Skip to content

Capability tiers (C0–C4)

A replay pack declares what it can replay. The tiers are additive: a higher tier stores more per-walker channels, so it can answer more questions — but every tier shares the same walk and the same geometry.

Tier Name What it stores What you can decode
C0 Gradient the walk, as a compact trajectory basis any gradient waveform — PGSE, OGSE, STE, arbitrary G(t); any b, Δ, δ, direction; real RF / B₁ pulses (vector-Bloch replay, so refocusing follows the actual pulse)
C1 Bulk relaxation per-compartment T₂/T₁ any TE/TR weighting; a walk to T_max contains every shorter TE as a prefix
C2 Surface boundary local-time surface relaxivity ρ at the walls, and the apparent T₂ it produces
C3 Field per-walker susceptibility basis (13-channel) any B₀ and χ (isotropic + anisotropic), any orientation; GRE dephases, spin-echo refocuses
C4 MT two-pool qMT parameters (f_b, k_f, T₂ᵇ) magnetization-transfer ratio, Z-spectrum

The field tier is subtle on purpose

C3 is where a lossy shortcut can quietly break the physics. A spin-echo refocuses the static field to machine precision regardless of substrate shape; only diffusing spins carry a residual (the diffusion×susceptibility cross-term). That cancellation is delicate — a codec that looks fine on a gradient-echo can destroy it. So the fidelity check grades any C3 codec against the substrate's own field-noise floor, on the spin-echo cross-term, not just the easy gradient-echo case.

One walk, all tiers

A single master walk backs C0–C4 at once. The tiers are what got recorded alongside the trajectory, not separate simulations — and each is stored as a compact basis the decode contracts against, not an expanded trace. That is why replay is cheap and a pack is small.

The replay invariant