Skip to content

FAQ

Does OddsFox Graph host data?

No. This repository is Hypertrial-owned MIT software. Documentation does not host datasets. Operators supply their own Pipeline golden mart export and local model weights.

Why is infer so slow?

LLM inference dominates wall-clock time. Deterministic topology skips the LLM for template-covered events (~91% on WC2026). For residual events, prefer --llm-backend inprocess (default) or mlx for single-machine decode speed; use --llm-backend server --concurrency N when you want concurrent request pipelining. See Inference backends.

Do I need Metal?

Metal-accelerated llama-cpp-python is recommended on Apple Silicon. CI and CPU-only environments can install the prebuilt CPU wheel instead. See Linux / CPU-only setup.

Are topology and markets connected?

Yes for markets covered by the proposition compiler. Those OUTCOME nodes carry a formal Proposition and link into topology via REFERS_TO (plus PRICES / COMPLEMENT / EXACTLY_ONE). Deterministic rules then add IMPLIES / EQUIVALENT / MUTEX. Residual / unrecognized market types may still lack propositions — see Logical layer and Known limitations.

Where do configuration defaults live?

In oddsgraph/config.py. The Configuration page mirrors those code defaults.

See also