Findings

These findings come from our papers and experiments on trained models in wide use today. Each names the programme that produced it.

Updated
4 October 2026

All findings

4 findings

  1. 27 June 2026

    The Forlais Theorem

    For trained models as they are commonly stored and shared, there is a firm limit to how far conventional exact compression can go, and it can be stated before any compression is attempted.

    Paper, FCN Compression

  2. 27 June 2026

    The Entropy Wall

    Conventional exact compression of trained models stops at a limit, and the limit is shared across the model families we measured.

    Paper, FCN Compression

  3. 27 June 2026

    The Mi Paradox

    A signal in trained models that appeared to offer a route to better compression turned out to be largely an artefact of measurement.

    Paper, FCN Compression

  4. 27 June 2026

    The r=2 Neural Law

    Trained models from several families and developers share the same regularity.

    Paper, FCN Compression