batru

Official Python client for batru.gg — real, calibrated win-rate predictions and pick recommendations across three games (Dota 2, Deadlock, Marvel Rivals), served by production models trained on real matches.

Early preview (alpha). The surface is intentionally tiny and may change. This package is the home of the official batru.gg SDK; the MCP server for LLMs lives separately.

Install

pip install batru

Use

import batru

# Calibrated win probability for a (partial) draft — hero shortNames:
batru.winrate(radiant=["antimage", "axe"], dire=["lina", "tinker"])
# {'radiant_win_rate': 52.31, 'dire_win_rate': 47.69}

# Best next picks for our (Radiant) team:
batru.recommend(radiant=["antimage"], dire=["lina", "tinker"])

# Hero roster, to map display names to the shortNames used above:
[h["shortName"] for h in batru.heroes()][:3]

Deadlock & Marvel Rivals

Deadlock and Marvel Rivals use full 6v6 drafts keyed by integer hero ids (map names to ids with the *_heroes() rosters):

import batru

# Rosters: [{"id": int, "name": str}, ...]
batru.deadlock_heroes()
batru.marvel_heroes()

# Meta content (dicts, returned as-is):
batru.deadlock_tier_list()      # or marvel_tier_list()
batru.deadlock_counters()       # worst matchups per hero
batru.deadlock_synergy()        # best teammates per hero

# Calibrated comp-only draft win probability (6 ids per team) — this models
# the hero composition, not player skill:
batru.deadlock_predict_draft([1, 2, 3, 4, 5, 6], [7, 8, 9, 10, 11, 12])
# {'win_prob_team0': 0.53}
batru.marvel_predict_draft([1, 2, 3, 4, 5, 6], [7, 8, 9, 10, 11, 12])

Probabilities are calibrated: a reported 60% corresponds to an empirically observed ~60% win rate on holdout data — see the methodology.

Read-only, no key required. Please keep request volume reasonable.

Analysis only, no betting.

License

MIT © Batru (batru.gg)