dsh-model-arena 0.2.0 → 0.3.0

package/lib/store.js

@@ -1,4 +1,4 @@
-import { writeFileSync, readFileSync, existsSync, mkdirSync, readdirSync } from 'node:fs';
+import { writeFileSync, readFileSync, existsSync, mkdirSync, readdirSync, unlinkSync } from 'node:fs';
 import { join } from 'node:path';
 import { createHash } from 'node:crypto';
 export class ArenaStore {
@@ -40,6 +40,13 @@ export class ArenaStore {
             return null;
         return JSON.parse(readFileSync(path, 'utf8'));
     }
+    deleteRun(id) {
+        const path = join(this.arenaDir, `${id}.json`);
+        if (!existsSync(path))
+            return false;
+        unlinkSync(path);
+        return true;
+    }
     static generateId(prompt, models) {
         const hash = createHash('sha256');
         hash.update(prompt);

package/lib/types.d.ts

@@ -5,6 +5,7 @@ export interface ModelResult {
     completionTokens: number;
     latencyMs: number;
     error?: string;
+    rating?: number;
 }
 export interface Score {
     userRating?: number;
@@ -18,6 +19,7 @@ export interface ArenaRun {
     timestamp: number;
     results: ModelResult[];
     scores: Record<string, Score>;
+    reference?: string | undefined;
 }
 export interface Suite {
     name: string;
@@ -36,11 +38,53 @@ export interface RegressionReport {
         currentOutput: string;
     }[];
 }
-export interface PanelPayload {
-    recentRuns: {
-        id: string;
-        prompt: string;
-        timestamp: number;
-        modelCount: number;
+export interface LeaderboardEntry {
+    model: string;
+    elo: number;
+    wins: number;
+    losses: number;
+    draws: number;
+    games: number;
+    totalRuns: number;
+    avgRating: number;
+    avgLatencyMs: number;
+    avgTokens: number;
+}
+export interface RunSummary {
+    id: string;
+    prompt: string;
+    timestamp: number;
+    modelCount: number;
+    winner?: string | undefined;
+}
+export interface ComparePayload {
+    runId: string;
+    modelA: string;
+    modelB: string;
+    outputA: string;
+    outputB: string;
+    diff: {
+        type: 'same' | 'added' | 'removed';
+        text: string;
     }[];
+    summary: {
+        added: number;
+        removed: number;
+        changed: number;
+    };
+    scoreA: Score;
+    scoreB: Score;
+}
+export interface ArenaStats {
+    totalRuns: number;
+    totalModels: number;
+    uniqueModels: number;
+    uniquePrompts: number;
+    totalRatings: number;
+    avgModelsPerRun: number;
+}
+export interface PanelPayload {
+    recentRuns: RunSummary[];
+    leaderboard: LeaderboardEntry[];
+    stats: ArenaStats;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-model-arena",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "description": "Run the same prompt through every model you have, and see the difference without squinting.",
   "type": "module",
   "main": "lib/index.js",
@@ -41,8 +41,9 @@
     "client": {
       "platform": "web",
       "inject": [
-        "slots",
-        "connection"
+        "@deepseek-ai/dsh-client-ui-renderer",
+        "@deepseek-ai/dsh-client-locale",
+        "@deepseek-ai/dsh-client-ui-settings"
       ]
     }
   },