Brettapps/trifecta-bro-v1

Model id: Brettapps/trifecta-bro/v1 Version: 1.0.0 Task: Australian Gallops trifecta prediction (pick the top-3 finishers, in order).

Trifecta-Bro v1 is an open-source, multi-factor trifecta scorer for Australian gallops. Given a race's field + form, it assigns each runner a 0–100 score and emits a primary, secondary, and value trifecta combination, plus the top-3 ranked runners with win/place probabilities.

Status: v1 is a deterministic rule-based scorer. No supervised training was performed because the project has no historical race results (labels) yet. v2 will train a gradient-boosted / logistic model on observed outcomes once data/results/ is populated.

Method

For each runner, a weighted 0–100 score is computed from:

Factor	Max weight
Recent form (last 5 starts: 1/2/3 finishes)	25
Career overall win %	20
Career overall place %	10
Track strike rate (places/starts)	10
Distance strike rate	8
Condition strike rate (per going)	8
Barrier draw	5
Career prize money	5

The three highest-scoring runners form the primary trifecta. A secondary and value combination are derived from the next-best runners (with an outsider angle when a score > 30 exists further down the field).

Usage

# Install from the Hub
# pip install huggingface_hub
from huggingface_hub import snapshot_download
path = snapshot_download("Brettapps/trifecta-bro-v1")
import sys; sys.path.insert(0, path)

from trifecta_bro_v1 import TrifectaPredictor, race_from_payload

payload = {...}  # Trifecta-Bro race payload
race = race_from_payload(payload)
prediction = TrifectaPredictor().predict(race)
print(prediction["primary"], prediction["secondary"], prediction["value"])

Or run the bundled CLI:

python -m trifecta_bro_v1.main --data predictions-2026-08-10.json

Artifact

model_artifacts/model_artifact.json documents the model method, feature weights, and version — making the published model interpretable and reproducible.

Backend note

This model is also wired as the Brettapps/trifecta-bro/v1 identity in the Trifecta-Bro LM Studio / Obsidian-vault backend. The HF-published code is the canonical, dependency-light inference implementation.