agent-duelist 0.1.0 → 0.2.0

package/dist/cli.js

@@ -49,14 +49,154 @@ function bold(s5) {
 function dim(s5) {
   return `${dimCode}${s5}${reset}`;
 }
-function colorScore(value) {
-  const pct = Math.round(value * 100);
-  const str = `${pct}%`;
-  if (value >= 0.8) return `${green}${str}${reset}`;
-  if (value >= 0.5) return `${yellow}${str}${reset}`;
-  return `${red}${str}${reset}`;
-}
-function consoleReporter(results) {
+function stripAnsi(s5) {
+  return s5.replace(/\x1b\[[0-9;]*m/g, "");
+}
+function displayWidth(s5) {
+  const stripped = stripAnsi(s5);
+  let width = 0;
+  for (const ch of stripped) {
+    const code = ch.codePointAt(0) ?? 0;
+    if (code >= 126976) width += 2;
+    else if (code >= 9728 && code <= 10175) width += 2;
+    else width += 1;
+  }
+  return width;
+}
+function padCell(str, targetWidth, align) {
+  const dw = displayWidth(str);
+  const padding = Math.max(0, targetWidth - dw);
+  if (align === "right") return " ".repeat(padding) + str;
+  return str + " ".repeat(padding);
+}
+function sparkBar(ratio, width = 8) {
+  const clamped = Math.max(0, Math.min(1, ratio));
+  const fillLen = Math.round(clamped * width);
+  const fill = "\u2593".repeat(fillLen);
+  const track = "\u2591".repeat(width - fillLen);
+  return { fill, track };
+}
+function drawTableLine(widths, position) {
+  const totalInner = widths.reduce((sum, w4) => sum + w4 + 2, 0) + widths.length - 1;
+  if (position === "bottom") {
+    return dim(`\u2514${"\u2500".repeat(totalInner)}\u2518`);
+  }
+  if (position === "merge") {
+    return dim(`\u251C${"\u2500".repeat(totalInner)}\u2524`);
+  }
+  const segments = widths.map((w4) => "\u2500".repeat(w4 + 2));
+  if (position === "top") {
+    return dim(`\u250C${segments.join("\u252C")}\u2510`);
+  }
+  return dim(`\u251C${segments.join("\u253C")}\u2524`);
+}
+function drawTableRow(cells, widths, aligns) {
+  const parts = cells.map(
+    (cell, i7) => " " + padCell(cell, widths[i7], aligns[i7]) + " "
+  );
+  return dim("\u2502") + parts.join(dim("\u2502")) + dim("\u2502");
+}
+function drawSpanRow(content, widths) {
+  const totalInner = widths.reduce((sum, w4) => sum + w4 + 2, 0) + widths.length - 1;
+  const dw = displayWidth(content);
+  const padding = Math.max(0, totalInner - dw - 1);
+  return dim("\u2502") + " " + content + " ".repeat(padding) + dim("\u2502");
+}
+function computeColumnStats(providerData, scorerNames) {
+  const stats = /* @__PURE__ */ new Map();
+  const valid = providerData.filter((p5) => !p5.allErrors);
+  if (scorerNames.includes("latency")) {
+    const values = /* @__PURE__ */ new Map();
+    for (const p5 of providerData) {
+      values.set(p5.providerId, p5.allErrors ? void 0 : p5.latencyMs);
+    }
+    const nums = valid.map((p5) => p5.latencyMs).filter((v4) => v4 !== void 0);
+    stats.set("latency", {
+      values,
+      best: nums.length > 0 ? Math.min(...nums) : void 0,
+      worst: nums.length > 0 ? Math.max(...nums) : void 0
+    });
+  }
+  if (scorerNames.includes("cost")) {
+    const costValues = /* @__PURE__ */ new Map();
+    const tokenValues = /* @__PURE__ */ new Map();
+    for (const p5 of providerData) {
+      costValues.set(p5.providerId, p5.allErrors ? void 0 : p5.avgDetails.costUsd);
+      tokenValues.set(p5.providerId, p5.allErrors ? void 0 : p5.avgDetails.totalTokens);
+    }
+    const costNums = valid.map((p5) => p5.avgDetails.costUsd).filter((v4) => v4 !== void 0);
+    const tokenNums = valid.map((p5) => p5.avgDetails.totalTokens).filter((v4) => v4 !== void 0);
+    stats.set("cost", {
+      values: costValues,
+      best: costNums.length > 0 ? Math.min(...costNums) : void 0,
+      worst: costNums.length > 0 ? Math.max(...costNums) : void 0
+    });
+    stats.set("tokens", {
+      values: tokenValues,
+      best: tokenNums.length > 0 ? Math.min(...tokenNums) : void 0,
+      worst: tokenNums.length > 0 ? Math.max(...tokenNums) : void 0
+    });
+  }
+  for (const name of scorerNames) {
+    if (name === "latency" || name === "cost") continue;
+    const values = /* @__PURE__ */ new Map();
+    for (const p5 of providerData) {
+      values.set(p5.providerId, p5.allErrors ? void 0 : p5.avgScores[name]);
+    }
+    const nums = valid.map((p5) => p5.avgScores[name]).filter((v4) => v4 !== void 0);
+    stats.set(name, {
+      values,
+      best: nums.length > 0 ? Math.max(...nums) : void 0,
+      worst: nums.length > 0 ? Math.min(...nums) : void 0
+    });
+  }
+  return stats;
+}
+function colorByRank(text, value, colStats, providerCount) {
+  if (value === void 0) return dim("\u2014");
+  if (providerCount < 2) return text;
+  if (colStats.best === void 0 || colStats.worst === void 0) return text;
+  if (colStats.best === colStats.worst) return text;
+  if (value === colStats.best) return `${brightGreen}${boldCode}${text}${reset}`;
+  if (value === colStats.worst) return `${red}${text}${reset}`;
+  return `${yellow}${text}${reset}`;
+}
+function computeMedals(columnStats, providerIds) {
+  const medals = /* @__PURE__ */ new Map();
+  if (providerIds.length < 2) {
+    for (const id of providerIds) medals.set(id, "");
+    return medals;
+  }
+  const wins = /* @__PURE__ */ new Map();
+  for (const id of providerIds) wins.set(id, 0);
+  for (const [, colStats] of columnStats) {
+    if (colStats.best === void 0) continue;
+    for (const [providerId, value] of colStats.values) {
+      if (value !== void 0 && value === colStats.best) {
+        wins.set(providerId, (wins.get(providerId) ?? 0) + 1);
+      }
+    }
+  }
+  const totalWins = [...wins.values()].reduce((a7, b3) => a7 + b3, 0);
+  if (totalWins === 0) {
+    for (const id of providerIds) medals.set(id, "");
+    return medals;
+  }
+  const sorted = [...wins.entries()].sort(
+    (a7, b3) => b3[1] - a7[1] || a7[0].localeCompare(b3[0])
+  );
+  const medalList = ["\u{1F947}", "\u{1F948}", "\u{1F949}"];
+  let rank = 0;
+  for (let i7 = 0; i7 < sorted.length; i7++) {
+    if (i7 > 0 && sorted[i7][1] < sorted[i7 - 1][1]) {
+      rank = i7;
+    }
+    medals.set(sorted[i7][0], rank < medalList.length ? medalList[rank] : "");
+  }
+  return medals;
+}
+function consoleReporter(results, options) {
+  const showSparklines = options?.sparklines ?? true;
   if (results.length === 0) {
     console.log("\nNo results to display.\n");
     return;
@@ -66,78 +206,155 @@ function consoleReporter(results) {
   const scorerNames = [...new Set(results.flatMap((r3) => r3.scores.map((s5) => s5.name)))];
   const hasCost = scorerNames.includes("cost");
   const hasErrors = results.some((r3) => r3.error);
+  const multi = providers.length >= 2;
   const runsPerCell = Math.max(...results.map((r3) => r3.run));
-  const runLabel = runsPerCell > 1 ? ` (${runsPerCell} runs each)` : "";
+  const runLabel = runsPerCell > 1 ? `  ${dim(`(${runsPerCell} runs each)`)}` : "";
   console.log("");
-  console.log(`  ${bold(`\u2B21 Agent Duelist Results${runLabel}`)}`);
-  console.log(`  ${dim("\u2500".repeat(70))}`);
+  console.log(`  ${brightWhite}${boldCode}\u2B21  Agent Duelist${reset}${runLabel}`);
+  console.log(`  ${dim("\u2501".repeat(72))}`);
   console.log("");
   for (const task of tasks) {
     console.log(`  ${bold(`Task: ${task}`)}`);
-    const cols = [{ label: "Provider", width: 22, align: "left" }];
-    for (const name of scorerNames) {
-      if (name === "latency") cols.push({ label: "Latency", width: 10, align: "right" });
-      else if (name === "cost") {
-        cols.push({ label: "Cost", width: 12, align: "right" });
-        cols.push({ label: "Tokens", width: 9, align: "right" });
-      } else if (name === "correctness") cols.push({ label: "Match", width: 8, align: "right" });
-      else if (name === "schema-correctness") cols.push({ label: "Schema", width: 8, align: "right" });
-      else if (name === "fuzzy-similarity") cols.push({ label: "Fuzzy", width: 8, align: "right" });
-      else if (name === "llm-judge-correctness") cols.push({ label: "Judge", width: 8, align: "right" });
-      else if (name === "tool-usage") cols.push({ label: "Tool", width: 8, align: "right" });
-      else cols.push({ label: name, width: 10, align: "right" });
-    }
-    if (hasErrors) cols.push({ label: "Status", width: 8, align: "left" });
-    const totalWidth = cols.reduce((sum, c3) => sum + c3.width + 2, 0);
-    console.log(`  ${dim(cols.map((c3) => pad(c3.label, c3.width + 2, c3.align)).join(""))}`);
-    console.log(`  ${dim("\u2500".repeat(totalWidth))}`);
-    for (const provider of providers) {
-      const taskResults = results.filter(
-        (r3) => r3.taskName === task && r3.providerId === provider
-      );
+    console.log("");
+    const providerData = providers.map((providerId) => {
+      const taskResults = results.filter((r3) => r3.taskName === task && r3.providerId === providerId);
       const errorResults2 = taskResults.filter((r3) => r3.error);
       const successResults = taskResults.filter((r3) => !r3.error);
-      if (successResults.length === 0 && errorResults2.length > 0) {
-        const cells2 = [pad(provider, 24, "left")];
-        for (const name of scorerNames) {
-          if (name === "cost") {
-            cells2.push(pad("\u2014", 14, "right"));
-            cells2.push(pad("\u2014", 11, "right"));
-          } else cells2.push(pad("\u2014", cols.find((c3) => c3.label !== "Provider").width + 2, "right"));
-        }
-        if (hasErrors) cells2.push(`  ${red}FAIL${reset}`);
-        console.log(`  ${cells2.join("")}`);
-        continue;
+      if (successResults.length === 0) {
+        return {
+          providerId,
+          avgScores: {},
+          avgDetails: { costUsd: void 0, totalTokens: void 0 },
+          latencyMs: void 0,
+          allErrors: errorResults2.length > 0,
+          errorCount: errorResults2.length
+        };
       }
-      const avgScores = averageScores(successResults);
-      const avgDetails = averageDetails(successResults);
-      const latencyMs = average(successResults.map((r3) => r3.raw.latencyMs));
-      const cells = [pad(provider, 24, "left")];
-      for (const name of scorerNames) {
-        if (name === "latency") {
-          cells.push(pad(latencyMs !== void 0 ? `${Math.round(latencyMs)}ms` : "\u2014", 12, "right"));
-        } else if (name === "cost") {
-          cells.push(pad(formatCost(avgDetails.costUsd), 14, "right"));
-          cells.push(pad(avgDetails.totalTokens !== void 0 ? `${avgDetails.totalTokens}` : "\u2014", 11, "right"));
-        } else {
-          const val = avgScores[name];
-          if (val === void 0) cells.push(pad("\u2014", 10, "right"));
-          else cells.push(pad(colorScore(val), 10 + colorLen(colorScore(val)), "right"));
+      return {
+        providerId,
+        avgScores: averageScores(successResults),
+        avgDetails: averageDetails(successResults),
+        latencyMs: average(successResults.map((r3) => r3.raw.latencyMs)),
+        allErrors: false,
+        errorCount: errorResults2.length
+      };
+    });
+    const columnStats = computeColumnStats(providerData, scorerNames);
+    const medals = computeMedals(columnStats, providers);
+    const maxProviderLen = Math.max(...providers.map((id) => id.length));
+    const providerWidth = Math.min(35, Math.max(22, maxProviderLen + 5));
+    const cols = [
+      { label: "Provider", width: providerWidth, align: "left" }
+    ];
+    for (const name of scorerNames) {
+      if (name === "latency") {
+        cols.push({ label: "Latency", width: 10, align: "right", statsKey: "latency" });
+      } else if (name === "cost") {
+        cols.push({ label: "Cost", width: 12, align: "right", statsKey: "cost" });
+        cols.push({ label: "Tokens", width: 9, align: "right", statsKey: "tokens" });
+      } else {
+        const label = name === "correctness" ? "Match" : name === "schema-correctness" ? "Schema" : name === "fuzzy-similarity" ? "Fuzzy" : name === "llm-judge-correctness" ? "Judge" : name === "tool-usage" ? "Tool" : name;
+        cols.push({ label, width: showSparklines ? 15 : 8, align: "right", statsKey: name });
+      }
+    }
+    if (hasErrors) {
+      cols.push({ label: "Status", width: 8, align: "left" });
+    }
+    const widths = cols.map((c3) => c3.width);
+    const aligns = cols.map((c3) => c3.align);
+    console.log(`  ${drawTableLine(widths, "top")}`);
+    const headerCells = cols.map((c3) => bold(c3.label));
+    console.log(`  ${drawTableRow(headerCells, widths, aligns)}`);
+    console.log(`  ${drawTableLine(widths, "header")}`);
+    for (const pd of providerData) {
+      const medal = medals.get(pd.providerId) ?? "";
+      const providerCell = medal ? `${medal} ${pd.providerId}` : pd.providerId;
+      const cells = [providerCell];
+      if (pd.allErrors) {
+        for (const col of cols.slice(1)) {
+          if (col.label === "Status") {
+            cells.push(`${red}FAIL${reset}`);
+          } else {
+            cells.push(dim("\u2014"));
+          }
+        }
+      } else {
+        for (const col of cols.slice(1)) {
+          if (col.label === "Status") {
+            cells.push(
+              pd.errorCount > 0 ? `${yellow}${pd.errorCount} err${reset}` : `${green}OK${reset}`
+            );
+            continue;
+          }
+          const statsKey = col.statsKey;
+          const colStats = columnStats.get(statsKey);
+          if (statsKey === "latency") {
+            const ms = pd.latencyMs;
+            if (ms === void 0) {
+              cells.push(dim("\u2014"));
+            } else {
+              const text = `${Math.round(ms)}ms`;
+              cells.push(colStats ? colorByRank(text, ms, colStats, providers.length) : text);
+            }
+          } else if (statsKey === "cost") {
+            const cost = pd.avgDetails.costUsd;
+            if (cost === void 0) {
+              cells.push(dim("\u2014"));
+            } else {
+              const text = formatCost(cost);
+              cells.push(colStats ? colorByRank(text, cost, colStats, providers.length) : text);
+            }
+          } else if (statsKey === "tokens") {
+            const tokens = pd.avgDetails.totalTokens;
+            if (tokens === void 0) {
+              cells.push(dim("\u2014"));
+            } else {
+              const text = `${tokens}`;
+              cells.push(colStats ? colorByRank(text, tokens, colStats, providers.length) : text);
+            }
+          } else {
+            const val = pd.avgScores[statsKey];
+            if (val === void 0) {
+              cells.push(dim("\u2014"));
+            } else {
+              const pctStr = `${Math.round(val * 100)}%`.padStart(4);
+              let coloredPct;
+              if (multi && colStats) {
+                coloredPct = colorByRank(pctStr, val, colStats, providers.length);
+              } else {
+                if (val >= 0.8) coloredPct = `${green}${pctStr}${reset}`;
+                else if (val >= 0.5) coloredPct = `${yellow}${pctStr}${reset}`;
+                else coloredPct = `${red}${pctStr}${reset}`;
+              }
+              if (showSparklines) {
+                const { fill, track } = sparkBar(val);
+                const barColor = val >= 0.8 ? green : val >= 0.5 ? yellow : red;
+                cells.push(`${coloredPct} ${barColor}${fill}${reset}${dim(track)}`);
+              } else {
+                cells.push(coloredPct);
+              }
+            }
+          }
         }
       }
-      if (hasErrors) {
-        const failCount = errorResults2.length;
-        cells.push(failCount > 0 ? `  ${yellow}${failCount} err${reset}` : `  ${green}OK${reset}`);
+      console.log(`  ${drawTableRow(cells, widths, aligns)}`);
+    }
+    if (multi && providerData.some((p5) => !p5.allErrors)) {
+      const winnerId = [...medals.entries()].find(([, m8]) => m8 === "\u{1F947}")?.[0];
+      if (winnerId) {
+        console.log(`  ${drawTableLine(widths, "merge")}`);
+        const winnerText = `${brightGreen}${boldCode}\u{1F3C6}  Winner: ${winnerId}${reset} ${dim(providerLabel(winnerId))}`;
+        console.log(`  ${drawSpanRow(winnerText, widths)}`);
       }
-      console.log(`  ${cells.join("")}`);
     }
+    console.log(`  ${drawTableLine(widths, "bottom")}`);
     console.log("");
   }
   printSummary(results, providers);
   const errorResults = results.filter((r3) => r3.error);
   if (errorResults.length > 0) {
     console.log(`  ${bold("Errors")}`);
-    console.log(`  ${dim("\u2500".repeat(70))}`);
+    console.log(`  ${dim("\u2501".repeat(72))}`);
     const seen = /* @__PURE__ */ new Set();
     for (const r3 of errorResults) {
       const key = `${r3.providerId}::${r3.error}`;
@@ -145,7 +362,7 @@ function consoleReporter(results) {
       seen.add(key);
       const count = errorResults.filter((e5) => e5.providerId === r3.providerId && e5.error === r3.error).length;
       const suffix = count > 1 ? ` (\xD7${count})` : "";
-      console.log(`  ${red}\u2717${reset} ${r3.providerId}: ${r3.error}${suffix}`);
+      console.log(`  ${red}\u2716${reset} ${r3.providerId}: ${r3.error}${suffix}`);
       const hint = apiKeyHint(r3.providerId, r3.error ?? "");
       if (hint) console.log(`    ${dim(hint)}`);
     }
@@ -159,15 +376,20 @@ function consoleReporter(results) {
 function printSummary(results, providers) {
   const successResults = results.filter((r3) => !r3.error);
   if (successResults.length === 0) return;
-  console.log(`  ${dim("\u2500".repeat(70))}`);
   console.log(`  ${bold("Summary")}`);
+  console.log(`  ${dim("\u2501".repeat(72))}`);
   console.log("");
   const single = providers.length === 1;
   const correctnessKey = successResults.some((r3) => r3.scores.some((s5) => s5.name === "llm-judge-correctness" && s5.value >= 0)) ? "llm-judge-correctness" : "correctness";
   const byCorrectness = rankProviders(successResults, providers, correctnessKey);
   if (byCorrectness) {
-    const label = single ? "Avg correctness" : `Most correct: ${bold(byCorrectness.id)} ${dim(providerLabel(byCorrectness.id))}`;
-    console.log(`  ${cyan}\u25C6${reset} ${label} (avg ${colorScore(byCorrectness.avg)})`);
+    const medal = single ? `${cyan}\u25C6${reset}` : "\u{1F947}";
+    const pctStr = `${Math.round(byCorrectness.avg * 100)}%`;
+    if (single) {
+      console.log(`  ${medal} Avg correctness:  ${brightGreen}${boldCode}${pctStr}${reset}`);
+    } else {
+      console.log(`  ${medal} Most correct:  ${bold(byCorrectness.id)} ${dim(providerLabel(byCorrectness.id))}  ${brightGreen}${boldCode}${pctStr}${reset}`);
+    }
   }
   const byLatency = providers.map((id) => {
     const runs = successResults.filter((r3) => r3.providerId === id);
@@ -175,8 +397,13 @@ function printSummary(results, providers) {
     return { id, avg: avg ?? Infinity };
   }).sort((a7, b3) => a7.avg - b3.avg)[0];
   if (byLatency && byLatency.avg !== Infinity) {
-    const label = single ? "Avg latency" : `Fastest: ${bold(byLatency.id)} ${dim(providerLabel(byLatency.id))}`;
-    console.log(`  ${cyan}\u25C6${reset} ${label} (avg ${Math.round(byLatency.avg)}ms)`);
+    const medal = single ? `${cyan}\u25C6${reset}` : "\u{1F947}";
+    const msStr = `${Math.round(byLatency.avg)}ms`;
+    if (single) {
+      console.log(`  ${medal} Avg latency:     ${brightGreen}${boldCode}${msStr}${reset}`);
+    } else {
+      console.log(`  ${medal} Fastest:       ${bold(byLatency.id)} ${dim(providerLabel(byLatency.id))}  ${brightGreen}${boldCode}${msStr}${reset}`);
+    }
   }
   const byCost = providers.map((id) => {
     const runs = successResults.filter((r3) => r3.providerId === id);
@@ -188,8 +415,32 @@ function printSummary(results, providers) {
     return { id, avg };
   }).filter((p5) => p5.avg !== void 0).sort((a7, b3) => a7.avg - b3.avg)[0];
   if (byCost?.avg !== void 0) {
-    const label = single ? "Avg cost" : `Cheapest: ${bold(byCost.id)} ${dim(providerLabel(byCost.id))}`;
-    console.log(`  ${cyan}\u25C6${reset} ${label} (avg ${formatCost(byCost.avg)})`);
+    const medal = single ? `${cyan}\u25C6${reset}` : "\u{1F947}";
+    const costStr = formatCost(byCost.avg);
+    if (single) {
+      console.log(`  ${medal} Avg cost:        ${brightGreen}${boldCode}${costStr}${reset}`);
+    } else {
+      console.log(`  ${medal} Cheapest:      ${bold(byCost.id)} ${dim(providerLabel(byCost.id))}  ${brightGreen}${boldCode}${costStr}${reset}`);
+    }
+  }
+  if (!single) {
+    const wins = /* @__PURE__ */ new Map();
+    for (const id of providers) wins.set(id, 0);
+    if (byCorrectness) wins.set(byCorrectness.id, (wins.get(byCorrectness.id) ?? 0) + 1);
+    if (byLatency && byLatency.avg !== Infinity) wins.set(byLatency.id, (wins.get(byLatency.id) ?? 0) + 1);
+    if (byCost?.avg !== void 0) wins.set(byCost.id, (wins.get(byCost.id) ?? 0) + 1);
+    const maxWins = Math.max(...wins.values());
+    if (maxWins > 0) {
+      const topProviders = [...wins.entries()].filter(([, w4]) => w4 === maxWins);
+      console.log("");
+      if (topProviders.length === 1) {
+        const [winnerId, winCount] = topProviders[0];
+        console.log(`  \u{1F3C6} Overall:      ${brightGreen}${boldCode}${winnerId}${reset} ${dim(providerLabel(winnerId))}  ${dim(`(${winCount}/3 categories)`)}`);
+      } else {
+        const names = topProviders.map(([id]) => bold(id)).join(dim(", "));
+        console.log(`  \u{1F3C6} Overall:      ${names}  ${dim(`(tied at ${maxWins}/3)`)}`);
+      }
+    }
   }
   console.log("");
 }
@@ -251,14 +502,6 @@ function formatCost(usd) {
   const digits = Math.max(4, -Math.floor(Math.log10(usd)) + 1);
   return `~$${usd.toFixed(digits).replace(/0+$/, "")}`;
 }
-function pad(str, width, align) {
-  if (align === "right") return str.padStart(width);
-  return str.padEnd(width);
-}
-function colorLen(str) {
-  const stripped = str.replace(/\x1b\[[0-9;]*m/g, "");
-  return str.length - stripped.length;
-}
 function apiKeyHint(providerId, error) {
   const lower = error.toLowerCase();
   const isAuthError = lower.includes("api key") || lower.includes("401") || lower.includes("unauthorized") || lower.includes("authentication") || lower.includes("incorrect api key") || lower.includes("apikey");
@@ -328,7 +571,7 @@ function providerLabel(providerId) {
       return `(${prefix})`;
   }
 }
-var reset, boldCode, dimCode, green, red, yellow, cyan;
+var reset, boldCode, dimCode, green, red, yellow, cyan, brightGreen, brightWhite;
 var init_console = __esm({
   "src/reporter/console.ts"() {
     "use strict";
@@ -339,6 +582,8 @@ var init_console = __esm({
     red = "\x1B[31m";
     yellow = "\x1B[33m";
     cyan = "\x1B[36m";
+    brightGreen = "\x1B[92m";
+    brightWhite = "\x1B[97m";
   }
 });
 
@@ -375,6 +620,425 @@ var init_json = __esm({
   }
 });
 
+// src/ci.ts
+var ci_exports = {};
+__export(ci_exports, {
+  compareResults: () => compareResults,
+  computeCostSummary: () => computeCostSummary,
+  computeScorerStats: () => computeScorerStats,
+  computeStats: () => computeStats,
+  loadBaseline: () => loadBaseline,
+  saveBaseline: () => saveBaseline
+});
+import { readFileSync, writeFileSync, mkdirSync } from "fs";
+import { dirname } from "path";
+function tCritical(df) {
+  if (df <= 0) return 1.96;
+  if (T_CRITICAL_95[df] !== void 0) return T_CRITICAL_95[df];
+  const keys = Object.keys(T_CRITICAL_95).map(Number).sort((a7, b3) => a7 - b3);
+  if (df > keys[keys.length - 1]) return 1.96;
+  for (let i7 = 0; i7 < keys.length - 1; i7++) {
+    if (df > keys[i7] && df < keys[i7 + 1]) {
+      const low = keys[i7], high = keys[i7 + 1];
+      const ratio = (df - low) / (high - low);
+      return T_CRITICAL_95[low] + ratio * (T_CRITICAL_95[high] - T_CRITICAL_95[low]);
+    }
+  }
+  return 1.96;
+}
+function computeScorerStats(samples) {
+  const n3 = samples.length;
+  if (n3 === 0) {
+    return { mean: 0, stddev: 0, cv: 0, n: 0, ci95Lower: 0, ci95Upper: 0 };
+  }
+  const mean = samples.reduce((a7, b3) => a7 + b3, 0) / n3;
+  if (n3 === 1) {
+    return { mean, stddev: 0, cv: 0, n: 1, ci95Lower: mean, ci95Upper: mean };
+  }
+  const variance = samples.reduce((sum, x) => sum + (x - mean) ** 2, 0) / (n3 - 1);
+  const stddev = Math.sqrt(variance);
+  const cv = mean !== 0 ? stddev / Math.abs(mean) : 0;
+  const se4 = stddev / Math.sqrt(n3);
+  const t3 = tCritical(n3 - 1);
+  return {
+    mean,
+    stddev,
+    cv,
+    n: n3,
+    ci95Lower: mean - t3 * se4,
+    ci95Upper: mean + t3 * se4
+  };
+}
+function groupKey(providerId, taskName, scorerName) {
+  return `${providerId}::${taskName}::${scorerName}`;
+}
+function computeStats(results) {
+  const grouped = /* @__PURE__ */ new Map();
+  for (const r3 of results) {
+    if (r3.error) continue;
+    for (const score of r3.scores) {
+      if (score.value < 0) continue;
+      const key = groupKey(r3.providerId, r3.taskName, score.name);
+      if (!grouped.has(key)) grouped.set(key, []);
+      grouped.get(key).push(score.value);
+    }
+  }
+  const stats = /* @__PURE__ */ new Map();
+  for (const [key, samples] of grouped) {
+    stats.set(key, computeScorerStats(samples));
+  }
+  return stats;
+}
+function computeCostSummary(results, budget) {
+  let totalUsd = 0;
+  const perProvider = /* @__PURE__ */ new Map();
+  for (const r3 of results) {
+    if (r3.error) continue;
+    const costScore = r3.scores.find((s5) => s5.name === "cost");
+    if (!costScore || costScore.value < 0) continue;
+    const details = costScore.details;
+    const usd = details?.estimatedUsd ?? 0;
+    if (usd <= 0) continue;
+    totalUsd += usd;
+    perProvider.set(r3.providerId, (perProvider.get(r3.providerId) ?? 0) + usd);
+  }
+  return {
+    totalUsd,
+    perProvider,
+    budget,
+    overBudget: budget !== void 0 && totalUsd > budget
+  };
+}
+function compareResults(baselineStats, currentStats, thresholds, budget, currentResults) {
+  const comparisons = [];
+  const failureReasons = [];
+  for (const [key, current] of currentStats) {
+    const [providerId, taskName, scorerName] = key.split("::");
+    const baseline = baselineStats?.get(key) ?? null;
+    let delta = null;
+    let regressed = false;
+    let improved = false;
+    if (baseline) {
+      delta = current.mean - baseline.mean;
+      const threshold = thresholds.get(scorerName);
+      if (threshold !== void 0) {
+        const lowerIsBetter = LOWER_IS_BETTER.has(scorerName);
+        regressed = detectRegression(baseline, current, threshold, lowerIsBetter);
+        improved = detectImprovement(baseline, current, threshold, lowerIsBetter);
+      }
+    }
+    const flaky = current.n > 1 && current.cv > FLAKY_CV_THRESHOLD;
+    comparisons.push({
+      providerId,
+      taskName,
+      scorerName,
+      baseline,
+      current,
+      delta,
+      regressed,
+      improved,
+      flaky
+    });
+  }
+  const cost = computeCostSummary(currentResults ?? [], budget);
+  const regressions = comparisons.filter((c3) => c3.regressed);
+  if (regressions.length > 0) {
+    for (const r3 of regressions) {
+      failureReasons.push(
+        `${r3.providerId} \xD7 ${r3.taskName}: ${r3.scorerName} regressed by ${formatDelta(r3.delta)}`
+      );
+    }
+  }
+  if (cost.overBudget) {
+    failureReasons.push(
+      `Total cost $${cost.totalUsd.toFixed(4)} exceeds budget $${cost.budget.toFixed(2)}`
+    );
+  }
+  const flakyResults = comparisons.filter((c3) => c3.flaky);
+  const failed = failureReasons.length > 0;
+  return { comparisons, cost, failed, flakyResults, failureReasons };
+}
+function detectRegression(baseline, current, threshold, lowerIsBetter) {
+  if (baseline.n === 1 && current.n === 1) {
+    const delta = current.mean - baseline.mean;
+    if (lowerIsBetter) return delta > threshold;
+    return delta < -threshold;
+  }
+  if (lowerIsBetter) {
+    return current.ci95Lower - baseline.ci95Upper > threshold;
+  }
+  return baseline.ci95Upper - current.ci95Lower > threshold && current.mean < baseline.mean;
+}
+function detectImprovement(baseline, current, threshold, lowerIsBetter) {
+  if (baseline.n === 1 && current.n === 1) {
+    const delta = current.mean - baseline.mean;
+    if (lowerIsBetter) return delta < -threshold;
+    return delta > threshold;
+  }
+  if (lowerIsBetter) {
+    return baseline.ci95Lower - current.ci95Upper > threshold;
+  }
+  return current.ci95Lower - baseline.ci95Upper > threshold;
+}
+function formatDelta(delta) {
+  const sign = delta >= 0 ? "+" : "";
+  return `${sign}${delta.toFixed(4)}`;
+}
+function loadBaseline(path) {
+  try {
+    const raw = readFileSync(path, "utf-8");
+    const data = JSON.parse(raw);
+    const results = data.results ?? data;
+    if (!Array.isArray(results)) return null;
+    return {
+      timestamp: data.timestamp ?? "unknown",
+      results
+    };
+  } catch {
+    return null;
+  }
+}
+function saveBaseline(path, results) {
+  mkdirSync(dirname(path), { recursive: true });
+  const data = {
+    timestamp: (/* @__PURE__ */ new Date()).toISOString(),
+    results
+  };
+  writeFileSync(path, JSON.stringify(data, null, 2));
+}
+var LOWER_IS_BETTER, FLAKY_CV_THRESHOLD, T_CRITICAL_95;
+var init_ci = __esm({
+  "src/ci.ts"() {
+    "use strict";
+    LOWER_IS_BETTER = /* @__PURE__ */ new Set(["cost"]);
+    FLAKY_CV_THRESHOLD = 0.3;
+    T_CRITICAL_95 = {
+      1: 12.706,
+      2: 4.303,
+      3: 3.182,
+      4: 2.776,
+      5: 2.571,
+      6: 2.447,
+      7: 2.365,
+      8: 2.306,
+      9: 2.262,
+      10: 2.228,
+      15: 2.131,
+      20: 2.086,
+      25: 2.06,
+      30: 2.042
+    };
+  }
+});
+
+// src/reporter/markdown.ts
+var markdown_exports = {};
+__export(markdown_exports, {
+  COMMENT_MARKER: () => COMMENT_MARKER,
+  markdownComparisonTable: () => markdownComparisonTable,
+  markdownCostSummary: () => markdownCostSummary,
+  markdownReporter: () => markdownReporter
+});
+function markdownReporter(report, _current) {
+  const lines = [COMMENT_MARKER, ""];
+  const status = report.failed ? "\u{1F534} Failed" : "\u{1F7E2} Passed";
+  lines.push(`## \u2B21 Agent Duelist CI \u2014 ${status}`);
+  lines.push("");
+  if (report.comparisons.length > 0) {
+    lines.push(markdownComparisonTable(report.comparisons));
+    lines.push("");
+  }
+  if (report.cost.totalUsd > 0 || report.cost.budget !== void 0) {
+    lines.push(markdownCostSummary(report.cost));
+    lines.push("");
+  }
+  if (report.flakyResults.length > 0) {
+    lines.push("### \u26A0\uFE0F Flaky Results");
+    lines.push("");
+    lines.push("These scorer/task combinations have high variance (CV > 0.3). Consider increasing `runs` or tightening prompts.");
+    lines.push("");
+    for (const f6 of report.flakyResults) {
+      lines.push(`- **${f6.providerId}** \xD7 ${f6.taskName} \u2192 ${f6.scorerName} (CV = ${f6.current.cv.toFixed(2)})`);
+    }
+    lines.push("");
+  }
+  if (report.failureReasons.length > 0) {
+    lines.push("### Failure Reasons");
+    lines.push("");
+    for (const reason of report.failureReasons) {
+      lines.push(`- ${reason}`);
+    }
+    lines.push("");
+  }
+  lines.push("---");
+  lines.push("*Generated by [agent-duelist](https://github.com/DataGobes/agent-duelist) CI*");
+  return lines.join("\n");
+}
+function markdownComparisonTable(comparisons) {
+  const lines = [];
+  lines.push("| Provider | Task | Scorer | Baseline | Current | Delta | Status |");
+  lines.push("|----------|------|--------|----------|---------|-------|--------|");
+  for (const c3 of comparisons) {
+    const baselineStr = c3.baseline ? formatStats(c3.baseline) : "\u2014";
+    const currentStr = formatStats(c3.current);
+    const deltaStr = c3.delta !== null ? formatDelta2(c3.delta) : "\u2014";
+    const status = statusIndicator(c3);
+    lines.push(`| ${c3.providerId} | ${c3.taskName} | ${c3.scorerName} | ${baselineStr} | ${currentStr} | ${deltaStr} | ${status} |`);
+  }
+  return lines.join("\n");
+}
+function markdownCostSummary(cost) {
+  const lines = [];
+  lines.push("### \u{1F4B0} Cost Summary");
+  lines.push("");
+  lines.push(`**Total:** $${cost.totalUsd.toFixed(4)}`);
+  if (cost.budget !== void 0) {
+    const pct = cost.budget > 0 ? (cost.totalUsd / cost.budget * 100).toFixed(0) : "\u221E";
+    const indicator = cost.overBudget ? "\u{1F534}" : "\u{1F7E2}";
+    lines.push(`**Budget:** $${cost.budget.toFixed(2)} (${pct}% used) ${indicator}`);
+  }
+  if (cost.perProvider.size > 1) {
+    lines.push("");
+    lines.push("| Provider | Cost |");
+    lines.push("|----------|------|");
+    for (const [provider, usd] of cost.perProvider) {
+      lines.push(`| ${provider} | $${usd.toFixed(4)} |`);
+    }
+  }
+  return lines.join("\n");
+}
+function formatStats(stats) {
+  if (stats.n > 1) {
+    const margin = (stats.ci95Upper - stats.ci95Lower) / 2;
+    return `${stats.mean.toFixed(3)} \xB1 ${margin.toFixed(3)}`;
+  }
+  return stats.mean.toFixed(3);
+}
+function formatDelta2(delta) {
+  const sign = delta >= 0 ? "+" : "";
+  return `${sign}${delta.toFixed(3)}`;
+}
+function statusIndicator(c3) {
+  if (c3.regressed) return "\u{1F534} regressed";
+  if (c3.improved) return "\u{1F7E2} improved";
+  if (c3.baseline === null) return "\u{1F195} new";
+  return "\u26AA unchanged";
+}
+var COMMENT_MARKER;
+var init_markdown = __esm({
+  "src/reporter/markdown.ts"() {
+    "use strict";
+    COMMENT_MARKER = "<!-- duelist-ci-report -->";
+  }
+});
+
+// src/github.ts
+var github_exports = {};
+__export(github_exports, {
+  detectGitHubContext: () => detectGitHubContext,
+  findExistingComment: () => findExistingComment,
+  upsertPrComment: () => upsertPrComment
+});
+import { readFileSync as readFileSync2 } from "fs";
+function detectGitHubContext() {
+  const token = process.env.GITHUB_TOKEN;
+  const repository = process.env.GITHUB_REPOSITORY;
+  const eventPath = process.env.GITHUB_EVENT_PATH;
+  if (!token || !repository) return null;
+  const [owner, repo] = repository.split("/");
+  if (!owner || !repo) return null;
+  let prNumber;
+  if (eventPath) {
+    try {
+      const event = JSON.parse(readFileSync2(eventPath, "utf-8"));
+      if (event.pull_request && typeof event.pull_request === "object") {
+        const pr = event.pull_request;
+        prNumber = pr.number;
+      }
+      if (!prNumber && event.issue && typeof event.issue === "object") {
+        const issue = event.issue;
+        if (issue.pull_request) {
+          prNumber = issue.number;
+        }
+      }
+    } catch {
+    }
+  }
+  if (!prNumber && process.env.DUELIST_PR_NUMBER) {
+    prNumber = parseInt(process.env.DUELIST_PR_NUMBER, 10);
+  }
+  if (!prNumber) return null;
+  return { token, owner, repo, prNumber };
+}
+async function findExistingComment(ctx, marker) {
+  let page = 1;
+  const perPage = 50;
+  while (true) {
+    const url = `${API_BASE}/repos/${ctx.owner}/${ctx.repo}/issues/${ctx.prNumber}/comments?per_page=${perPage}&page=${page}`;
+    const res = await fetch(url, {
+      headers: {
+        Authorization: `Bearer ${ctx.token}`,
+        Accept: "application/vnd.github+json",
+        "X-GitHub-Api-Version": "2022-11-28"
+      }
+    });
+    if (!res.ok) return null;
+    const comments = await res.json();
+    if (comments.length === 0) break;
+    for (const comment of comments) {
+      if (comment.body?.includes(marker)) {
+        return comment.id;
+      }
+    }
+    if (comments.length < perPage) break;
+    page++;
+  }
+  return null;
+}
+async function upsertPrComment(ctx, body, marker) {
+  const existingId = await findExistingComment(ctx, marker);
+  if (existingId) {
+    const url = `${API_BASE}/repos/${ctx.owner}/${ctx.repo}/issues/comments/${existingId}`;
+    const res = await fetch(url, {
+      method: "PATCH",
+      headers: {
+        Authorization: `Bearer ${ctx.token}`,
+        Accept: "application/vnd.github+json",
+        "Content-Type": "application/json",
+        "X-GitHub-Api-Version": "2022-11-28"
+      },
+      body: JSON.stringify({ body })
+    });
+    if (!res.ok) {
+      const text = await res.text();
+      console.warn(`Failed to update PR comment: ${res.status} ${text}`);
+    }
+  } else {
+    const url = `${API_BASE}/repos/${ctx.owner}/${ctx.repo}/issues/${ctx.prNumber}/comments`;
+    const res = await fetch(url, {
+      method: "POST",
+      headers: {
+        Authorization: `Bearer ${ctx.token}`,
+        Accept: "application/vnd.github+json",
+        "Content-Type": "application/json",
+        "X-GitHub-Api-Version": "2022-11-28"
+      },
+      body: JSON.stringify({ body })
+    });
+    if (!res.ok) {
+      const text = await res.text();
+      console.warn(`Failed to create PR comment: ${res.status} ${text}`);
+    }
+  }
+}
+var API_BASE;
+var init_github = __esm({
+  "src/github.ts"() {
+    "use strict";
+    API_BASE = "https://api.github.com";
+  }
+});
+
 // node_modules/tsx/dist/temporary-directory-CwHp0_NW.mjs
 import r from "path";
 import o from "os";
@@ -6530,10 +7194,10 @@ var init_api = __esm({
 // src/cli.ts
 import "dotenv/config";
 import { Command } from "commander";
-import { readFileSync, writeFileSync, existsSync } from "fs";
-import { resolve, join, dirname } from "path";
+import { readFileSync as readFileSync3, writeFileSync as writeFileSync2, existsSync } from "fs";
+import { resolve, join, dirname as dirname2 } from "path";
 import { pathToFileURL, fileURLToPath } from "url";
-var __dirname2 = dirname(fileURLToPath(import.meta.url));
+var __dirname2 = dirname2(fileURLToPath(import.meta.url));
 var program = new Command();
 program.name("duelist").description("Pit LLM providers against each other on agent tasks.").version(getVersion());
 program.command("init").description("Scaffold an arena.config.ts in the current directory").option("--force", "Overwrite existing config file").action((opts) => {
@@ -6545,11 +7209,11 @@ program.command("init").description("Scaffold an arena.config.ts in the current
   const templatePath = join(__dirname2, "..", "templates", "arena.config.ts");
   let template;
   if (existsSync(templatePath)) {
-    template = readFileSync(templatePath, "utf-8");
+    template = readFileSync3(templatePath, "utf-8");
   } else {
     template = DEFAULT_TEMPLATE;
   }
-  writeFileSync(target, template);
+  writeFileSync2(target, template);
   console.log(existsSync(target) && opts.force ? "Overwrote arena.config.ts" : "Created arena.config.ts");
   console.log("");
   console.log("Next steps:");
@@ -6557,17 +7221,119 @@ program.command("init").description("Scaffold an arena.config.ts in the current
   console.log("  2. npx duelist run");
 });
 program.command("run").description("Run benchmarks defined in your arena config").option("-c, --config <path>", "Path to config file", "arena.config.ts").option("--reporter <type>", "Output format: console or json", "console").option("-q, --quiet", "Suppress per-result progress (show only final report)").action(async (opts) => {
-  const configPath = resolve(opts.config);
+  if (!["console", "json"].includes(opts.reporter)) {
+    console.error(`Unknown reporter "${opts.reporter}". Use "console" or "json".`);
+    process.exit(1);
+  }
+  const typedArena = await loadArenaConfig(opts.config);
+  try {
+    const showProgress = opts.reporter === "console" && !opts.quiet;
+    const onResult = showProgress ? logResult : void 0;
+    const results = await typedArena.run({ onResult });
+    const { consoleReporter: consoleReporter2 } = await Promise.resolve().then(() => (init_console(), console_exports));
+    const { jsonReporter: jsonReporter2 } = await Promise.resolve().then(() => (init_json(), json_exports));
+    if (opts.reporter === "json") {
+      console.log(jsonReporter2(results));
+    } else {
+      console.log("");
+      consoleReporter2(results, { sparklines: typedArena.config?.sparklines });
+    }
+    const allFailed = results.length > 0 && results.every((r3) => r3.error);
+    if (allFailed) process.exit(1);
+  } catch (err) {
+    const message = err instanceof Error ? err.message : String(err);
+    console.error(`Benchmark failed: ${message}`);
+    process.exit(1);
+  }
+});
+function collectThreshold(value, previous) {
+  const [scorer, delta] = value.split("=");
+  if (!scorer || delta === void 0 || isNaN(Number(delta))) {
+    console.error(`Invalid threshold format: "${value}". Expected scorer=delta (e.g., correctness=0.1)`);
+    process.exit(1);
+  }
+  previous.set(scorer, Number(delta));
+  return previous;
+}
+program.command("ci").description("Run benchmarks, compare against baseline, and enforce quality gates").option("-c, --config <path>", "Path to config file", "arena.config.ts").option("--baseline <path>", "Baseline JSON file", ".duelist/baseline.json").option("--budget <dollars>", "Max total cost in USD", parseFloat).option("--threshold <scorer=delta>", "Regression threshold (repeatable)", collectThreshold, /* @__PURE__ */ new Map()).option("--update-baseline", "Save results as new baseline after passing").option("--comment", "Post results as GitHub PR comment").option("-q, --quiet", "Suppress per-result progress").action(async (opts) => {
+  const ciOpts = {
+    configPath: opts.config,
+    baselinePath: resolve(opts.baseline),
+    budget: opts.budget,
+    thresholds: opts.threshold,
+    updateBaseline: opts.updateBaseline ?? false,
+    comment: opts.comment ?? false,
+    quiet: opts.quiet ?? false
+  };
+  const typedArena = await loadArenaConfig(ciOpts.configPath);
+  console.log("Running benchmarks...");
+  const onResult = ciOpts.quiet ? void 0 : logResult;
+  let results;
+  try {
+    results = await typedArena.run({ onResult });
+  } catch (err) {
+    const message = err instanceof Error ? err.message : String(err);
+    console.error(`Benchmark failed: ${message}`);
+    process.exit(1);
+  }
+  const { loadBaseline: loadBaseline2, saveBaseline: saveBaseline2, computeStats: computeStats2, compareResults: compareResults2 } = await Promise.resolve().then(() => (init_ci(), ci_exports));
+  const baseline = loadBaseline2(ciOpts.baselinePath);
+  const baselineStats = baseline ? computeStats2(baseline.results) : null;
+  if (baseline) {
+    console.log(`Loaded baseline from ${ciOpts.baselinePath} (${baseline.timestamp})`);
+  } else {
+    console.log("No baseline found \u2014 this run establishes the first baseline.");
+  }
+  const currentStats = computeStats2(results);
+  const report = compareResults2(baselineStats, currentStats, ciOpts.thresholds, ciOpts.budget, results);
+  const { consoleReporter: consoleReporter2 } = await Promise.resolve().then(() => (init_console(), console_exports));
+  console.log("");
+  consoleReporter2(results, { sparklines: typedArena.config?.sparklines });
+  const { markdownReporter: markdownReporter2, COMMENT_MARKER: COMMENT_MARKER2 } = await Promise.resolve().then(() => (init_markdown(), markdown_exports));
+  if (report.flakyResults.length > 0) {
+    console.log(`\u26A0  ${report.flakyResults.length} flaky result(s) detected (high variance)`);
+  }
+  if (report.cost.overBudget) {
+    console.log(`\u{1F534} Budget exceeded: $${report.cost.totalUsd.toFixed(4)} > $${report.cost.budget.toFixed(2)}`);
+  }
+  for (const reason of report.failureReasons) {
+    console.log(`\u{1F534} ${reason}`);
+  }
+  if (!report.failed) {
+    console.log("\u{1F7E2} CI passed");
+  }
+  if (ciOpts.comment) {
+    const { detectGitHubContext: detectGitHubContext2, upsertPrComment: upsertPrComment2 } = await Promise.resolve().then(() => (init_github(), github_exports));
+    const ghCtx = detectGitHubContext2();
+    if (ghCtx) {
+      const markdown = markdownReporter2(report, results);
+      try {
+        await upsertPrComment2(ghCtx, markdown, COMMENT_MARKER2);
+        console.log("Posted results to PR comment.");
+      } catch (err) {
+        console.warn(`Failed to post PR comment: ${err instanceof Error ? err.message : err}`);
+      }
+    } else {
+      console.warn("--comment: not in a GitHub Actions PR context, skipping.");
+    }
+  }
+  if (ciOpts.updateBaseline && !report.failed) {
+    saveBaseline2(ciOpts.baselinePath, results);
+    console.log(`Baseline saved to ${ciOpts.baselinePath}`);
+  } else if (ciOpts.updateBaseline && report.failed) {
+    console.log("Baseline not updated (CI failed).");
+  }
+  process.exit(report.failed ? 1 : 0);
+});
+program.parse();
+async function loadArenaConfig(configOpt) {
+  const configPath = resolve(configOpt);
   if (!existsSync(configPath)) {
     console.error(`Config not found: ${configPath}`);
     console.error("");
     console.error("Create one with: npx duelist init");
     process.exit(1);
   }
-  if (!["console", "json"].includes(opts.reporter)) {
-    console.error(`Unknown reporter "${opts.reporter}". Use "console" or "json".`);
-    process.exit(1);
-  }
   let mod;
   try {
     if (configPath.endsWith(".ts")) {
@@ -6591,35 +7357,16 @@ program.command("run").description("Run benchmarks defined in your arena config"
     console.error(`Loaded from: ${configPath}`);
     process.exit(1);
   }
-  try {
-    const typedArena = arena;
-    const showProgress = opts.reporter === "console" && !opts.quiet;
-    const onResult = showProgress ? (result) => {
-      if (result.error) {
-        console.log(`  ${result.providerId} \xD7 ${result.taskName}: ERROR ${result.error}`);
-      } else {
-        const scores = result.scores.map((s5) => `${s5.name}=${formatScoreForLog(s5)}`).join(" ");
-        console.log(`  ${result.providerId} \xD7 ${result.taskName}: ${scores}`);
-      }
-    } : void 0;
-    const results = await typedArena.run({ onResult });
-    const { consoleReporter: consoleReporter2 } = await Promise.resolve().then(() => (init_console(), console_exports));
-    const { jsonReporter: jsonReporter2 } = await Promise.resolve().then(() => (init_json(), json_exports));
-    if (opts.reporter === "json") {
-      console.log(jsonReporter2(results));
-    } else {
-      console.log("");
-      consoleReporter2(results);
-    }
-    const allFailed = results.length > 0 && results.every((r3) => r3.error);
-    if (allFailed) process.exit(1);
-  } catch (err) {
-    const message = err instanceof Error ? err.message : String(err);
-    console.error(`Benchmark failed: ${message}`);
-    process.exit(1);
+  return arena;
+}
+function logResult(result) {
+  if (result.error) {
+    console.log(`  ${result.providerId} \xD7 ${result.taskName}: ERROR ${result.error}`);
+  } else {
+    const scores = result.scores.map((s5) => `${s5.name}=${formatScoreForLog(s5)}`).join(" ");
+    console.log(`  ${result.providerId} \xD7 ${result.taskName}: ${scores}`);
   }
-});
-program.parse();
+}
 async function importTypeScript(filePath) {
   try {
     await Promise.resolve().then(() => (init_api(), api_exports));
@@ -6652,7 +7399,7 @@ function formatScoreForLog(s5) {
 }
 function getVersion() {
   try {
-    const pkg = readFileSync(join(__dirname2, "..", "package.json"), "utf-8");
+    const pkg = readFileSync3(join(__dirname2, "..", "package.json"), "utf-8");
     return JSON.parse(pkg).version ?? "0.0.0";
   } catch {
     return "0.0.0";
@@ -6674,12 +7421,12 @@ import { z } from 'zod'
 
 export default defineArena({
   providers: [
-    openai('gpt-4o-mini'),
+    openai('gpt-5-mini'),
     // Add more providers to compare:
-    // openai('gpt-4o'),
-    // azureOpenai('gpt-4o-mini'),
-    // anthropic('claude-sonnet-4-20250514'),
-    // gemini('gemini-2.5-flash'),
+    // openai('gpt-5.2'),
+    // azureOpenai('gpt-5-mini'),
+    // anthropic('claude-sonnet-4.6'),
+    // gemini('gemini-3-flash-preview'),
   ],
 
   tasks: [