# Zeno Divergent

> A learned model of model-relative surprise. It does not forecast the world; it
> forecasts when the models that describe the world stop being valid. Runs as a
> sidecar to existing predictors, simulators, bounds and policy rules.

Checkpoint: hub `main` = v0.3.1 — the v0.3.0 real-corpus checkpoint (USGS FDSN catalogue, promotion gate passed), loadable with `trust_remote_code=False`. The in-browser ONNX path on this site still runs the v0.2.3 synthetic-corpus artifact — advisory use only.

## Integration

- [Integration skill](https://zenodivergent.dev/zeno-divergent.skill.md): canonical instructions — input contract, normalization, fusion, output contract, misuse rules.
- [agents.md](https://zenodivergent.dev/agents.md): the same content under the conventional agent filename.
- [Integrate page](https://zenodivergent.dev/integrate): human-readable version with copyable snippets.

## Artifacts

- [Model repo](https://huggingface.co/mbarbosa1/zeno-divergent-v1): `mbarbosa1/zeno-divergent-v1`, loads without `trust_remote_code` via `pip install zeno-divergent`.
- [Demo Space](https://huggingface.co/spaces/mbarbosa1/zeno-divergent)
- [ONNX checkpoint](https://zenodivergent.dev/models/zeno-divergent-v1.onnx): dynamic reference axes, input `references` of shape `[1, T, K, 6]`.
- [Provenance](https://zenodivergent.dev/models/zeno-divergent-v1.provenance.json)
- [Model card](https://zenodivergent.dev/zeno-divergent-model-card.md)

## Method

- [Methods & schemas](https://zenodivergent.dev/methods)
- [Benchmark protocol](https://zenodivergent.dev/benchmark)
- [How it works](https://zenodivergent.dev/how-it-works)