Interval.fromDateTimes(ws, we).intersection(range)
+    if (!block) continue
+    while (i < merged.length && merged[i].end <= block.start) i++
+    let cursor = block.start
+    for (let j = i; j < merged.length && merged[j].start < block.end; j++) {
+      const bs = merged[j].start > block.start ? merged[j].start : block.start
+      if (bs > cursor) {
+        emit(cursor, bs)
+        cursor = bs
+      }
+      if (merged[j].end > cursor) {
+        const be = merged[j].end < block.end ? merged[j].end : block.end
+        cursor = be
+      }
+      if (cursor >= block.end) break
+    }
+    if (cursor < block.end) emit(cursor, block.end)
+  }
+  return out
 }
-  return out;
-};
 export default findAvailableSlots;

tests/8_json_validator/outputs/anthropic_claude-sonnet-4.5 TEMP_0.7.js

@@ -1,15 +1,17 @@
 const validateJSON = async (data, schema) => {
-  const { default: Ajv } = await import('https://cdn.jsdelivr.net/npm/ajv@8/dist/ajv.bundle.js');
+  const { default: Ajv } = await import('https://esm.sh/ajv@8.12.0');
-  
+  const ajv = new Ajv({ allErrors: true });
-  const ajv = new Ajv({ allErrors: true, verbose: true });
   const validate = ajv.compile(schema);
   const valid = validate(data);
   
   return {
     valid,
-    errors: valid ? [] : validate.errors.map(e => 
+    errors: valid ? [] : validate.errors.map(e => ({
-      `${e.instancePath || '/'} ${e.message}${e.params ? ': ' + JSON.stringify(e.params) : ''}`
+      path: e.instancePath || '/',
-    )
+      message: e.message,
+      keyword: e.keyword,
+      params: e.params
+    }))
   };
 };
 export default validateJSON;

tests/8_json_validator/outputs/google_gemini-2.5-pro.js

@@ -1,35 +1,17 @@
-const validateJSON = async (jsonObject, schemaObject) => {
+const validateJSON = async (data, schema) => {
   try {
-    const [{
+    const { default: Ajv } = await import('https://esm.sh/ajv@8');
-      default: Ajv
+    const validate = new Ajv({ allErrors: true }).compile(schema);
-    }, {
+    const valid = validate(data);
-      default: addFormats
-    }] = await Promise.all([
-      import('https://esm.sh/ajv'),
-      import('https://esm.sh/ajv-formats')
-    ]);
-
-    const ajv = new Ajv({
-      allErrors: true
-    });
-    addFormats(ajv);
-
-    const validate = ajv.compile(schemaObject);
-    const valid = validate(jsonObject);
-
-    const errors = valid ? [] : validate.errors.map(
-      e => `${e.instancePath || '<root>'}: ${e.message}`
-    );
 
     return {
       valid,
-      errors
+      errors: valid ? [] : (validate.errors ?? []).map(
+        ({ instancePath, message }) => `${instancePath || 'root'} ${message}`.trim()
+      ),
     };
   } catch (e) {
-    return {
+    return { valid: false, errors: [e.message] };
-      valid: false,
-      errors: [`Validation setup failed: ${e.message}`]
-    };
   }
 };
 export default validateJSON;

tests/8_json_validator/outputs/openai_gpt-5.1-chat.js

@@ -1,8 +1,8 @@
-async function validateJSON(o, s) {
+async function validateJSON(d,s){
-  const { default: Ajv } = await import('https://esm.sh/ajv@8');
+  const {default:Ajv}=await import('https://cdn.skypack.dev/ajv');
   const a=new Ajv({allErrors:true});
   const v=a.compile(s);
-  const ok = v(o);
+  const ok=v(d);
-  return { valid: ok, errors: ok ? [] : v.errors.map(e => e.message) };
+  return{valid:ok,errors:ok?[]:(v.errors||[]).map(e=>e.message||'')};
 }
 export default validateJSON;

tests/8_json_validator/outputs/openai_gpt-5.1-codex.js

@@ -1,10 +1,9 @@
-const validateJSON=async(j,s)=>{
+export async function validateJSON(data, schema) {
-  const{default:Ajv}=await import('https://esm.sh/ajv@8')
+  const { default: Ajv } = await import('https://esm.sh/ajv@8?bundle');
-  const{default:f}=await import('https://esm.sh/ajv-formats@2')
+  const ajv = new Ajv({ allErrors: true, strict: false });
-  const a=new Ajv({allErrors:1,strict:false})
+  const validate = ajv.compile(schema);
-  f(a)
+  const valid = validate(data);
-  const v=a.compile(s)
+  const errors = valid ? [] : (validate.errors || []).map(e => `${e.instancePath || '/'} ${e.message || ''}`.trim());
-  const ok=v(j)
+  return { valid, errors };
-  return{valid:ok,errors:ok?[]:(v.errors||[]).map(e=>`${e.instancePath||'/'} ${e.message}`.trim())}
 }
 export default validateJSON;

tests/9_stream_visualizer/outputs/anthropic_claude-sonnet-4.5 TEMP_0.7.js

@@ -0,0 +1,30 @@
+export async function createStreamVisualizer(asyncIterable, options) {
+  const { maxPoints, alpha, width, height, yDomain } = options;
+  const d3 = await import('https://cdn.jsdelivr.net/npm/d3@7/+esm');
+  
+  const data = [];
+  let prevEma;
+  
+  for await (const { timestamp, value } of asyncIterable) {
+    const ema = prevEma === undefined ? value : alpha * value + (1 - alpha) * prevEma;
+    data.push({ timestamp, value, ema });
+    if (data.length > maxPoints) data.shift();
+    prevEma = ema;
+  }
+  
+  const xScale = d3.scaleLinear()
+    .domain([data[0].timestamp, data[data.length - 1].timestamp])
+    .range([0, width]);
+  
+  const yScale = d3.scaleLinear()
+    .domain(yDomain)
+    .range([height, 0]);
+  
+  const path = d3.line()
+    .x(d => xScale(d.timestamp))
+    .y(d => yScale(d.ema))
+    (data);
+  
+  return { data, path };
+}
+export default createStreamVisualizer;

tests/9_stream_visualizer/outputs/google_gemini-2.5-pro.js

@@ -0,0 +1,60 @@
+async function createStreamVisualizer(asyncIterable, {
+  maxPoints,
+  alpha,
+  width,
+  height,
+  yDomain
+}) {
+  const {
+    scaleLinear,
+    line
+  } = await import('d3');
+
+  const data = [];
+  let prevEma = null;
+
+  for await (const {
+      timestamp,
+      value
+    } of asyncIterable) {
+    const ema = prevEma === null ?
+      value :
+      alpha * value + (1 - alpha) * prevEma;
+
+    data.push({
+      timestamp,
+      value,
+      ema
+    });
+    prevEma = ema;
+
+    if (data.length > maxPoints) {
+      data.shift();
+    }
+  }
+
+  if (data.length < 2) {
+    return {
+      data,
+      path: ''
+    };
+  }
+
+  const x = scaleLinear()
+    .domain([data[0].timestamp, data.at(-1).timestamp])
+    .range([0, width]);
+
+  const y = scaleLinear()
+    .domain(yDomain)
+    .range([height, 0]);
+
+  const path = line()
+    .x(d => x(d.timestamp))
+    .y(d => y(d.ema))(data);
+
+  return {
+    data,
+    path
+  };
+}
+export default createStreamVisualizer;

tests/9_stream_visualizer/outputs/openai_gpt-5.1-chat.js

@@ -0,0 +1,23 @@
+export async function createStreamVisualizer(iter, o) {
+  const { maxPoints, alpha, width, height, yDomain } = o
+  const d3 = await import('https://cdn.jsdelivr.net/npm/d3@7/+esm')
+  const data = []
+  let ema
+  for await (const { timestamp, value } of iter) {
+    ema = ema == null ? value : alpha * value + (1 - alpha) * ema
+    data.push({ timestamp, value, ema })
+    if (data.length > maxPoints) data.shift()
+  }
+  const xs = d3.scaleLinear()
+    .domain([data[0].timestamp, data[data.length - 1].timestamp])
+    .range([0, width])
+  const ys = d3.scaleLinear()
+    .domain(yDomain)
+    .range([height, 0])
+  const line = d3.line()
+    .x(d => xs(d.timestamp))
+    .y(d => ys(d.ema))
+  const path = line(data) || ''
+  return { data, path }
+}
+export default createStreamVisualizer;

tests/9_stream_visualizer/outputs/openai_gpt-5.1-codex.js

@@ -0,0 +1,23 @@
+async function createStreamVisualizer(iterable, {
+  maxPoints = 100,
+  alpha = .5,
+  width = 300,
+  height = 150,
+  yDomain = [0, 1]
+} = {}) {
+  const { scaleLinear, line } = await import('https://cdn.jsdelivr.net/npm/d3@7/+esm')
+  const data = []
+  let ema
+  for await (const point of iterable) {
+    const { timestamp, value } = point
+    ema = ema == null ? value : alpha * value + (1 - alpha) * ema
+    data.push({ timestamp, value, ema })
+    data.length > maxPoints && data.shift()
+  }
+  if (!data.length) return { data, path: '' }
+  const x = scaleLinear().domain([data[0].timestamp, data.at(-1).timestamp]).range([0, width])
+  const y = scaleLinear().domain(yDomain).range([height, 0])
+  const path = line().x(d => x(d.timestamp)).y(d => y(d.ema))(data) || ''
+  return { data, path }
+}
+export default createStreamVisualizer;