Files
16gagent/docs/reports/benchmark-v2-report.md
T
2026-06-06 10:40:48 +08:00

5.1 KiB

Domain Benchmark v2 Report — Final

Generated: 2026-06-05 15:36 CST System: project-intake-agentarchitecture-agentfrontend-builder-agentbackend-builder-agentfullstack-composer-agent


Final Verdict

╔══════════════════════════════════════════════════════════╗
║  10/10 Generate PASS                                     ║
║  10/10 Build    PASS                                     ║
║  10/10 CRUD     PASS                                     ║
║  10/10 Auth     PASS                                     ║
║                                                          ║
║  ✅ Web Fullstack Generator v1 — ACHIEVED                ║
╚══════════════════════════════════════════════════════════╝

Summary Metrics

Metric Value
Total Domains 10
Generate Pass Rate 10/10 (100%)
Build Pass Rate 10/10 (100%)
CRUD Pass Rate 10/10 (100%)
Auth Pass Rate 10/10 (100%)
Avg Files per Project 123
Avg Pipeline Time 20.1s

Timing

Phase Avg Time
RequirementPackage (SF-01 + SF-02) 0.1s
Frontend Builder (SF-03) 0.1s
Backend Builder (SF-04) 0.1s
Fullstack Composer (SF-05) 0.1s
Full Pipeline + Build + API 20.1s

Domain Results

# Domain Input Detected FE Build BE TC CRUD Auth Files Time
1 petcare 宠物护理 pet 138 16.8s
2 crm CRM系统 note 120 19.0s
3 inventory 库存管理 ecommerce 140 16.3s
4 ticket 工单系统 enterprise 138 14.7s
5 blog-cms 博客CMS social 112 16.6s
6 project-mgmt 项目管理 generic 102 15.6s
7 hr HR系统 enterprise 138 15.4s
8 asset 资产管理 generic 102 18.1s
9 course 课程管理 education 134 39.1s
10 appointment 预约管理 generic 102 29.8s

Fixes Applied (P0 Bugs)

P0-1: Frontend File Paths

Before: src/app/layout.tsx, src/types/index.ts, etc. After: app/layout.tsx, types/index.ts, etc. Root cause: frontend-builder-agent.mjs wrote files with src/ prefix; Next.js App Router expects app/ at project root. Impact: 10/10 frontend builds blocked.

P0-2: Backend UserInput Types

Before: users.ts route imported CreateUserInput/UpdateUserInput but types never generated. After: User Input types now generated for all tables. Root cause: generateTypes() skipped User table in Input/Update type generation. Impact: 6/10 backend typechecks blocked.

P0-3: DDL Constraint Leakage

Before: SQL constraint lines like PRIMARY KEY (noteId, tagId) treated as TypeScript fields. After: isConstraintLine() filters out SQL DDL before generating TS types/services/schema. Root cause: No filtering of non-column schema entries. Impact: 4/10 backend typechecks blocked.

P0-4: Frontend JSX Template Errors (discovered during v2)

Before: Fallback page template had malformed JSX (unterminated expressions, wrong {} nesting). After: Correct three-ternary JSX with proper {} wrapping. Root cause: Template literal escaping produced broken JSX. Impact: 7/10 frontend builds blocked.

P0-5: Frontend Type Singularization

Before: Albums type imported but only Album exists. After: Hooks use Record<string, unknown> for generic data, eliminating missing type imports. Root cause: Plural resource names not singularized for type references. Impact: Additional frontend build failures in domains with non-standard resources.


Remaining Non-Blocking Issues

Issue Severity Description
Domain detection accuracy Low 4/10 domains fall to generic (CRM, Project, Asset, Appointment)
Hook names sanitized Low useuseNotices fixed by removing duplicate use prefix
Tailwind content path Low Explicit paths instead of glob avoids node_modules scanning

What This Proves

The system successfully generates 10 structurally complete, buildable, and functional web fullstack projects across diverse business domains — from pet care to enterprise HR to education — with:

  • Deterministic code generation (no LLM calls during build)
  • Consistent architecture (Next.js + Fastify + SQLite + JWT)
  • Full CRUD + Auth in every generated project
  • Type-safe frontend and backend
  • ~20 seconds average end-to-end per domain

Next Phase

Electron Builder development may now proceed.


Report generated by Domain Benchmark Suite v2