What is scio?

It is an open-source Python library for designing, using and evaluating confidence scores in Neural Networks. This topic has recently gained significant attention, as deploying of Deep Learning models raises pressing reliability concerns – specifically for critical applications.

It is a friendly fork of the original project, initiated by Élie Goudout in 2024, internally at Thales cortAIx Labs France, and its first open-source release dates back to July 25th, 2025, under the MIT License.

The name “scio” carries two meanings.

1. In Latin, sciō can mean to understand, to have practical knowledge (sharing etymology with science). The well-known saying “scio me nihil scire” (“I know that I know nothing”) aptly illustrates the purpose of confidence scores: identifying predictions that should not be trusted due to insufficient informed support.

2. It is also an acronym for Statistical Confidence from Internal Observations, reflecting both our motivation and approach.

The following sections introduce the broader context and outline the conceptual framework that guides the development and usage of scio.