Uma Team Trials Builder

Build optimal Team Trials teams in Uma Musume: Pretty Derby from the characters on your account — and see exactly what the top 100 ranked players actually run.

🐎 Open the app →

---

Why this exists

Team Trials scoring is unusual. Two things dominate a team's score:

• Guaranteed-activation skills. Skills that always fire bank reliable points every race, and gold skills are worth ~2.4× a normal skill. A horse stuffed with relevant guaranteed skills out-scores a flashier one whose skills rarely trigger.
• Distinct running styles. Two horses on the same style fight each other for positioning bonuses — duplicating a style throws away ~4 000 points. So a good team runs three different styles.

This tool encodes those rules. You tell it what you own; it tells you the best teams and why.

Features

🗂️ Roster tracking	Mark owned cards, stars (★), potential level (1–5), with search & filters. Saved in your browser.
🏇 5 teams	Suggests 3 horses for each race category: Sprint / Mile / Medium / Long / Dirt.
🎯 Smart assignment	Picks 3 distinct running styles per team and assigns each horse the style (aptitude ≥ your threshold) where its skills score best.
🥇 Skill-tier scoring	Only guaranteed-activation white/gold skills count: golds ≫ normals, lower-priority ("bracketed") skills discounted. Uniques and raw aptitude grades are not scored.
🔄 Quick Change	Each slot has a dropdown of the next best-fitting horses (sorted by score) to swap in manually.
🔁 Unique-across-teams toggle	Keep every horse to one team (real Team Trials rules), or allow reuse.
🔒 Locks	Exclude a horse globally or per race category.
🧪 Event-skill toggle	Count training-event skills or ignore them.
💡 "Why this horse?"	Every pick explains its matched skills (origin · tier · reason) and the full score math.
🏆 Top-100 meta	A dedicated tab showing what the top 100 ranked players actually field — per race & running style. (see below)
✨ Meta-aware swaps	Horses popular in the top-100 are tagged Meta #N right in the swap dropdown, with an optional "prioritize meta" mode.

🏆 Top-100 meta — what the best actually run

Most team builders score horses in a vacuum. This one goes further: it analyzes the real teams of the 100 highest-ranked Team Trials players and turns them into a meta breakdown you can act on.

For every race × running style (Sprint / Mile / Medium / Long / Dirt × Front / Pace / Late / End) it shows:

• 🐎 Most-used horses — ranked by how many of the top 100 field them there, with usage %.
• 🎴 Support-card builds — the actual support cards the top players trained those horses on.
• 📊 Average stats — the real Speed / Stamina / Power / Guts / Wisdom the top teams hit.
• ⚡ Most-used skills — the guaranteed-activation skills that win at the top, by frequency.

And it feeds back into team-building: in the swap dropdown, top-100 picks are flagged Meta #N, so you can match the meta using the horses you already own.

As far as we know, no other Team Trials tool surfaces the live top-100 like this — usage, builds, supports, stats and skills, broken down by race and style. The meta dataset (public/data/tt_meta.json) is aggregated, fully anonymous (counts only), and refreshed from the current top 100 global ranking.

How a horse is scored

flowchart LR
  A[Card skills<br/>inherit · potential · event · unique] --> B{Guaranteed<br/>activation?}
  B -- no --> X[ignored]
  B -- yes --> C{Relevant to assigned<br/>style + race category?}
  C -- no --> X
  C -- yes --> D[Weighted score<br/>gold ≫ normal · brackets ↓]
  D --> F[Team = best 3 cards<br/>with 3 distinct styles]

Loading

• Guaranteed lists (general / per-style / per-distance / Trick) — src/scoring/guaranteedSkills.ts
• Matcher & weights (normalises ◎/○ badges, drops × debuffs, matches official name_en) — src/scoring/classify.ts
• Team-assignment search (per category, optional uniqueness across teams) — src/scoring/optimizer.ts

Tech

React + TypeScript + Vite + Tailwind v4, Zustand state, fully static (no backend), deployed to GitHub Pages via Actions.

npm install
npm run dev      # http://localhost:5173/TeamTrials_Builder/
npm run build    # production build → dist/

Data & auto-updates

The dataset (public/data/dataset.json + skills.json) is generated from gametora's English (global) game data plus the community master.mdb (for potential unlock ranks), and committed so the app needs no network at build time.

A scheduled GitHub Action (update-data.yml) re-pulls the data weekly, so new global horses and skills appear automatically — gametora tags each card with its global release date, and the Roster's Global (EN) only filter picks them up.

Support-card names (public/data/supports.json) come from the same gametora data layer. The top-100 meta (public/data/tt_meta.json) is an aggregated, anonymous snapshot (usage counts per race & style — no player identities) of the current global top 100.

Refresh manually any time:

bash scripts/fetch_raw.sh     # re-download raw data + rebuild public/data
# or rebuild from existing data/raw:
python3 scripts/build_dataset.py

Data field reference

• aptitude order: [turf, dirt, short, mile, medium, long, front, pace, late, end]
• skill rarity: 1 normal (white) · 2 gold · 3/4/5 unique · 6 evolution
• potential rank: 0 inherit (innate) · 1–5 awakening / talent level

Disclaimer

Unofficial fan tool. Uma Musume: Pretty Derby, all game data, and images are © Cygames. Not affiliated with or endorsed by Cygames. Character art and skill icons are loaded from gametora.com.