Christian Arango
Side projects

Side project · Desktop app

Guardian Tales companion app

Team advice for Guardian Tales mixes official notes, wikis, and tier lists of uneven quality. I built a Windows app that recommends teams and gear by game mode from the heroes I own, and labels every fact with how far its source can be trusted.

Author
Christian Arango
Role
Product Owner and builder, with Claude Code
Platform
Windows · Python
Status
Personal build · September 2026
The full hero catalog, ready to add to your collection.

The problem

Opinions get passed off as facts

Blogs and forum threads repeat tier lists as if they were measured. I wanted advice that shows where each recommendation comes from and never claims more certainty than its data has.

What I built

The real catalog, my roster, and a team builder

  • Catalog

    The game’s real catalog

    171 heroes, 155 cards, and the weapons, shields, accessories, merch, and relics, each read from the official wiki’s catalog pages.

  • Roster

    Each hero’s progress

    Level, limit breaks, evolution, and weapon progress for every hero, with the level caps the game really applies.

  • Teams

    A team builder by game mode

    Pick heroes and gear slot by slot, or load the best-sourced composition for that mode from the heroes you own.

One click loads the best-sourced team; gear stays empty until a source names it.
See the five trust levels in the data model
src/gt_agent/models.pypy
class SourceTier(str, Enum):
    """How much to trust a piece of data. Every fact in the DB carries one."""

    OFFICIAL = "official"          # dev letters, patch notes, official Discord
    STATS = "stats"                # gtales.top raid/comp data, datamined values
    WIKI = "wiki"                  # Fandom / NamuWiki reference data (manually verified)
    CORROBORATED_COMMUNITY = "corroborated_community"  # 2+ independent community sources agree
    COMMUNITY = "community"        # a single tier-list/blog/video — FYI only, never authoritative
Every fact in the database carries one of these levels.

Decisions

Trust is part of the data model

  • Five trust levels

    Official, stats, wiki, corroborated community, and a single community source. A lone tier list is for reference only.

  • No guessing to fill gaps

    The team leader is filled in only when the source names it, and a slot the game doesn’t have stays empty.

  • Rules instead of machine learning

    Recommendations rank by source trust first, then by how much of your roster covers them. The engine needs only Python’s standard library.

Where it stands

Usable today; the recommendations need more data

The catalog and roster tracking are complete, but only one sourced composition is loaded so far. Growing it is careful data intake, not more code.