Research-use software. CranioSwift is in active development. Outputs are reviewed by humans before any downstream use. Not intended for clinical decision making, treatment planning, or regulatory submission.
CranioSwift Try demo →

Phase summary · public version

An honest evaluation, summarised.

CranioSwift's model-development arc ran as a series of bounded research phases. Each phase tested one hypothesis end-to-end and closed with a written verdict. Negative results were documented as carefully as positive ones. This page is the public, high-level version — the full per-phase detail, including dataset counts, experimental configurations, and per-phase metrics, is shared under NDA via the investor materials packet.

The arc, in one paragraph

Several research phases tested the path to commercial-grade performance on the four-module workflow. One phase produced the shipping bundle that drives the live demo today. Subsequent phases identified the binding constraint for the next step: the size and diversity of clinical-distribution training data, not the model architecture or the loss design. The active phase builds the product surface and the fundraising case; the next phase resumes ML work once private clinical data and funding are in place.

What's visible publicly

  • The shipping bundle is live on /demo/ — six fixed-tail SkullFix research cases, four reviewable module-evidence cards, zero skull-overlap voxels.
  • The active surface is what you're looking at — the public investor view, the methodology page, this summary.
  • The Monash Innovation IDF channel is in flight; method-level technical disclosure runs through that channel under institutional confidentiality.
  • Research-line corpus expansion is in active follow-up with the relevant dataset maintainers.

What's under NDA

The detailed per-phase narrative — hypotheses tested, configurations tried, datasets used, exact metrics achieved, architectural choices made and rejected — lives in the confidential investor materials packet. We do not publish that detail publicly for two reasons:

  • IP posture: until the institutional provisional is filed via the Monash Innovation IDF, public-facing materials stay at outcome-level granularity. See /methodology/ §9 for the disclosure rule.
  • Operational discretion: detailed negative-result framing is useful diligence material under NDA but not appropriate as open-internet competitive intelligence.

To request the full investor materials packet (including the per-phase summary, dataset and licensing detail, and the use-of-funds breakdown), contact [email protected]. For the public methodology framing, see /methodology/. For the live shipping bundle, see /demo/.