Skip to content

ADR 0001 — Repository scope: a pulsed-FRC kinematic and RTL hot-path laboratory

Status

Accepted.

Context

SCPN-MIF-CORE is one of six sibling repositories in the SCPN ecosystem. Several siblings already own canonical physics and control capability:

  • SCPN-FUSION-CORE owns the plasma solvers — Hall-MHD, FRC equilibrium, magneto-Rayleigh-Taylor, tilt stability, non-adiabatic flux evolution.
  • SCPN-CONTROL owns the control facade — Petri-net runtime, NMPC, replay, AER ingest.
  • SC-NEUROCORE owns the spiking-network-to-SystemVerilog emitter, the Q8.8 quantiser, and the AER HDL.
  • SCPN-PHASE-ORCHESTRATOR owns the Kuramoto family, distance coupling, the Doppler term, and the lock monitors.
  • SCPN-QUANTUM-CONTROL owns QAOA-MPC, pulse shaping, the QRNG, and the post-quantum trigger signer.

Without an explicit boundary, MIF-CORE would drift into re-implementing those solvers, producing divergent copies that fall out of sync and inflate the maintenance and audit surface.

Decision

MIF-CORE is scoped as the pulsed-FRC kinematic and RTL hot-path laboratory. It owns, and may only own: FRC kinematic merging, the chamber-fixed moving-frame layer, the merge-window monitor, the pulsed-shot lifecycle, the capacitor-bank model, the AER sensor bridge, the sub-50-nanosecond trigger fabric, the timing-aware formal property set, the MIF Lean safety proofs, and the Faraday recovery carrier.

Anything that falls under a sibling's canonical scope must be upstreamed there, not duplicated here. The anti-duplication checklist lives in the internal scope_and_ownership document and is enforced at review time.

Consequences

  • The repository stays small and focused; its public surface is the hot-path and integration layer, not a third plasma solver.
  • Capability that genuinely belongs here but overlaps a sibling is implemented as an upstream-pending local carrier (see ADR 0003).
  • Cross-repository physics enters through typed contracts and prescribed inputs, never by copying a sibling's solver.
  • Reviewers must check every new module against the ownership table before it lands.

Alternatives considered

  • Monolithic fusion package. Fold all physics, control, and RTL into one repository. Rejected: it destroys the per-domain ownership that lets each sibling evolve and be audited independently.
  • Pure integration shim with no owned capability. Rejected: the kinematic merging, lifecycle, and trigger fabric are genuinely MIF-specific and have no natural sibling home; a shim would force them into a sibling where they do not belong.