8. Community, governance and scale
8.1 Governance and community
- Open source from day one, developed in the open.
- Fiscal sponsorship through an organization like NumFOCUS, so the project can hold funds and outlive any one grant or lab.
- A focused core team. Open consortia are good at breadth but slow at the first, tightly coupled years. The core (worm Tracks A and C, the first platform extraction) is best run by a dedicated, milestone-driven team, for example as a focused research organization (FRO) or within one host institute, with the open consortium growing around it as interfaces stabilize.
- Steering council elected by active contributors; a code of conduct.
- CAKE Enhancement Proposals (CEPs), like Python PEPs and NumPy NEPs, for any change to the CKL language, file formats, coupler interfaces or benchmark protocols.
- Working groups, each owning part of the code and its validation:
- physics layers (electrical, synaptic, chemistry, body/environment, observation);
- organisms (worm, fly, fish, later mammals);
- inference and experiment design;
- infrastructure and performance;
- benchmarks and NeuroMIP.
- Contribution model: biology enters as CKL mechanisms and organism recipes, each with tests and provenance. Physics components enter through coupler interfaces, each with a verification suite. As in OpenMM force fields or Geant4 physics lists, nothing merges without validation.
- Affiliated packages (as in Astropy) for analysis tools, viewers and organism-specific extensions that live outside the core.
- Wet-lab partners are first-class members: they supply blind-round data and run experiments from the active-learning queue.
- Credit: a software paper per major release (e.g. JOSS), a Zenodo DOI per release, and authorship rules for consortium papers decided up front.
- Documentation organized as tutorials, how-to guides, reference and explanations, with an annual workshop and hackathon.
8.2 Scale and timeline (rough)
| Period | Scope | Team (indicative) |
|---|---|---|
| Year 0–1 | Worm M0–M3 and Track B1 on existing tools (Jaxley, MJX); commission datasets; first papers (identifiability, H2) | ~10–15 engineers and scientists, 2–3 partner labs |
| Year 2–3 | Worm M4–M7, decision on H*; extract CKL, coupler and data model from working code; fly VNC + body; first NeuroMIP protocol; first blind round | ~30–50 across several institutions |
| Year 4–6 | Full fly CNS; larval zebrafish; volunteer computing; regular blind rounds | Consortium of ~100+ contributors |
| Year 6+ | Mammalian cortical volumes; CAKE as standard infrastructure for connectome-based emulation | Community-governed |
Team sizes are indicative only; they depend on funding and on whether the worm milestones confirm H*.