agent-duelist 0.2.0 → 0.2.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/index.js CHANGED
@@ -1410,11 +1410,13 @@ var fuzzySimilarityScorer = ({ task, result }) => {
1410
1410
  }
1411
1411
  const a = stringify(task.expected);
1412
1412
  const b = stringify(result.output);
1413
- const similarity = jaccardSimilarity(tokenize(a), tokenize(b));
1413
+ const setA = tokenize(a);
1414
+ const setB = tokenize(b);
1415
+ const similarity = jaccardSimilarity(setA, setB);
1414
1416
  return {
1415
1417
  name: "fuzzy-similarity",
1416
1418
  value: Math.round(similarity * 100) / 100,
1417
- details: { method: "jaccard", expectedTokens: tokenize(a).size, actualTokens: tokenize(b).size }
1419
+ details: { method: "jaccard", expectedTokens: setA.size, actualTokens: setB.size }
1418
1420
  };
1419
1421
  };
1420
1422
  function stringify(value) {
@@ -1440,6 +1442,19 @@ import OpenAI2, { AzureOpenAI as AzureOpenAI2 } from "openai";
1440
1442
  // src/providers/openai.ts
1441
1443
  import OpenAI, { AzureOpenAI } from "openai";
1442
1444
  import { zodToJsonSchema } from "zod-to-json-schema";
1445
+
1446
+ // src/providers/shared.ts
1447
+ var SCHEMA_SYSTEM_MESSAGE = "Respond with valid JSON matching the requested schema.";
1448
+ function parseSchemaOutput(rawContent, hasSchema) {
1449
+ if (!hasSchema) return rawContent;
1450
+ try {
1451
+ return JSON.parse(rawContent);
1452
+ } catch {
1453
+ return rawContent;
1454
+ }
1455
+ }
1456
+
1457
+ // src/providers/openai.ts
1443
1458
  var REQUEST_TIMEOUT_MS = 6e4;
1444
1459
  function openai(model, options) {
1445
1460
  const client = new OpenAI({
@@ -1486,7 +1501,7 @@ function makeProvider(id, name, model, client, requestModel, stripThinking) {
1486
1501
  if (input.schema) {
1487
1502
  params.response_format = { type: "json_object" };
1488
1503
  params.messages = [
1489
- { role: "system", content: "Respond with valid JSON matching the requested schema." },
1504
+ { role: "system", content: SCHEMA_SYSTEM_MESSAGE },
1490
1505
  ...params.messages
1491
1506
  ];
1492
1507
  }
@@ -1539,13 +1554,7 @@ function makeProvider(id, name, model, client, requestModel, stripThinking) {
1539
1554
  if (stripThinking) {
1540
1555
  rawContent = rawContent.replace(/<think>[\s\S]*?<\/think>\s*/, "");
1541
1556
  }
1542
- let output = rawContent;
1543
- if (input.schema) {
1544
- try {
1545
- output = JSON.parse(rawContent);
1546
- } catch {
1547
- }
1548
- }
1557
+ const output = parseSchemaOutput(rawContent, !!input.schema);
1549
1558
  return {
1550
1559
  output,
1551
1560
  usage: {
@@ -1559,6 +1568,20 @@ function makeProvider(id, name, model, client, requestModel, stripThinking) {
1559
1568
  }
1560
1569
  };
1561
1570
  }
1571
+ function gemini(model, options) {
1572
+ const apiKey = options?.apiKey ?? process.env.GOOGLE_API_KEY;
1573
+ if (!apiKey) {
1574
+ throw new Error(
1575
+ `Missing API key for google/${model}. Set GOOGLE_API_KEY or pass apiKey option.`
1576
+ );
1577
+ }
1578
+ const client = new OpenAI({
1579
+ apiKey,
1580
+ baseURL: "https://generativelanguage.googleapis.com/v1beta/openai/",
1581
+ timeout: options?.timeoutMs ?? REQUEST_TIMEOUT_MS
1582
+ });
1583
+ return makeProvider(`google/${model}`, "Google AI", model, client, model);
1584
+ }
1562
1585
  function toolDefToOpenAI(tool) {
1563
1586
  return {
1564
1587
  type: "function",
@@ -1636,8 +1659,7 @@ function createLlmJudgeScorer(judgeModel, timeoutMs = REQUEST_TIMEOUT_MS) {
1636
1659
  const response = await client.chat.completions.create({
1637
1660
  model,
1638
1661
  messages: [{ role: "user", content: prompt }],
1639
- temperature: 0,
1640
- max_tokens: 2048
1662
+ max_completion_tokens: 2048
1641
1663
  });
1642
1664
  const content = response.choices[0]?.message?.content?.trim() ?? "";
1643
1665
  const parsed = {};
@@ -1733,118 +1755,173 @@ async function runBenchmarks(options) {
1733
1755
  const timeout = options.timeout ?? DEFAULT_TIMEOUT_MS;
1734
1756
  const results = [];
1735
1757
  for (const task of tasks) {
1736
- for (const provider of providers) {
1737
- for (let run = 1; run <= runs; run++) {
1738
- let result;
1739
- try {
1740
- const taskResult = await withTimeout((signal) => provider.run({
1741
- prompt: task.prompt,
1742
- schema: task.schema,
1743
- tools: task.tools,
1744
- signal
1745
- }), timeout);
1746
- const scores = await Promise.all(
1747
- scorers.map((scorer) => scorer({ task, result: taskResult }, provider.id))
1748
- );
1749
- result = {
1750
- providerId: provider.id,
1751
- taskName: task.name,
1752
- run,
1753
- scores,
1754
- raw: {
1755
- output: taskResult.output,
1756
- latencyMs: taskResult.latencyMs,
1757
- usage: taskResult.usage,
1758
- toolCalls: taskResult.toolCalls
1759
- }
1760
- };
1761
- } catch (err) {
1762
- const message = err instanceof Error ? err.message : String(err);
1763
- result = {
1764
- providerId: provider.id,
1765
- taskName: task.name,
1766
- run,
1767
- scores: [],
1768
- error: message,
1769
- raw: { output: "", latencyMs: 0 }
1770
- };
1771
- }
1772
- results.push(result);
1773
- onResult?.(result);
1774
- }
1758
+ for (let run = 1; run <= runs; run++) {
1759
+ const runResults = await Promise.all(
1760
+ providers.map(async (provider) => {
1761
+ let result;
1762
+ try {
1763
+ const taskResult = await withTimeout((signal) => provider.run({
1764
+ prompt: task.prompt,
1765
+ schema: task.schema,
1766
+ tools: task.tools,
1767
+ signal
1768
+ }), timeout);
1769
+ const scores = await Promise.all(
1770
+ scorers.map((scorer) => scorer({ task, result: taskResult }, provider.id))
1771
+ );
1772
+ result = {
1773
+ providerId: provider.id,
1774
+ taskName: task.name,
1775
+ run,
1776
+ scores,
1777
+ raw: {
1778
+ output: taskResult.output,
1779
+ latencyMs: taskResult.latencyMs,
1780
+ usage: taskResult.usage,
1781
+ toolCalls: taskResult.toolCalls
1782
+ }
1783
+ };
1784
+ } catch (err) {
1785
+ const message = err instanceof Error ? err.message : String(err);
1786
+ result = {
1787
+ providerId: provider.id,
1788
+ taskName: task.name,
1789
+ run,
1790
+ scores: [],
1791
+ error: message,
1792
+ raw: { output: "", latencyMs: 0 }
1793
+ };
1794
+ }
1795
+ onResult?.(result);
1796
+ return result;
1797
+ })
1798
+ );
1799
+ results.push(...runResults);
1775
1800
  }
1776
1801
  }
1777
1802
  return results;
1778
1803
  }
1779
1804
 
1780
- // src/reporter/console.ts
1781
- var reset = "\x1B[0m";
1782
- var boldCode = "\x1B[1m";
1783
- var dimCode = "\x1B[2m";
1784
- var green = "\x1B[32m";
1785
- var red = "\x1B[31m";
1786
- var yellow = "\x1B[33m";
1787
- var cyan = "\x1B[36m";
1788
- var brightGreen = "\x1B[92m";
1789
- var brightWhite = "\x1B[97m";
1790
- function bold(s) {
1791
- return `${boldCode}${s}${reset}`;
1792
- }
1793
- function dim(s) {
1794
- return `${dimCode}${s}${reset}`;
1805
+ // src/utils/format.ts
1806
+ var MAX_FRACTION_DIGITS = 100;
1807
+ function formatCost(usd) {
1808
+ if (usd === void 0) return "\u2014";
1809
+ if (usd === 0) return "$0.00";
1810
+ if (usd >= 0.01) return `~$${usd.toFixed(2)}`;
1811
+ const digits = Math.min(
1812
+ MAX_FRACTION_DIGITS,
1813
+ Math.max(4, -Math.floor(Math.log10(Math.abs(usd))) + 1)
1814
+ );
1815
+ return `~$${usd.toFixed(digits).replace(/0+$/, "")}`;
1795
1816
  }
1796
- function stripAnsi(s) {
1797
- return s.replace(/\x1b\[[0-9;]*m/g, "");
1817
+ function formatDelta(delta, precision = 4) {
1818
+ const sign = delta >= 0 ? "+" : "";
1819
+ return `${sign}${delta.toFixed(precision)}`;
1798
1820
  }
1799
- function displayWidth(s) {
1800
- const stripped = stripAnsi(s);
1801
- let width = 0;
1802
- for (const ch of stripped) {
1803
- const code = ch.codePointAt(0) ?? 0;
1804
- if (code >= 126976) width += 2;
1805
- else if (code >= 9728 && code <= 10175) width += 2;
1806
- else width += 1;
1821
+
1822
+ // src/reporter/shared.ts
1823
+ function groupResults(results) {
1824
+ const taskSet = /* @__PURE__ */ new Set();
1825
+ const providerSet = /* @__PURE__ */ new Set();
1826
+ const scorerSet = /* @__PURE__ */ new Set();
1827
+ const grouped = /* @__PURE__ */ new Map();
1828
+ const byProvider = /* @__PURE__ */ new Map();
1829
+ let hasErrors = false;
1830
+ let maxRun = 0;
1831
+ for (const r of results) {
1832
+ taskSet.add(r.taskName);
1833
+ providerSet.add(r.providerId);
1834
+ for (const s of r.scores) scorerSet.add(s.name);
1835
+ if (r.error) hasErrors = true;
1836
+ if (r.run > maxRun) maxRun = r.run;
1837
+ const key = `${r.taskName}::${r.providerId}`;
1838
+ let group = grouped.get(key);
1839
+ if (!group) {
1840
+ group = [];
1841
+ grouped.set(key, group);
1842
+ }
1843
+ group.push(r);
1844
+ let provGroup = byProvider.get(r.providerId);
1845
+ if (!provGroup) {
1846
+ provGroup = [];
1847
+ byProvider.set(r.providerId, provGroup);
1848
+ }
1849
+ provGroup.push(r);
1807
1850
  }
1808
- return width;
1809
- }
1810
- function padCell(str, targetWidth, align) {
1811
- const dw = displayWidth(str);
1812
- const padding = Math.max(0, targetWidth - dw);
1813
- if (align === "right") return " ".repeat(padding) + str;
1814
- return str + " ".repeat(padding);
1815
- }
1816
- function sparkBar(ratio, width = 8) {
1817
- const clamped = Math.max(0, Math.min(1, ratio));
1818
- const fillLen = Math.round(clamped * width);
1819
- const fill = "\u2593".repeat(fillLen);
1820
- const track = "\u2591".repeat(width - fillLen);
1821
- return { fill, track };
1851
+ return {
1852
+ tasks: [...taskSet],
1853
+ providers: [...providerSet],
1854
+ scorerNames: [...scorerSet],
1855
+ grouped,
1856
+ byProvider,
1857
+ hasErrors,
1858
+ maxRun
1859
+ };
1822
1860
  }
1823
- function drawTableLine(widths, position) {
1824
- const totalInner = widths.reduce((sum, w) => sum + w + 2, 0) + widths.length - 1;
1825
- if (position === "bottom") {
1826
- return dim(`\u2514${"\u2500".repeat(totalInner)}\u2518`);
1861
+ function aggregateProviderTask(providerId, grouped, task) {
1862
+ const taskResults = grouped.get(`${task}::${providerId}`) ?? [];
1863
+ const errorResults = taskResults.filter((r) => r.error);
1864
+ const successResults = taskResults.filter((r) => !r.error);
1865
+ if (successResults.length === 0) {
1866
+ return {
1867
+ providerId,
1868
+ avgScores: {},
1869
+ avgDetails: { costUsd: void 0, totalTokens: void 0 },
1870
+ latencyMs: void 0,
1871
+ allErrors: errorResults.length > 0,
1872
+ errorCount: errorResults.length
1873
+ };
1827
1874
  }
1828
- if (position === "merge") {
1829
- return dim(`\u251C${"\u2500".repeat(totalInner)}\u2524`);
1875
+ return {
1876
+ providerId,
1877
+ avgScores: averageScores(successResults),
1878
+ avgDetails: averageDetails(successResults),
1879
+ latencyMs: average(successResults.map((r) => r.raw.latencyMs)),
1880
+ allErrors: false,
1881
+ errorCount: errorResults.length
1882
+ };
1883
+ }
1884
+ function averageScores(results) {
1885
+ const sums = {};
1886
+ const counts = {};
1887
+ for (const result of results) {
1888
+ for (const score of result.scores) {
1889
+ if (score.value < 0) continue;
1890
+ sums[score.name] = (sums[score.name] ?? 0) + score.value;
1891
+ counts[score.name] = (counts[score.name] ?? 0) + 1;
1892
+ }
1830
1893
  }
1831
- const segments = widths.map((w) => "\u2500".repeat(w + 2));
1832
- if (position === "top") {
1833
- return dim(`\u250C${segments.join("\u252C")}\u2510`);
1894
+ const avgs = {};
1895
+ for (const name of Object.keys(sums)) {
1896
+ avgs[name] = sums[name] / counts[name];
1834
1897
  }
1835
- return dim(`\u251C${segments.join("\u253C")}\u2524`);
1898
+ return avgs;
1836
1899
  }
1837
- function drawTableRow(cells, widths, aligns) {
1838
- const parts = cells.map(
1839
- (cell, i) => " " + padCell(cell, widths[i], aligns[i]) + " "
1840
- );
1841
- return dim("\u2502") + parts.join(dim("\u2502")) + dim("\u2502");
1900
+ function averageDetails(results) {
1901
+ let costSum = 0;
1902
+ let costCount = 0;
1903
+ let tokenSum = 0;
1904
+ let tokenCount = 0;
1905
+ for (const result of results) {
1906
+ const costScore = result.scores.find((s) => s.name === "cost");
1907
+ const details = costScore?.details;
1908
+ if (details?.estimatedUsd != null) {
1909
+ costSum += details.estimatedUsd;
1910
+ costCount++;
1911
+ }
1912
+ if (details?.totalTokens != null) {
1913
+ tokenSum += details.totalTokens;
1914
+ tokenCount++;
1915
+ }
1916
+ }
1917
+ return {
1918
+ costUsd: costCount > 0 ? costSum / costCount : void 0,
1919
+ totalTokens: tokenCount > 0 ? Math.round(tokenSum / tokenCount) : void 0
1920
+ };
1842
1921
  }
1843
- function drawSpanRow(content, widths) {
1844
- const totalInner = widths.reduce((sum, w) => sum + w + 2, 0) + widths.length - 1;
1845
- const dw = displayWidth(content);
1846
- const padding = Math.max(0, totalInner - dw - 1);
1847
- return dim("\u2502") + " " + content + " ".repeat(padding) + dim("\u2502");
1922
+ function average(nums) {
1923
+ if (nums.length === 0) return void 0;
1924
+ return nums.reduce((a, b) => a + b, 0) / nums.length;
1848
1925
  }
1849
1926
  function computeColumnStats(providerData, scorerNames) {
1850
1927
  const stats = /* @__PURE__ */ new Map();
@@ -1896,62 +1973,235 @@ function computeColumnStats(providerData, scorerNames) {
1896
1973
  }
1897
1974
  return stats;
1898
1975
  }
1899
- function colorByRank(text, value, colStats, providerCount) {
1900
- if (value === void 0) return dim("\u2014");
1901
- if (providerCount < 2) return text;
1902
- if (colStats.best === void 0 || colStats.worst === void 0) return text;
1903
- if (colStats.best === colStats.worst) return text;
1904
- if (value === colStats.best) return `${brightGreen}${boldCode}${text}${reset}`;
1905
- if (value === colStats.worst) return `${red}${text}${reset}`;
1906
- return `${yellow}${text}${reset}`;
1907
- }
1908
1976
  function computeMedals(columnStats, providerIds) {
1909
1977
  const medals = /* @__PURE__ */ new Map();
1910
1978
  if (providerIds.length < 2) {
1911
- for (const id of providerIds) medals.set(id, "");
1979
+ for (const id of providerIds) medals.set(id, "none");
1912
1980
  return medals;
1913
1981
  }
1914
1982
  const wins = /* @__PURE__ */ new Map();
1915
1983
  for (const id of providerIds) wins.set(id, 0);
1916
1984
  for (const [, colStats] of columnStats) {
1917
1985
  if (colStats.best === void 0) continue;
1918
- for (const [providerId, value] of colStats.values) {
1919
- if (value !== void 0 && value === colStats.best) {
1920
- wins.set(providerId, (wins.get(providerId) ?? 0) + 1);
1921
- }
1986
+ const bestProviders = [...colStats.values.entries()].filter(([, v]) => v !== void 0 && v === colStats.best);
1987
+ if (bestProviders.length === 1) {
1988
+ wins.set(bestProviders[0][0], (wins.get(bestProviders[0][0]) ?? 0) + 1);
1922
1989
  }
1923
1990
  }
1924
1991
  const totalWins = [...wins.values()].reduce((a, b) => a + b, 0);
1925
1992
  if (totalWins === 0) {
1926
- for (const id of providerIds) medals.set(id, "");
1993
+ for (const id of providerIds) medals.set(id, "none");
1927
1994
  return medals;
1928
1995
  }
1929
1996
  const sorted = [...wins.entries()].sort(
1930
1997
  (a, b) => b[1] - a[1] || a[0].localeCompare(b[0])
1931
1998
  );
1932
- const medalList = ["\u{1F947}", "\u{1F948}", "\u{1F949}"];
1999
+ const medalList = ["gold", "silver", "bronze"];
1933
2000
  let rank = 0;
1934
2001
  for (let i = 0; i < sorted.length; i++) {
1935
2002
  if (i > 0 && sorted[i][1] < sorted[i - 1][1]) {
1936
2003
  rank = i;
1937
2004
  }
1938
- medals.set(sorted[i][0], rank < medalList.length ? medalList[rank] : "");
2005
+ const hasWins = sorted[i][1] > 0;
2006
+ medals.set(sorted[i][0], hasWins && rank < medalList.length ? medalList[rank] : "none");
1939
2007
  }
1940
2008
  return medals;
1941
2009
  }
2010
+ function providerLabel(providerId) {
2011
+ const prefix = providerId.split("/")[0];
2012
+ switch (prefix) {
2013
+ case "azure":
2014
+ return "(OpenAI via Azure)";
2015
+ case "openai":
2016
+ return "(OpenAI)";
2017
+ case "anthropic":
2018
+ return "(Anthropic)";
2019
+ case "google":
2020
+ return "(Google)";
2021
+ case "mistral":
2022
+ return "(Mistral)";
2023
+ case "meta":
2024
+ return "(Meta)";
2025
+ case "deepseek":
2026
+ return "(DeepSeek)";
2027
+ case "cohere":
2028
+ return "(Cohere)";
2029
+ case "qwen":
2030
+ return "(Qwen)";
2031
+ case "xai":
2032
+ return "(xAI)";
2033
+ case "minimax":
2034
+ return "(MiniMax)";
2035
+ case "moonshot":
2036
+ return "(Moonshot / Kimi)";
2037
+ case "perplexity":
2038
+ return "(Perplexity)";
2039
+ case "amazon":
2040
+ return "(Amazon)";
2041
+ case "nvidia":
2042
+ return "(NVIDIA)";
2043
+ case "microsoft":
2044
+ return "(Microsoft)";
2045
+ case "ai21":
2046
+ return "(AI21 Labs)";
2047
+ case "bytedance":
2048
+ return "(ByteDance)";
2049
+ case "together":
2050
+ return "(Together AI)";
2051
+ case "fireworks":
2052
+ return "(Fireworks AI)";
2053
+ case "groq":
2054
+ return "(Groq)";
2055
+ case "cerebras":
2056
+ return "(Cerebras)";
2057
+ default:
2058
+ return `(${prefix})`;
2059
+ }
2060
+ }
2061
+ function apiKeyHint(providerId, error) {
2062
+ const lower = error.toLowerCase();
2063
+ const isAuthError = lower.includes("api key") || lower.includes("401") || lower.includes("unauthorized") || lower.includes("authentication") || lower.includes("incorrect api key") || lower.includes("apikey");
2064
+ if (!isAuthError) return void 0;
2065
+ const prefix = providerId.split("/")[0];
2066
+ switch (prefix) {
2067
+ case "openai":
2068
+ return "Set: export OPENAI_API_KEY=sk-...";
2069
+ case "azure":
2070
+ return "Set: export AZURE_OPENAI_API_KEY=... and AZURE_OPENAI_ENDPOINT=...";
2071
+ case "anthropic":
2072
+ return "Set: export ANTHROPIC_API_KEY=sk-ant-...";
2073
+ case "google":
2074
+ return "Set: export GOOGLE_API_KEY=...";
2075
+ default:
2076
+ return `Check the API key for ${providerId}`;
2077
+ }
2078
+ }
2079
+ function rankProviders(successByProvider, providers, scorerName) {
2080
+ const ranked = providers.map((id) => {
2081
+ const runs = successByProvider.get(id) ?? [];
2082
+ const scores = runs.flatMap((r) => r.scores.filter((s) => s.name === scorerName && s.value >= 0)).map((s) => s.value);
2083
+ const avg = scores.length > 0 ? scores.reduce((a, b) => a + b, 0) / scores.length : void 0;
2084
+ return { id, avg };
2085
+ }).filter((p) => p.avg !== void 0).sort((a, b) => b.avg - a.avg);
2086
+ return ranked[0] ? { id: ranked[0].id, avg: ranked[0].avg } : void 0;
2087
+ }
2088
+ function scorerLabel(name) {
2089
+ switch (name) {
2090
+ case "correctness":
2091
+ return "Match";
2092
+ case "schema-correctness":
2093
+ return "Schema";
2094
+ case "fuzzy-similarity":
2095
+ return "Fuzzy";
2096
+ case "llm-judge-correctness":
2097
+ return "Judge";
2098
+ case "tool-usage":
2099
+ return "Tool";
2100
+ default:
2101
+ return name;
2102
+ }
2103
+ }
2104
+ function medalEmoji(medal) {
2105
+ switch (medal) {
2106
+ case "gold":
2107
+ return "\u{1F947}";
2108
+ case "silver":
2109
+ return "\u{1F948}";
2110
+ case "bronze":
2111
+ return "\u{1F949}";
2112
+ case "none":
2113
+ return "";
2114
+ }
2115
+ }
2116
+
2117
+ // src/reporter/console.ts
2118
+ var reset = "\x1B[0m";
2119
+ var boldCode = "\x1B[1m";
2120
+ var dimCode = "\x1B[2m";
2121
+ var green = "\x1B[32m";
2122
+ var red = "\x1B[31m";
2123
+ var yellow = "\x1B[33m";
2124
+ var cyan = "\x1B[36m";
2125
+ var brightGreen = "\x1B[92m";
2126
+ var brightWhite = "\x1B[97m";
2127
+ function bold(s) {
2128
+ return `${boldCode}${s}${reset}`;
2129
+ }
2130
+ function dim(s) {
2131
+ return `${dimCode}${s}${reset}`;
2132
+ }
2133
+ function stripAnsi(s) {
2134
+ return s.replace(/\x1b\[[0-9;]*m/g, "");
2135
+ }
2136
+ function displayWidth(s) {
2137
+ const stripped = stripAnsi(s);
2138
+ let width = 0;
2139
+ for (const ch of stripped) {
2140
+ const code = ch.codePointAt(0) ?? 0;
2141
+ if (code >= 126976) width += 2;
2142
+ else if (code >= 9728 && code <= 10175) width += 2;
2143
+ else width += 1;
2144
+ }
2145
+ return width;
2146
+ }
2147
+ function padCell(str, targetWidth, align) {
2148
+ const dw = displayWidth(str);
2149
+ const padding = Math.max(0, targetWidth - dw);
2150
+ if (align === "right") return " ".repeat(padding) + str;
2151
+ return str + " ".repeat(padding);
2152
+ }
2153
+ function sparkBar(ratio, width = 8) {
2154
+ const clamped = Math.max(0, Math.min(1, ratio));
2155
+ const fillLen = Math.round(clamped * width);
2156
+ const fill = "\u2593".repeat(fillLen);
2157
+ const track = "\u2591".repeat(width - fillLen);
2158
+ return { fill, track };
2159
+ }
2160
+ function drawTableLine(widths, position) {
2161
+ const totalInner = widths.reduce((sum, w) => sum + w + 2, 0) + widths.length - 1;
2162
+ if (position === "bottom") {
2163
+ return dim(`\u2514${"\u2500".repeat(totalInner)}\u2518`);
2164
+ }
2165
+ if (position === "merge") {
2166
+ return dim(`\u251C${"\u2500".repeat(totalInner)}\u2524`);
2167
+ }
2168
+ const segments = widths.map((w) => "\u2500".repeat(w + 2));
2169
+ if (position === "top") {
2170
+ return dim(`\u250C${segments.join("\u252C")}\u2510`);
2171
+ }
2172
+ return dim(`\u251C${segments.join("\u253C")}\u2524`);
2173
+ }
2174
+ function drawTableRow(cells, widths, aligns) {
2175
+ const parts = cells.map(
2176
+ (cell, i) => " " + padCell(cell, widths[i], aligns[i]) + " "
2177
+ );
2178
+ return dim("\u2502") + parts.join(dim("\u2502")) + dim("\u2502");
2179
+ }
2180
+ function drawSpanRow(content, widths) {
2181
+ const totalInner = widths.reduce((sum, w) => sum + w + 2, 0) + widths.length - 1;
2182
+ const dw = displayWidth(content);
2183
+ const padding = Math.max(0, totalInner - dw - 1);
2184
+ return dim("\u2502") + " " + content + " ".repeat(padding) + dim("\u2502");
2185
+ }
2186
+ function colorByRank(text, value, colStats, providerCount) {
2187
+ if (value === void 0) return dim("\u2014");
2188
+ if (providerCount < 2) return text;
2189
+ if (colStats.best === void 0 || colStats.worst === void 0) return text;
2190
+ if (colStats.best === colStats.worst) return text;
2191
+ if (value === colStats.best) return `${brightGreen}${boldCode}${text}${reset}`;
2192
+ if (value === colStats.worst) return `${red}${text}${reset}`;
2193
+ return `${yellow}${text}${reset}`;
2194
+ }
1942
2195
  function consoleReporter(results, options) {
1943
2196
  const showSparklines = options?.sparklines ?? true;
1944
2197
  if (results.length === 0) {
1945
2198
  console.log("\nNo results to display.\n");
1946
2199
  return;
1947
2200
  }
1948
- const tasks = [...new Set(results.map((r) => r.taskName))];
1949
- const providers = [...new Set(results.map((r) => r.providerId))];
1950
- const scorerNames = [...new Set(results.flatMap((r) => r.scores.map((s) => s.name)))];
2201
+ const { tasks, providers, scorerNames, grouped, byProvider, hasErrors, maxRun } = groupResults(results);
1951
2202
  const hasCost = scorerNames.includes("cost");
1952
- const hasErrors = results.some((r) => r.error);
1953
2203
  const multi = providers.length >= 2;
1954
- const runsPerCell = Math.max(...results.map((r) => r.run));
2204
+ const runsPerCell = maxRun;
1955
2205
  const runLabel = runsPerCell > 1 ? ` ${dim(`(${runsPerCell} runs each)`)}` : "";
1956
2206
  console.log("");
1957
2207
  console.log(` ${brightWhite}${boldCode}\u2B21 Agent Duelist${reset}${runLabel}`);
@@ -1960,29 +2210,9 @@ function consoleReporter(results, options) {
1960
2210
  for (const task of tasks) {
1961
2211
  console.log(` ${bold(`Task: ${task}`)}`);
1962
2212
  console.log("");
1963
- const providerData = providers.map((providerId) => {
1964
- const taskResults = results.filter((r) => r.taskName === task && r.providerId === providerId);
1965
- const errorResults2 = taskResults.filter((r) => r.error);
1966
- const successResults = taskResults.filter((r) => !r.error);
1967
- if (successResults.length === 0) {
1968
- return {
1969
- providerId,
1970
- avgScores: {},
1971
- avgDetails: { costUsd: void 0, totalTokens: void 0 },
1972
- latencyMs: void 0,
1973
- allErrors: errorResults2.length > 0,
1974
- errorCount: errorResults2.length
1975
- };
1976
- }
1977
- return {
1978
- providerId,
1979
- avgScores: averageScores(successResults),
1980
- avgDetails: averageDetails(successResults),
1981
- latencyMs: average(successResults.map((r) => r.raw.latencyMs)),
1982
- allErrors: false,
1983
- errorCount: errorResults2.length
1984
- };
1985
- });
2213
+ const providerData = providers.map(
2214
+ (providerId) => aggregateProviderTask(providerId, grouped, task)
2215
+ );
1986
2216
  const columnStats = computeColumnStats(providerData, scorerNames);
1987
2217
  const medals = computeMedals(columnStats, providers);
1988
2218
  const maxProviderLen = Math.max(...providers.map((id) => id.length));
@@ -1997,8 +2227,7 @@ function consoleReporter(results, options) {
1997
2227
  cols.push({ label: "Cost", width: 12, align: "right", statsKey: "cost" });
1998
2228
  cols.push({ label: "Tokens", width: 9, align: "right", statsKey: "tokens" });
1999
2229
  } else {
2000
- const label = name === "correctness" ? "Match" : name === "schema-correctness" ? "Schema" : name === "fuzzy-similarity" ? "Fuzzy" : name === "llm-judge-correctness" ? "Judge" : name === "tool-usage" ? "Tool" : name;
2001
- cols.push({ label, width: showSparklines ? 15 : 8, align: "right", statsKey: name });
2230
+ cols.push({ label: scorerLabel(name), width: showSparklines ? 15 : 8, align: "right", statsKey: name });
2002
2231
  }
2003
2232
  }
2004
2233
  if (hasErrors) {
@@ -2011,7 +2240,7 @@ function consoleReporter(results, options) {
2011
2240
  console.log(` ${drawTableRow(headerCells, widths, aligns)}`);
2012
2241
  console.log(` ${drawTableLine(widths, "header")}`);
2013
2242
  for (const pd of providerData) {
2014
- const medal = medals.get(pd.providerId) ?? "";
2243
+ const medal = medalEmoji(medals.get(pd.providerId) ?? "none");
2015
2244
  const providerCell = medal ? `${medal} ${pd.providerId}` : pd.providerId;
2016
2245
  const cells = [providerCell];
2017
2246
  if (pd.allErrors) {
@@ -2084,7 +2313,7 @@ function consoleReporter(results, options) {
2084
2313
  console.log(` ${drawTableRow(cells, widths, aligns)}`);
2085
2314
  }
2086
2315
  if (multi && providerData.some((p) => !p.allErrors)) {
2087
- const winnerId = [...medals.entries()].find(([, m]) => m === "\u{1F947}")?.[0];
2316
+ const winnerId = [...medals.entries()].find(([, m]) => m === "gold")?.[0];
2088
2317
  if (winnerId) {
2089
2318
  console.log(` ${drawTableLine(widths, "merge")}`);
2090
2319
  const winnerText = `${brightGreen}${boldCode}\u{1F3C6} Winner: ${winnerId}${reset} ${dim(providerLabel(winnerId))}`;
@@ -2094,7 +2323,7 @@ function consoleReporter(results, options) {
2094
2323
  console.log(` ${drawTableLine(widths, "bottom")}`);
2095
2324
  console.log("");
2096
2325
  }
2097
- printSummary(results, providers);
2326
+ printSummary(results, providers, byProvider);
2098
2327
  const errorResults = results.filter((r) => r.error);
2099
2328
  if (errorResults.length > 0) {
2100
2329
  console.log(` ${bold("Errors")}`);
@@ -2117,15 +2346,19 @@ function consoleReporter(results, options) {
2117
2346
  console.log("");
2118
2347
  }
2119
2348
  }
2120
- function printSummary(results, providers) {
2349
+ function printSummary(results, providers, byProvider) {
2121
2350
  const successResults = results.filter((r) => !r.error);
2122
2351
  if (successResults.length === 0) return;
2352
+ const successByProvider = /* @__PURE__ */ new Map();
2353
+ for (const id of providers) {
2354
+ successByProvider.set(id, (byProvider.get(id) ?? []).filter((r) => !r.error));
2355
+ }
2123
2356
  console.log(` ${bold("Summary")}`);
2124
2357
  console.log(` ${dim("\u2501".repeat(72))}`);
2125
2358
  console.log("");
2126
2359
  const single = providers.length === 1;
2127
2360
  const correctnessKey = successResults.some((r) => r.scores.some((s) => s.name === "llm-judge-correctness" && s.value >= 0)) ? "llm-judge-correctness" : "correctness";
2128
- const byCorrectness = rankProviders(successResults, providers, correctnessKey);
2361
+ const byCorrectness = rankProviders(successByProvider, providers, correctnessKey);
2129
2362
  if (byCorrectness) {
2130
2363
  const medal = single ? `${cyan}\u25C6${reset}` : "\u{1F947}";
2131
2364
  const pctStr = `${Math.round(byCorrectness.avg * 100)}%`;
@@ -2136,7 +2369,7 @@ function printSummary(results, providers) {
2136
2369
  }
2137
2370
  }
2138
2371
  const byLatency = providers.map((id) => {
2139
- const runs = successResults.filter((r) => r.providerId === id);
2372
+ const runs = successByProvider.get(id) ?? [];
2140
2373
  const avg = average(runs.map((r) => r.raw.latencyMs));
2141
2374
  return { id, avg: avg ?? Infinity };
2142
2375
  }).sort((a, b) => a.avg - b.avg)[0];
@@ -2150,7 +2383,7 @@ function printSummary(results, providers) {
2150
2383
  }
2151
2384
  }
2152
2385
  const byCost = providers.map((id) => {
2153
- const runs = successResults.filter((r) => r.providerId === id);
2386
+ const runs = successByProvider.get(id) ?? [];
2154
2387
  const costs = runs.map((r) => {
2155
2388
  const s = r.scores.find((s2) => s2.name === "cost");
2156
2389
  return s && s.value >= 0 ? s.value : void 0;
@@ -2188,133 +2421,6 @@ function printSummary(results, providers) {
2188
2421
  }
2189
2422
  console.log("");
2190
2423
  }
2191
- function rankProviders(results, providers, scorerName) {
2192
- const ranked = providers.map((id) => {
2193
- const runs = results.filter((r) => r.providerId === id);
2194
- const scores = runs.flatMap((r) => r.scores.filter((s) => s.name === scorerName && s.value >= 0)).map((s) => s.value);
2195
- const avg = scores.length > 0 ? scores.reduce((a, b) => a + b, 0) / scores.length : void 0;
2196
- return { id, avg };
2197
- }).filter((p) => p.avg !== void 0).sort((a, b) => b.avg - a.avg);
2198
- return ranked[0] ? { id: ranked[0].id, avg: ranked[0].avg } : void 0;
2199
- }
2200
- function averageScores(results) {
2201
- const sums = {};
2202
- const counts = {};
2203
- for (const result of results) {
2204
- for (const score of result.scores) {
2205
- if (score.value < 0) continue;
2206
- sums[score.name] = (sums[score.name] ?? 0) + score.value;
2207
- counts[score.name] = (counts[score.name] ?? 0) + 1;
2208
- }
2209
- }
2210
- const avgs = {};
2211
- for (const name of Object.keys(sums)) {
2212
- avgs[name] = sums[name] / counts[name];
2213
- }
2214
- return avgs;
2215
- }
2216
- function averageDetails(results) {
2217
- let costSum = 0;
2218
- let costCount = 0;
2219
- let tokenSum = 0;
2220
- let tokenCount = 0;
2221
- for (const result of results) {
2222
- const costScore = result.scores.find((s) => s.name === "cost");
2223
- const details = costScore?.details;
2224
- if (details?.estimatedUsd != null) {
2225
- costSum += details.estimatedUsd;
2226
- costCount++;
2227
- }
2228
- if (details?.totalTokens != null) {
2229
- tokenSum += details.totalTokens;
2230
- tokenCount++;
2231
- }
2232
- }
2233
- return {
2234
- costUsd: costCount > 0 ? costSum / costCount : void 0,
2235
- totalTokens: tokenCount > 0 ? Math.round(tokenSum / tokenCount) : void 0
2236
- };
2237
- }
2238
- function average(nums) {
2239
- if (nums.length === 0) return void 0;
2240
- return nums.reduce((a, b) => a + b, 0) / nums.length;
2241
- }
2242
- function formatCost(usd) {
2243
- if (usd === void 0) return "\u2014";
2244
- if (usd === 0) return "$0.00";
2245
- if (usd >= 0.01) return `~$${usd.toFixed(2)}`;
2246
- const digits = Math.max(4, -Math.floor(Math.log10(usd)) + 1);
2247
- return `~$${usd.toFixed(digits).replace(/0+$/, "")}`;
2248
- }
2249
- function apiKeyHint(providerId, error) {
2250
- const lower = error.toLowerCase();
2251
- const isAuthError = lower.includes("api key") || lower.includes("401") || lower.includes("unauthorized") || lower.includes("authentication") || lower.includes("incorrect api key") || lower.includes("apikey");
2252
- if (!isAuthError) return void 0;
2253
- const prefix = providerId.split("/")[0];
2254
- switch (prefix) {
2255
- case "openai":
2256
- return "Set: export OPENAI_API_KEY=sk-...";
2257
- case "azure":
2258
- return "Set: export AZURE_OPENAI_API_KEY=... and AZURE_OPENAI_ENDPOINT=...";
2259
- case "anthropic":
2260
- return "Set: export ANTHROPIC_API_KEY=sk-ant-...";
2261
- case "google":
2262
- return "Set: export GOOGLE_API_KEY=...";
2263
- default:
2264
- return `Check the API key for ${providerId}`;
2265
- }
2266
- }
2267
- function providerLabel(providerId) {
2268
- const prefix = providerId.split("/")[0];
2269
- switch (prefix) {
2270
- case "azure":
2271
- return "(OpenAI via Azure)";
2272
- case "openai":
2273
- return "(OpenAI)";
2274
- case "anthropic":
2275
- return "(Anthropic)";
2276
- case "google":
2277
- return "(Google)";
2278
- case "mistral":
2279
- return "(Mistral)";
2280
- case "meta":
2281
- return "(Meta)";
2282
- case "deepseek":
2283
- return "(DeepSeek)";
2284
- case "cohere":
2285
- return "(Cohere)";
2286
- case "qwen":
2287
- return "(Qwen)";
2288
- case "xai":
2289
- return "(xAI)";
2290
- case "minimax":
2291
- return "(MiniMax)";
2292
- case "moonshot":
2293
- return "(Moonshot / Kimi)";
2294
- case "perplexity":
2295
- return "(Perplexity)";
2296
- case "amazon":
2297
- return "(Amazon)";
2298
- case "nvidia":
2299
- return "(NVIDIA)";
2300
- case "microsoft":
2301
- return "(Microsoft)";
2302
- case "ai21":
2303
- return "(AI21 Labs)";
2304
- case "bytedance":
2305
- return "(ByteDance)";
2306
- case "together":
2307
- return "(Together AI)";
2308
- case "fireworks":
2309
- return "(Fireworks AI)";
2310
- case "groq":
2311
- return "(Groq)";
2312
- case "cerebras":
2313
- return "(Cerebras)";
2314
- default:
2315
- return `(${prefix})`;
2316
- }
2317
- }
2318
2424
 
2319
2425
  // src/reporter/json.ts
2320
2426
  function jsonReporter(results) {
@@ -2379,7 +2485,7 @@ function anthropic(model, options) {
2379
2485
  model,
2380
2486
  async run(input) {
2381
2487
  const start = Date.now();
2382
- const systemMessage = input.schema ? "Respond with valid JSON matching the requested schema." : void 0;
2488
+ const systemMessage = input.schema ? SCHEMA_SYSTEM_MESSAGE : void 0;
2383
2489
  const response = await client.messages.create({
2384
2490
  model,
2385
2491
  max_tokens: maxTokens,
@@ -2389,13 +2495,7 @@ function anthropic(model, options) {
2389
2495
  const latencyMs = Date.now() - start;
2390
2496
  const textBlock = response.content.find((b) => b.type === "text");
2391
2497
  const rawContent = textBlock?.type === "text" ? textBlock.text : "";
2392
- let output = rawContent;
2393
- if (input.schema) {
2394
- try {
2395
- output = JSON.parse(rawContent);
2396
- } catch {
2397
- }
2398
- }
2498
+ const output = parseSchemaOutput(rawContent, !!input.schema);
2399
2499
  return {
2400
2500
  output,
2401
2501
  usage: {
@@ -2409,23 +2509,6 @@ function anthropic(model, options) {
2409
2509
  };
2410
2510
  }
2411
2511
 
2412
- // src/providers/gemini.ts
2413
- import OpenAI3 from "openai";
2414
- function gemini(model, options) {
2415
- const apiKey = options?.apiKey ?? process.env.GOOGLE_API_KEY;
2416
- if (!apiKey) {
2417
- throw new Error(
2418
- `Missing API key for google/${model}. Set GOOGLE_API_KEY or pass apiKey option.`
2419
- );
2420
- }
2421
- const client = new OpenAI3({
2422
- apiKey,
2423
- baseURL: "https://generativelanguage.googleapis.com/v1beta/openai/",
2424
- timeout: options?.timeoutMs ?? REQUEST_TIMEOUT_MS
2425
- });
2426
- return makeProvider(`google/${model}`, "Google AI", model, client, model);
2427
- }
2428
-
2429
2512
  // src/reporter/markdown.ts
2430
2513
  var COMMENT_MARKER = "<!-- duelist-ci-report -->";
2431
2514
  function markdownReporter(report, _current) {
@@ -2470,7 +2553,7 @@ function markdownComparisonTable(comparisons) {
2470
2553
  for (const c of comparisons) {
2471
2554
  const baselineStr = c.baseline ? formatStats(c.baseline) : "\u2014";
2472
2555
  const currentStr = formatStats(c.current);
2473
- const deltaStr = c.delta !== null ? formatDelta(c.delta) : "\u2014";
2556
+ const deltaStr = c.delta !== null ? formatDelta(c.delta, 3) : "\u2014";
2474
2557
  const status = statusIndicator(c);
2475
2558
  lines.push(`| ${c.providerId} | ${c.taskName} | ${c.scorerName} | ${baselineStr} | ${currentStr} | ${deltaStr} | ${status} |`);
2476
2559
  }
@@ -2503,10 +2586,6 @@ function formatStats(stats) {
2503
2586
  }
2504
2587
  return stats.mean.toFixed(3);
2505
2588
  }
2506
- function formatDelta(delta) {
2507
- const sign = delta >= 0 ? "+" : "";
2508
- return `${sign}${delta.toFixed(3)}`;
2509
- }
2510
2589
  function statusIndicator(c) {
2511
2590
  if (c.regressed) return "\u{1F534} regressed";
2512
2591
  if (c.improved) return "\u{1F7E2} improved";
@@ -2514,6 +2593,644 @@ function statusIndicator(c) {
2514
2593
  return "\u26AA unchanged";
2515
2594
  }
2516
2595
 
2596
+ // src/reporter/html.ts
2597
+ function esc(s) {
2598
+ return s.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;").replace(/'/g, "&#39;");
2599
+ }
2600
+ function htmlReporter(results) {
2601
+ if (results.length === 0) {
2602
+ return emptyReport();
2603
+ }
2604
+ const { tasks, providers, scorerNames, grouped, byProvider, maxRun } = groupResults(results);
2605
+ const hasCost = scorerNames.includes("cost");
2606
+ const multi = providers.length >= 2;
2607
+ const runsLabel = maxRun > 1 ? `${maxRun} runs each` : "1 run";
2608
+ const taskSections = tasks.map((task) => {
2609
+ const providerData = providers.map((id) => aggregateProviderTask(id, grouped, task));
2610
+ const columnStats = computeColumnStats(providerData, scorerNames);
2611
+ const medals = computeMedals(columnStats, providers);
2612
+ const winnerId = multi ? [...medals.entries()].find(([, m]) => m === "gold")?.[0] : void 0;
2613
+ return { task, providerData, columnStats, medals, winnerId };
2614
+ });
2615
+ const successResults = results.filter((r) => !r.error);
2616
+ const successByProvider = /* @__PURE__ */ new Map();
2617
+ for (const id of providers) {
2618
+ successByProvider.set(id, (byProvider.get(id) ?? []).filter((r) => !r.error));
2619
+ }
2620
+ const correctnessKey = successResults.some(
2621
+ (r) => r.scores.some((s) => s.name === "llm-judge-correctness" && s.value >= 0)
2622
+ ) ? "llm-judge-correctness" : "correctness";
2623
+ const byCorrectness = rankProviders(successByProvider, providers, correctnessKey);
2624
+ const byLatency = providers.map((id) => {
2625
+ const runs = successByProvider.get(id) ?? [];
2626
+ const avg = average(runs.map((r) => r.raw.latencyMs));
2627
+ return { id, avg: avg ?? Infinity };
2628
+ }).sort((a, b) => a.avg - b.avg)[0];
2629
+ const byCost = providers.map((id) => {
2630
+ const runs = successByProvider.get(id) ?? [];
2631
+ const costs = runs.map((r) => {
2632
+ const s = r.scores.find((s2) => s2.name === "cost");
2633
+ return s && s.value >= 0 ? s.value : void 0;
2634
+ }).filter((c) => c !== void 0);
2635
+ const avg = costs.length > 0 ? costs.reduce((a, b) => a + b, 0) / costs.length : void 0;
2636
+ return { id, avg };
2637
+ }).filter((p) => p.avg !== void 0).sort((a, b) => a.avg - b.avg)[0];
2638
+ let overallWinner;
2639
+ if (multi) {
2640
+ const wins = /* @__PURE__ */ new Map();
2641
+ for (const id of providers) wins.set(id, 0);
2642
+ if (byCorrectness) wins.set(byCorrectness.id, (wins.get(byCorrectness.id) ?? 0) + 1);
2643
+ if (byLatency && byLatency.avg !== Infinity) wins.set(byLatency.id, (wins.get(byLatency.id) ?? 0) + 1);
2644
+ if (byCost?.avg !== void 0) wins.set(byCost.id, (wins.get(byCost.id) ?? 0) + 1);
2645
+ const maxWins = Math.max(...wins.values());
2646
+ if (maxWins > 0) {
2647
+ const tops = [...wins.entries()].filter(([, w]) => w === maxWins);
2648
+ if (tops.length === 1) overallWinner = tops[0][0];
2649
+ }
2650
+ }
2651
+ const errorResults = results.filter((r) => r.error);
2652
+ const deduped = dedupeErrors(errorResults);
2653
+ return `<!DOCTYPE html>
2654
+ <html lang="en">
2655
+ <head>
2656
+ <meta charset="UTF-8">
2657
+ <meta name="viewport" content="width=device-width, initial-scale=1">
2658
+ <title>Agent Duelist Report</title>
2659
+ <meta name="description" content="LLM provider benchmark results \u2014 ${providers.length} provider${providers.length !== 1 ? "s" : ""}, ${tasks.length} task${tasks.length !== 1 ? "s" : ""}">
2660
+ <meta property="og:title" content="Agent Duelist Report">
2661
+ <meta property="og:description" content="LLM provider benchmark: ${providers.map(esc).join(" vs ")}">
2662
+ <meta property="og:type" content="website">
2663
+ ${renderStyle()}
2664
+ </head>
2665
+ <body>
2666
+ <div class="bg-mesh"><div class="bg-mesh-extra"></div></div>
2667
+ <div class="report">
2668
+
2669
+ ${renderHeader(runsLabel, providers.length, tasks.length)}
2670
+
2671
+ ${tasks.length > 1 ? renderTabs(tasks) : ""}
2672
+
2673
+ <main>
2674
+ ${taskSections.map((s, i) => renderTaskSection(
2675
+ s.task,
2676
+ s.providerData,
2677
+ s.columnStats,
2678
+ s.medals,
2679
+ s.winnerId,
2680
+ scorerNames,
2681
+ hasCost,
2682
+ multi,
2683
+ i
2684
+ )).join("\n")}
2685
+ </main>
2686
+
2687
+ ${renderSummary(byCorrectness, byLatency, byCost, overallWinner, multi)}
2688
+
2689
+ ${deduped.length > 0 ? renderErrors(deduped) : ""}
2690
+
2691
+ ${renderFooter()}
2692
+
2693
+ </div>
2694
+ ${renderScript(tasks.length)}
2695
+ </body>
2696
+ </html>`;
2697
+ }
2698
+ function emptyReport() {
2699
+ return `<!DOCTYPE html>
2700
+ <html lang="en">
2701
+ <head>
2702
+ <meta charset="UTF-8">
2703
+ <meta name="viewport" content="width=device-width, initial-scale=1">
2704
+ <title>Agent Duelist Report</title>
2705
+ ${renderStyle()}
2706
+ </head>
2707
+ <body>
2708
+ <div class="bg-mesh"><div class="bg-mesh-extra"></div></div>
2709
+ <div class="report">
2710
+ ${renderHeader("0 runs", 0, 0)}
2711
+ <main><p class="empty-msg">No results to display.</p></main>
2712
+ ${renderFooter()}
2713
+ </div>
2714
+ </body>
2715
+ </html>`;
2716
+ }
2717
+ function dedupeErrors(errorResults) {
2718
+ const seen = /* @__PURE__ */ new Map();
2719
+ for (const r of errorResults) {
2720
+ const key = `${r.providerId}::${r.error}`;
2721
+ const existing = seen.get(key);
2722
+ if (existing) {
2723
+ existing.count++;
2724
+ } else {
2725
+ seen.set(key, {
2726
+ providerId: r.providerId,
2727
+ error: r.error ?? "Unknown error",
2728
+ count: 1,
2729
+ hint: apiKeyHint(r.providerId, r.error ?? "")
2730
+ });
2731
+ }
2732
+ }
2733
+ return [...seen.values()];
2734
+ }
2735
+ function renderStyle() {
2736
+ return `<style>
2737
+ :root {
2738
+ --bg: #0f172a;
2739
+ --bg-deep: #020617;
2740
+ --panel: rgba(15, 23, 42, 0.85);
2741
+ --accent: #f59e0b;
2742
+ --accent-soft: rgba(245, 158, 11, 0.15);
2743
+ --text: #e2e8f0;
2744
+ --muted: #94a3b8;
2745
+ --border: rgba(148, 163, 184, 0.15);
2746
+ --green: #22c55e;
2747
+ --red: #ef4444;
2748
+ --yellow: #eab308;
2749
+ --radius: 12px;
2750
+ --mono: ui-monospace, SFMono-Regular, "SF Mono", Menlo, Monaco, Consolas, monospace;
2751
+ --sans: system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
2752
+ }
2753
+ *, *::before, *::after { box-sizing: border-box; margin: 0; padding: 0; }
2754
+ html, body {
2755
+ font-family: var(--sans);
2756
+ background: var(--bg);
2757
+ color: var(--text);
2758
+ min-height: 100vh;
2759
+ }
2760
+ body { padding: 24px; display: flex; justify-content: center; }
2761
+
2762
+ /* Animated gradient mesh */
2763
+ .bg-mesh {
2764
+ position: fixed; inset: 0; z-index: 0;
2765
+ overflow: hidden; pointer-events: none;
2766
+ }
2767
+ .bg-mesh::before, .bg-mesh::after {
2768
+ content: ""; position: absolute; border-radius: 50%;
2769
+ filter: blur(120px); opacity: 0.4;
2770
+ }
2771
+ .bg-mesh::before {
2772
+ width: 600px; height: 600px;
2773
+ background: radial-gradient(circle, rgba(245,158,11,0.18), transparent 70%);
2774
+ top: -10%; left: -5%;
2775
+ animation: meshDrift1 18s ease-in-out infinite alternate;
2776
+ }
2777
+ .bg-mesh::after {
2778
+ width: 500px; height: 500px;
2779
+ background: radial-gradient(circle, rgba(139,92,246,0.12), transparent 70%);
2780
+ bottom: -10%; right: -5%;
2781
+ animation: meshDrift2 22s ease-in-out infinite alternate;
2782
+ }
2783
+ .bg-mesh-extra {
2784
+ position: absolute; width: 400px; height: 400px;
2785
+ border-radius: 50%; filter: blur(100px); opacity: 0.3;
2786
+ background: radial-gradient(circle, rgba(56,189,248,0.12), transparent 70%);
2787
+ top: 50%; left: 60%;
2788
+ animation: meshDrift3 15s ease-in-out infinite alternate;
2789
+ }
2790
+ @keyframes meshDrift1 { from { transform: translate(0,0) scale(1); } to { transform: translate(80px,60px) scale(1.15); } }
2791
+ @keyframes meshDrift2 { from { transform: translate(0,0) scale(1); } to { transform: translate(-60px,-50px) scale(1.1); } }
2792
+ @keyframes meshDrift3 { from { transform: translate(0,0) scale(1); } to { transform: translate(-40px,40px) scale(1.2); } }
2793
+
2794
+ /* Report container */
2795
+ .report {
2796
+ position: relative; z-index: 1;
2797
+ width: 100%; max-width: 960px;
2798
+ }
2799
+
2800
+ /* Header */
2801
+ .report-header {
2802
+ display: flex; justify-content: space-between; align-items: center;
2803
+ padding: 20px 0; margin-bottom: 8px;
2804
+ }
2805
+ .report-brand {
2806
+ display: flex; align-items: center; gap: 10px;
2807
+ text-decoration: none; color: var(--muted);
2808
+ font-weight: 600; font-size: 14px;
2809
+ letter-spacing: 0.04em; text-transform: uppercase;
2810
+ }
2811
+ .report-brand:hover { color: var(--text); }
2812
+ .brand-icon {
2813
+ width: 32px; height: 32px; border-radius: 8px;
2814
+ background: linear-gradient(135deg, var(--accent-soft), rgba(245,158,11,0.05));
2815
+ border: 1px solid rgba(245,158,11,0.3);
2816
+ display: flex; align-items: center; justify-content: center;
2817
+ font-size: 16px;
2818
+ }
2819
+ .report-meta {
2820
+ font-size: 12px; color: var(--muted);
2821
+ text-align: right; line-height: 1.6;
2822
+ }
2823
+
2824
+ /* Task tabs */
2825
+ .task-tabs {
2826
+ display: flex; gap: 6px; margin-bottom: 16px; flex-wrap: wrap;
2827
+ }
2828
+ .task-tab {
2829
+ padding: 6px 16px; border-radius: 999px;
2830
+ border: 1px solid var(--border);
2831
+ background: transparent; color: var(--muted);
2832
+ font-size: 13px; font-weight: 500; cursor: pointer;
2833
+ transition: all 150ms ease;
2834
+ }
2835
+ .task-tab:hover { border-color: rgba(245,158,11,0.3); color: var(--text); }
2836
+ .task-tab.active {
2837
+ background: var(--accent-soft);
2838
+ border-color: rgba(245,158,11,0.4);
2839
+ color: var(--accent);
2840
+ }
2841
+
2842
+ /* Task sections */
2843
+ .task-section { display: none; }
2844
+ .task-section.active { display: block; }
2845
+ .task-name {
2846
+ font-size: 18px; font-weight: 600;
2847
+ margin-bottom: 12px; letter-spacing: -0.01em;
2848
+ }
2849
+
2850
+ /* Results table */
2851
+ .results-table {
2852
+ width: 100%; border-collapse: collapse;
2853
+ font-size: 13px; margin-bottom: 16px;
2854
+ border-radius: var(--radius); overflow: hidden;
2855
+ border: 1px solid var(--border);
2856
+ }
2857
+ .results-table th, .results-table td {
2858
+ padding: 10px 14px;
2859
+ text-align: left;
2860
+ border-bottom: 1px solid var(--border);
2861
+ }
2862
+ .results-table th {
2863
+ background: rgba(0,0,0,0.3);
2864
+ font-size: 11px; font-weight: 600;
2865
+ text-transform: uppercase; letter-spacing: 0.05em;
2866
+ color: var(--muted); cursor: pointer;
2867
+ user-select: none; white-space: nowrap;
2868
+ }
2869
+ .results-table th:hover { color: var(--text); }
2870
+ .results-table th .sort-arrow { margin-left: 4px; font-size: 10px; }
2871
+ .results-table tbody tr {
2872
+ background: var(--panel);
2873
+ transition: background 120ms ease;
2874
+ }
2875
+ .results-table tbody tr:hover { background: rgba(15,23,42,0.95); }
2876
+ .results-table tbody tr:last-child td { border-bottom: none; }
2877
+
2878
+ /* Score cell with progress bar */
2879
+ .score-cell { position: relative; min-width: 90px; }
2880
+ .score-bar {
2881
+ position: absolute; left: 0; bottom: 0;
2882
+ height: 3px; border-radius: 2px;
2883
+ transition: width 300ms ease;
2884
+ }
2885
+ .score-val { position: relative; z-index: 1; font-family: var(--mono); font-size: 12px; }
2886
+
2887
+ /* Color ranking */
2888
+ .rank-best { color: var(--green); font-weight: 600; }
2889
+ .rank-worst { color: var(--red); }
2890
+ .rank-mid { color: var(--yellow); }
2891
+ .rank-neutral { color: var(--text); }
2892
+ .rank-error { color: var(--muted); }
2893
+
2894
+ /* Winner banner */
2895
+ .task-winner {
2896
+ display: flex; align-items: center; gap: 10px;
2897
+ padding: 12px 18px; margin-bottom: 20px;
2898
+ border-radius: var(--radius);
2899
+ background: linear-gradient(135deg, rgba(34,197,94,0.08), rgba(34,197,94,0.02));
2900
+ border: 1px solid rgba(34,197,94,0.2);
2901
+ font-size: 14px; font-weight: 500;
2902
+ }
2903
+ .task-winner .trophy { font-size: 20px; }
2904
+ .task-winner .winner-name { color: var(--green); font-weight: 600; }
2905
+ .task-winner .winner-label { color: var(--muted); font-size: 12px; margin-left: 4px; }
2906
+
2907
+ /* Summary cards */
2908
+ .summary-section { margin-top: 32px; }
2909
+ .summary-title {
2910
+ font-size: 16px; font-weight: 600;
2911
+ margin-bottom: 12px; color: var(--text);
2912
+ }
2913
+ .summary-cards {
2914
+ display: grid;
2915
+ grid-template-columns: repeat(auto-fill, minmax(200px, 1fr));
2916
+ gap: 12px;
2917
+ }
2918
+ .summary-card {
2919
+ padding: 16px; border-radius: var(--radius);
2920
+ border: 1px solid var(--border);
2921
+ background: var(--panel);
2922
+ }
2923
+ .summary-card .card-label {
2924
+ font-size: 11px; font-weight: 600;
2925
+ text-transform: uppercase; letter-spacing: 0.05em;
2926
+ color: var(--muted); margin-bottom: 6px;
2927
+ }
2928
+ .summary-card .card-value {
2929
+ font-size: 20px; font-weight: 700;
2930
+ color: var(--green); font-family: var(--mono);
2931
+ }
2932
+ .summary-card .card-provider {
2933
+ font-size: 12px; color: var(--muted); margin-top: 4px;
2934
+ }
2935
+
2936
+ /* Errors */
2937
+ .errors-section { margin-top: 24px; }
2938
+ .errors-title {
2939
+ font-size: 16px; font-weight: 600;
2940
+ margin-bottom: 8px; color: var(--red);
2941
+ cursor: pointer;
2942
+ }
2943
+ .errors-list {
2944
+ border-radius: var(--radius);
2945
+ border: 1px solid rgba(239,68,68,0.2);
2946
+ background: rgba(239,68,68,0.04);
2947
+ overflow: hidden;
2948
+ }
2949
+ .error-item {
2950
+ padding: 10px 16px;
2951
+ border-bottom: 1px solid rgba(239,68,68,0.1);
2952
+ font-size: 13px;
2953
+ }
2954
+ .error-item:last-child { border-bottom: none; }
2955
+ .error-provider { font-weight: 600; color: var(--text); }
2956
+ .error-msg { color: var(--muted); margin-left: 8px; }
2957
+ .error-count { color: var(--muted); font-size: 11px; }
2958
+ .error-hint { color: var(--muted); font-size: 12px; margin-top: 4px; font-style: italic; }
2959
+
2960
+ /* Footer */
2961
+ .report-footer {
2962
+ margin-top: 40px; padding: 20px 0;
2963
+ border-top: 1px solid var(--border);
2964
+ display: flex; justify-content: space-between; align-items: center;
2965
+ flex-wrap: wrap; gap: 12px;
2966
+ }
2967
+ .footer-brand {
2968
+ font-size: 13px; color: var(--muted);
2969
+ }
2970
+ .footer-brand a {
2971
+ color: var(--accent); text-decoration: none; font-weight: 500;
2972
+ }
2973
+ .footer-brand a:hover { text-decoration: underline; }
2974
+ .footer-cta {
2975
+ display: inline-flex; align-items: center; gap: 6px;
2976
+ padding: 6px 14px; border-radius: 8px;
2977
+ background: var(--accent-soft);
2978
+ border: 1px solid rgba(245,158,11,0.3);
2979
+ color: var(--accent); font-size: 12px; font-weight: 500;
2980
+ text-decoration: none;
2981
+ transition: transform 120ms ease, box-shadow 120ms ease;
2982
+ }
2983
+ .footer-cta:hover { transform: translateY(-1px); box-shadow: 0 4px 12px rgba(245,158,11,0.2); }
2984
+
2985
+ /* Empty state */
2986
+ .empty-msg {
2987
+ text-align: center; color: var(--muted);
2988
+ padding: 60px 20px; font-size: 16px;
2989
+ }
2990
+
2991
+ /* Responsive */
2992
+ @media (max-width: 640px) {
2993
+ body { padding: 12px; }
2994
+ .report-header { flex-direction: column; align-items: flex-start; gap: 8px; }
2995
+ .report-meta { text-align: left; }
2996
+ .summary-cards { grid-template-columns: 1fr; }
2997
+ .results-table { font-size: 12px; }
2998
+ .results-table th, .results-table td { padding: 8px 10px; }
2999
+ .report-footer { flex-direction: column; align-items: flex-start; }
3000
+ }
3001
+ </style>`;
3002
+ }
3003
+ function renderHeader(runsLabel, providerCount, taskCount) {
3004
+ const now = (/* @__PURE__ */ new Date()).toISOString().replace("T", " ").slice(0, 19) + " UTC";
3005
+ return `<header class="report-header">
3006
+ <a class="report-brand" href="https://github.com/DataGobes/agent-duelist" target="_blank" rel="noopener">
3007
+ <div class="brand-icon">&#x2B21;</div>
3008
+ <span>Agent Duelist</span>
3009
+ </a>
3010
+ <div class="report-meta">
3011
+ ${providerCount} provider${providerCount !== 1 ? "s" : ""} &middot;
3012
+ ${taskCount} task${taskCount !== 1 ? "s" : ""} &middot;
3013
+ ${esc(runsLabel)}<br>
3014
+ ${esc(now)}
3015
+ </div>
3016
+ </header>`;
3017
+ }
3018
+ function renderTabs(tasks) {
3019
+ const buttons = tasks.map(
3020
+ (t, i) => `<button class="task-tab${i === 0 ? " active" : ""}" data-task="${i}">${esc(t)}</button>`
3021
+ ).join("\n ");
3022
+ return `<nav class="task-tabs">
3023
+ ${buttons}
3024
+ </nav>`;
3025
+ }
3026
+ function renderTaskSection(task, providerData, columnStats, medals, winnerId, scorerNames, _hasCost, multi, index) {
3027
+ const cols = [
3028
+ { label: "Provider", key: "provider", isScore: false }
3029
+ ];
3030
+ for (const name of scorerNames) {
3031
+ if (name === "latency") {
3032
+ cols.push({ label: "Latency", key: "latency", isScore: false });
3033
+ } else if (name === "cost") {
3034
+ cols.push({ label: "Cost", key: "cost", isScore: false });
3035
+ cols.push({ label: "Tokens", key: "tokens", isScore: false });
3036
+ } else {
3037
+ cols.push({ label: scorerLabel(name), key: name, isScore: true });
3038
+ }
3039
+ }
3040
+ const ths = cols.map(
3041
+ (c) => `<th data-col="${esc(c.key)}">${esc(c.label)}<span class="sort-arrow"></span></th>`
3042
+ ).join("");
3043
+ const rows = providerData.map((pd) => {
3044
+ const medal = medalEmoji(medals.get(pd.providerId) ?? "none");
3045
+ const cells = [];
3046
+ const medalHtml = medal ? `${medal} ` : "";
3047
+ cells.push(`<td>${medalHtml}${esc(pd.providerId)}</td>`);
3048
+ if (pd.allErrors) {
3049
+ for (let ci = 1; ci < cols.length; ci++) {
3050
+ cells.push(`<td class="rank-error">&mdash;</td>`);
3051
+ }
3052
+ } else {
3053
+ for (const col of cols.slice(1)) {
3054
+ cells.push(renderDataCell(col.key, col.isScore, pd, columnStats, multi));
3055
+ }
3056
+ }
3057
+ return `<tr>${cells.join("")}</tr>`;
3058
+ }).join("\n");
3059
+ const winnerHtml = winnerId ? `<div class="task-winner">
3060
+ <span class="trophy">&#x1F3C6;</span>
3061
+ <span>Winner: <span class="winner-name">${esc(winnerId)}</span>
3062
+ <span class="winner-label">${esc(providerLabel(winnerId))}</span></span>
3063
+ </div>` : "";
3064
+ return `<section class="task-section${index === 0 ? " active" : ""}" data-task-idx="${index}">
3065
+ <h2 class="task-name">${esc(task)}</h2>
3066
+ <table class="results-table">
3067
+ <thead><tr>${ths}</tr></thead>
3068
+ <tbody>${rows}</tbody>
3069
+ </table>
3070
+ ${winnerHtml}
3071
+ </section>`;
3072
+ }
3073
+ function renderDataCell(key, _isScore, pd, columnStats, multi) {
3074
+ const colStats = columnStats.get(key);
3075
+ if (key === "latency") {
3076
+ const ms = pd.latencyMs;
3077
+ if (ms === void 0) return `<td class="rank-error">&mdash;</td>`;
3078
+ const rankClass = multi && colStats ? rankClass_(ms, colStats) : "rank-neutral";
3079
+ return `<td class="${rankClass}" data-sort-val="${ms}">${Math.round(ms)}ms</td>`;
3080
+ }
3081
+ if (key === "cost") {
3082
+ const cost = pd.avgDetails.costUsd;
3083
+ if (cost === void 0) return `<td class="rank-error">&mdash;</td>`;
3084
+ const rankClass = multi && colStats ? rankClass_(cost, colStats) : "rank-neutral";
3085
+ return `<td class="${rankClass}" data-sort-val="${cost}">${esc(formatCost(cost))}</td>`;
3086
+ }
3087
+ if (key === "tokens") {
3088
+ const tokens = pd.avgDetails.totalTokens;
3089
+ if (tokens === void 0) return `<td class="rank-error">&mdash;</td>`;
3090
+ const rankClass = multi && colStats ? rankClass_(tokens, colStats) : "rank-neutral";
3091
+ return `<td class="${rankClass}" data-sort-val="${tokens}">${tokens}</td>`;
3092
+ }
3093
+ const val = pd.avgScores[key];
3094
+ if (val === void 0) return `<td class="rank-error">&mdash;</td>`;
3095
+ const pct = Math.round(val * 100);
3096
+ let rankCls;
3097
+ if (multi && colStats) {
3098
+ rankCls = rankClass_(val, colStats);
3099
+ } else {
3100
+ rankCls = val >= 0.8 ? "rank-best" : val >= 0.5 ? "rank-mid" : "rank-worst";
3101
+ }
3102
+ const barColor = val >= 0.8 ? "var(--green)" : val >= 0.5 ? "var(--yellow)" : "var(--red)";
3103
+ return `<td class="score-cell ${rankCls}" data-sort-val="${val}">
3104
+ <span class="score-val">${pct}%</span>
3105
+ <div class="score-bar" style="width:${pct}%;background:${barColor}"></div>
3106
+ </td>`;
3107
+ }
3108
+ function rankClass_(value, colStats) {
3109
+ if (colStats.best === void 0 || colStats.worst === void 0) return "rank-neutral";
3110
+ if (colStats.best === colStats.worst) return "rank-neutral";
3111
+ if (value === colStats.best) return "rank-best";
3112
+ if (value === colStats.worst) return "rank-worst";
3113
+ return "rank-mid";
3114
+ }
3115
+ function renderSummary(byCorrectness, byLatency, byCost, overallWinner, multi) {
3116
+ const cards = [];
3117
+ if (byCorrectness) {
3118
+ const pct = `${Math.round(byCorrectness.avg * 100)}%`;
3119
+ const provider = multi ? `<div class="card-provider">${esc(byCorrectness.id)} ${esc(providerLabel(byCorrectness.id))}</div>` : "";
3120
+ cards.push(`<div class="summary-card">
3121
+ <div class="card-label">${multi ? "Most Correct" : "Avg Correctness"}</div>
3122
+ <div class="card-value">${pct}</div>
3123
+ ${provider}
3124
+ </div>`);
3125
+ }
3126
+ if (byLatency && byLatency.avg !== Infinity) {
3127
+ const ms = `${Math.round(byLatency.avg)}ms`;
3128
+ const provider = multi ? `<div class="card-provider">${esc(byLatency.id)} ${esc(providerLabel(byLatency.id))}</div>` : "";
3129
+ cards.push(`<div class="summary-card">
3130
+ <div class="card-label">${multi ? "Fastest" : "Avg Latency"}</div>
3131
+ <div class="card-value">${ms}</div>
3132
+ ${provider}
3133
+ </div>`);
3134
+ }
3135
+ if (byCost?.avg !== void 0) {
3136
+ const cost = esc(formatCost(byCost.avg));
3137
+ const provider = multi ? `<div class="card-provider">${esc(byCost.id)} ${esc(providerLabel(byCost.id))}</div>` : "";
3138
+ cards.push(`<div class="summary-card">
3139
+ <div class="card-label">${multi ? "Cheapest" : "Avg Cost"}</div>
3140
+ <div class="card-value">${cost}</div>
3141
+ ${provider}
3142
+ </div>`);
3143
+ }
3144
+ if (overallWinner) {
3145
+ cards.push(`<div class="summary-card">
3146
+ <div class="card-label">Overall Winner</div>
3147
+ <div class="card-value">&#x1F3C6;</div>
3148
+ <div class="card-provider">${esc(overallWinner)} ${esc(providerLabel(overallWinner))}</div>
3149
+ </div>`);
3150
+ }
3151
+ if (cards.length === 0) return "";
3152
+ return `<section class="summary-section">
3153
+ <h2 class="summary-title">Summary</h2>
3154
+ <div class="summary-cards">
3155
+ ${cards.join("\n ")}
3156
+ </div>
3157
+ </section>`;
3158
+ }
3159
+ function renderErrors(errors) {
3160
+ const items = errors.map((e) => {
3161
+ const suffix = e.count > 1 ? ` <span class="error-count">(&times;${e.count})</span>` : "";
3162
+ const hint = e.hint ? `<div class="error-hint">${esc(e.hint)}</div>` : "";
3163
+ return `<div class="error-item">
3164
+ <span class="error-provider">${esc(e.providerId)}:</span>
3165
+ <span class="error-msg">${esc(e.error)}</span>${suffix}
3166
+ ${hint}
3167
+ </div>`;
3168
+ }).join("\n");
3169
+ return `<section class="errors-section">
3170
+ <h2 class="errors-title" onclick="this.nextElementSibling.style.display=this.nextElementSibling.style.display==='none'?'block':'block'">Errors</h2>
3171
+ <div class="errors-list">
3172
+ ${items}
3173
+ </div>
3174
+ </section>`;
3175
+ }
3176
+ function renderFooter() {
3177
+ return `<footer class="report-footer">
3178
+ <div class="footer-brand">
3179
+ Powered by <a href="https://github.com/DataGobes/agent-duelist" target="_blank" rel="noopener">Agent Duelist</a>
3180
+ </div>
3181
+ <a class="footer-cta" href="https://github.com/DataGobes/agent-duelist" target="_blank" rel="noopener">
3182
+ &#x2B50; Star on GitHub
3183
+ </a>
3184
+ </footer>`;
3185
+ }
3186
+ function renderScript(taskCount) {
3187
+ return `<script>
3188
+ (function() {
3189
+ /* Tab switching */
3190
+ ${taskCount > 1 ? `
3191
+ var tabs = document.querySelectorAll('.task-tab');
3192
+ var sections = document.querySelectorAll('.task-section');
3193
+ tabs.forEach(function(tab) {
3194
+ tab.addEventListener('click', function() {
3195
+ var idx = parseInt(tab.getAttribute('data-task'));
3196
+ tabs.forEach(function(t) { t.classList.remove('active'); });
3197
+ sections.forEach(function(s) { s.classList.remove('active'); });
3198
+ tab.classList.add('active');
3199
+ sections[idx].classList.add('active');
3200
+ });
3201
+ });` : ""}
3202
+
3203
+ /* Column sorting */
3204
+ document.querySelectorAll('.results-table th').forEach(function(th, colIdx) {
3205
+ var table = th.closest('table');
3206
+ var asc = true;
3207
+ th.addEventListener('click', function() {
3208
+ var tbody = table.querySelector('tbody');
3209
+ var rows = Array.from(tbody.querySelectorAll('tr'));
3210
+ rows.sort(function(a, b) {
3211
+ var aCell = a.children[colIdx];
3212
+ var bCell = b.children[colIdx];
3213
+ var aVal = aCell.getAttribute('data-sort-val');
3214
+ var bVal = bCell.getAttribute('data-sort-val');
3215
+ if (aVal !== null && bVal !== null) {
3216
+ return asc ? parseFloat(aVal) - parseFloat(bVal) : parseFloat(bVal) - parseFloat(aVal);
3217
+ }
3218
+ var aText = aCell.textContent || '';
3219
+ var bText = bCell.textContent || '';
3220
+ return asc ? aText.localeCompare(bText) : bText.localeCompare(aText);
3221
+ });
3222
+ rows.forEach(function(row) { tbody.appendChild(row); });
3223
+
3224
+ /* Update sort arrows */
3225
+ table.querySelectorAll('th .sort-arrow').forEach(function(a) { a.textContent = ''; });
3226
+ th.querySelector('.sort-arrow').textContent = asc ? ' \\u25B2' : ' \\u25BC';
3227
+ asc = !asc;
3228
+ });
3229
+ });
3230
+ })();
3231
+ </script>`;
3232
+ }
3233
+
2517
3234
  // src/ci.ts
2518
3235
  import { readFileSync, writeFileSync, mkdirSync } from "fs";
2519
3236
  import { dirname } from "path";
@@ -2535,10 +3252,11 @@ var T_CRITICAL_95 = {
2535
3252
  25: 2.06,
2536
3253
  30: 2.042
2537
3254
  };
3255
+ var T_CRITICAL_KEYS = Object.keys(T_CRITICAL_95).map(Number).sort((a, b) => a - b);
2538
3256
  function tCritical(df) {
2539
3257
  if (df <= 0) return 1.96;
2540
3258
  if (T_CRITICAL_95[df] !== void 0) return T_CRITICAL_95[df];
2541
- const keys = Object.keys(T_CRITICAL_95).map(Number).sort((a, b) => a - b);
3259
+ const keys = T_CRITICAL_KEYS;
2542
3260
  if (df > keys[keys.length - 1]) return 1.96;
2543
3261
  for (let i = 0; i < keys.length - 1; i++) {
2544
3262
  if (df > keys[i] && df < keys[i + 1]) {
@@ -2648,7 +3366,7 @@ function compareResults(baselineStats, currentStats, thresholds, budget, current
2648
3366
  if (regressions.length > 0) {
2649
3367
  for (const r of regressions) {
2650
3368
  failureReasons.push(
2651
- `${r.providerId} \xD7 ${r.taskName}: ${r.scorerName} regressed by ${formatDelta2(r.delta)}`
3369
+ `${r.providerId} \xD7 ${r.taskName}: ${r.scorerName} regressed by ${formatDelta(r.delta)}`
2652
3370
  );
2653
3371
  }
2654
3372
  }
@@ -2683,10 +3401,6 @@ function detectImprovement(baseline, current, threshold, lowerIsBetter) {
2683
3401
  }
2684
3402
  return current.ci95Lower - baseline.ci95Upper > threshold;
2685
3403
  }
2686
- function formatDelta2(delta) {
2687
- const sign = delta >= 0 ? "+" : "";
2688
- return `${sign}${delta.toFixed(4)}`;
2689
- }
2690
3404
  function loadBaseline(path) {
2691
3405
  try {
2692
3406
  const raw = readFileSync(path, "utf-8");
@@ -2743,18 +3457,20 @@ function detectGitHubContext() {
2743
3457
  return { token, owner, repo, prNumber };
2744
3458
  }
2745
3459
  var API_BASE = "https://api.github.com";
3460
+ function ghHeaders(token, extra) {
3461
+ return {
3462
+ Authorization: `Bearer ${token}`,
3463
+ Accept: "application/vnd.github+json",
3464
+ "X-GitHub-Api-Version": "2022-11-28",
3465
+ ...extra
3466
+ };
3467
+ }
2746
3468
  async function findExistingComment(ctx, marker) {
2747
3469
  let page = 1;
2748
3470
  const perPage = 50;
2749
3471
  while (true) {
2750
3472
  const url = `${API_BASE}/repos/${ctx.owner}/${ctx.repo}/issues/${ctx.prNumber}/comments?per_page=${perPage}&page=${page}`;
2751
- const res = await fetch(url, {
2752
- headers: {
2753
- Authorization: `Bearer ${ctx.token}`,
2754
- Accept: "application/vnd.github+json",
2755
- "X-GitHub-Api-Version": "2022-11-28"
2756
- }
2757
- });
3473
+ const res = await fetch(url, { headers: ghHeaders(ctx.token) });
2758
3474
  if (!res.ok) return null;
2759
3475
  const comments = await res.json();
2760
3476
  if (comments.length === 0) break;
@@ -2774,12 +3490,7 @@ async function upsertPrComment(ctx, body, marker) {
2774
3490
  const url = `${API_BASE}/repos/${ctx.owner}/${ctx.repo}/issues/comments/${existingId}`;
2775
3491
  const res = await fetch(url, {
2776
3492
  method: "PATCH",
2777
- headers: {
2778
- Authorization: `Bearer ${ctx.token}`,
2779
- Accept: "application/vnd.github+json",
2780
- "Content-Type": "application/json",
2781
- "X-GitHub-Api-Version": "2022-11-28"
2782
- },
3493
+ headers: ghHeaders(ctx.token, { "Content-Type": "application/json" }),
2783
3494
  body: JSON.stringify({ body })
2784
3495
  });
2785
3496
  if (!res.ok) {
@@ -2790,12 +3501,7 @@ async function upsertPrComment(ctx, body, marker) {
2790
3501
  const url = `${API_BASE}/repos/${ctx.owner}/${ctx.repo}/issues/${ctx.prNumber}/comments`;
2791
3502
  const res = await fetch(url, {
2792
3503
  method: "POST",
2793
- headers: {
2794
- Authorization: `Bearer ${ctx.token}`,
2795
- Accept: "application/vnd.github+json",
2796
- "Content-Type": "application/json",
2797
- "X-GitHub-Api-Version": "2022-11-28"
2798
- },
3504
+ headers: ghHeaders(ctx.token, { "Content-Type": "application/json" }),
2799
3505
  body: JSON.stringify({ body })
2800
3506
  });
2801
3507
  if (!res.ok) {
@@ -2813,6 +3519,7 @@ export {
2813
3519
  defineArena,
2814
3520
  detectGitHubContext,
2815
3521
  gemini,
3522
+ htmlReporter,
2816
3523
  jsonReporter,
2817
3524
  loadBaseline,
2818
3525
  markdownReporter,