Product engineering · Full-stack delivery

Garater

A card search, deck-building, scoring, and visual-analysis product for Grand Archive players.

Open Garater ↗

1.45K30-day unique visitors · late Aug 2026 snapshot

661records in the public scoring library

~1 weekpart-time build to first launch

Product walkthrough

Garater: Official card search — **Official card search**Search suggestions, official card images, rules text, and deck destinations in one workflow.

Garater: Deck analytics — **Deck analytics**A visible scoring model with dimension totals, card-level contribution, and cost-curve analysis.

Situation

The problem behind the product

Grand Archive players had to move between card databases, deck lists, and subjective evaluation methods. The product brings official card search, deck construction, structural scoring, and analysis into one workflow.

Task

What success required

Design and ship a useful public product quickly, while keeping it accessible without an account and leaving room for optional cloud sync.

Action

How I built it

Unified the core workflow

Connected the official card search API to card detail, Main Deck, Material Deck, and Sideboard flows, then added text import, export, and share codes.

Turned judgment into an explainable model

Built a five-dimension scoring system and surfaced the result through totals, contribution views, cost curves, radar charts, and tags rather than a single opaque number.

Added production-grade persistence

Implemented browser-local saving first, then added Cloudflare Workers, D1, sessions, and Google OAuth for optional cross-device deck sync.

Decisions

Trade-offs and judgment

Local-first instead of login-first

Requiring an account would add friction before a player experienced value. The app therefore works anonymously; cloud accounts are an enhancement, not a gate.

Graceful degradation

If the cloud API is unavailable, local decks and share codes still work. This keeps the main deck-building workflow useful during partial service failures.

Fast delivery with owned judgment

AI-assisted development accelerated implementation, while requirements, scoring logic, architecture, security boundaries, testing, and production release remained explicitly owned.

Result

What changed

A late-August 2026 analytics snapshot showed 1.45K unique visitors over the preceding 30 days. The product grew into a bilingual, production-deployed tool with a 661-record public rating library, two construction formats, analytics, feedback paths, and optional cloud sync.

Tools & platforms

JavaScriptHTML/CSSGrand Archive REST APICloudflare WorkersD1 / SQLGoogle OAuthlocalStorageCloudflare observability