> ## Documentation Index
> Fetch the complete documentation index at: https://project-brain.rocheaiops.net/llms.txt
> Use this file to discover all available pages before exploring further.

# How B.R.A.I.N. changes its mind

> Corrections, conflicts, invalidation, and uncertainty without silent overwrite.

A correction is a new, attributable interpretation. It marks a predecessor **SUPERSEDED** or, where appropriate, **CONTESTED**; it does not rewrite history. Invalidation changes current applicability while keeping the source and prior processing available for inspection.

<a href="/assets/visuals/teach-recall-correct.svg" target="_blank" rel="noreferrer" aria-label="Open full-size diagram: Correction lineage"><img src="https://mintcdn.com/roche-ai-ops/zV65QLx2v4-e7KNO/assets/visuals/teach-recall-correct.svg?fit=max&auto=format&n=zV65QLx2v4-e7KNO&q=85&s=f358f5f2c3c0410a5a1773f4df279e93" alt="Correction lineage" noZoom width="1440" height="730" data-path="assets/visuals/teach-recall-correct.svg" /><span className="sr-only">Open full-size diagram: Correction lineage</span></a>

When equal-authority evidence conflicts or the intended context is unclear, B.R.A.I.N. should retain uncertainty or require clarification. It must not promote an inference to global truth merely because it is recent. See [provenance and lineage](/concepts/provenance-and-lineage).