feat: add types package

packages/types/package.json

@@ -0,0 +1,18 @@
+{
+  "name": "@duck/types",
+  "version": "0.1.0",
+  "private": true,
+  "type": "module",
+  "exports": {
+    ".": {
+      "types": "./src/index.ts",
+      "default": "./src/index.ts"
+    }
+  },
+  "devDependencies": {
+    "typescript": "^5.9.3"
+  },
+  "scripts": {
+    "typecheck": "tsc --noEmit"
+  }
+}

packages/types/src/index.ts

@@ -0,0 +1,117 @@
+// ── Tasks ───────────────────────────────────────────────
+
+export interface Task {
+  id: string;
+  question: string;
+  answer: string;
+}
+
+// ── Scenarios ───────────────────────────────────────────
+
+export type ScenarioName = "control" | "thinking" | "blind" | "mentor";
+
+export const SCENARIO_NAMES: readonly ScenarioName[] = [
+  "control",
+  "thinking",
+  "blind",
+  "mentor",
+];
+
+// ── Per-task evaluation row ─────────────────────────────
+
+export interface PerTaskRow {
+  id: string;
+  scenario: ScenarioName;
+  correct: boolean | null;
+  duckUsed: boolean;
+  toolCalls: number;
+  promptTokens: number;
+  genTokens: number;
+  duckTokens: number;
+  response: string;
+  expected: string;
+  question: string;
+}
+
+// ── Per-scenario summary (inside a model report) ────────
+
+export interface ScenarioSummary {
+  tasks: number;
+  reviewed: number;
+  correct: number;
+  accuracy: number;
+  pending: number;
+  duckUsed: number;
+  toolCalls: number;
+  avgCalls: number;
+  avgPromptTokens: number;
+  avgGenTokens: number;
+  avgDuckTokens: number;
+}
+
+// ── Per-model report ────────────────────────────────────
+
+export interface ModelReport {
+  generatedAt: string;
+  model: string;
+  mcpUrl: string;
+  scenarios: Record<ScenarioName, ScenarioSummary>;
+  prompts: Record<ScenarioName, string>;
+  perTask: PerTaskRow[];
+  meta: Record<string, unknown>;
+}
+
+// ── Aggregate ───────────────────────────────────────────
+
+export interface AggregateByModel {
+  tasks: number;
+  reviewed: number;
+  correct: number;
+  accuracy: number;
+  duckUsed: number;
+}
+
+export interface AggregateScenario {
+  tasks: number;
+  reviewed: number;
+  correct: number;
+  accuracy: number;
+  pending: number;
+  duckUsed: number;
+  toolCalls: number;
+  avgCalls: number;
+  byModel: Record<string, AggregateByModel>;
+}
+
+export interface AggregateModel {
+  tasks: number;
+  totalRows: number;
+  reviewed: number;
+  correct: number;
+  accuracy: number;
+  duckUsed: number;
+  toolCalls: number;
+  avgGenTokens: number;
+}
+
+export interface Aggregate {
+  perScenario: Record<ScenarioName, AggregateScenario>;
+  perModel: Record<string, AggregateModel>;
+}
+
+// ── Root report (report.json) ───────────────────────────
+
+export interface ReportRoot {
+  generatedAt: string;
+  mcpUrl: string;
+  prompts: Record<ScenarioName, string>;
+  models: Record<string, ModelReport>;
+  aggregate: Aggregate;
+  reviewedAt?: string;
+}
+
+// ── Reviews (review.json) ───────────────────────────────
+// Keys: "${model}|${taskId}|${scenario}"
+
+export type ReviewKey = `${string}|${string}|${string}`;
+export type Reviews = Record<ReviewKey, boolean>;

packages/types/tsconfig.json

@@ -0,0 +1,12 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ES2022",
+    "moduleResolution": "bundler",
+    "strict": true,
+    "noEmit": true,
+    "isolatedModules": true,
+    "skipLibCheck": true
+  },
+  "include": ["src"]
+}

pnpm-lock.yaml

@@ -253,6 +253,8 @@ importers:
         specifier: ^5
         version: 5.9.3
 
+  packages/types: {}
+
 packages:
 
   '@emnapi/runtime@1.11.3':

pnpm-workspace.yaml

@@ -1,2 +1,3 @@
 packages:
   - apps/*
+  - packages/*