Upload folder using huggingface_hub

README.md

@@ -22,7 +22,7 @@ S = pack.replay(waveform, relaxation=True)
 
 | Substrate | Replay tiers | Size | License |
 |---|---|---|---|
-| [`cactus-axons/single-bundle-g070-icvf061-3T`](cactus-axons/single-bundle-g070-icvf061-3T.md) | C0 C1 C2 | 5.84 MB | CC-BY-4.0 |
+| [`cactus-axons/single-bundle-g070-icvf061-3T`](cactus-axons/single-bundle-g070-icvf061-3T.md) | C0 C1 C2 C3 C4 | 46.65 MB | CC-BY-4.0 |
 | [`canonical-wm/g070-f055-3T`](canonical-wm/g070-f055-3T.md) | C0 | 18.23 MB | CC-BY-4.0 |
 
 Each substrate links to its own **substrate card** with geometry, provenance, fidelity and load/replay instructions. Tier legend: C0 gradient · C1 bulk relaxation · C2 surface relaxivity · C3 field · C4 magnetization transfer.

SHA256SUMS

@@ -1,2 +1,2 @@
-9a7dd9015b0a5cf5a4e0715c406c483404e7a10234572a76f0504d1142ada5d3  cactus-axons/single-bundle-g070-icvf061-3T.rpk
+5685ea727ea9ae67fa08055905c0c3de9202b7752a1514c2d854e248b4162adf  cactus-axons/single-bundle-g070-icvf061-3T.rpk
 bb57348ea115be2900cfb44eb0ad13b3ba307f1d2b967086136d9fbd23423106  canonical-wm/g070-f055-3T.rpk

cactus-axons/single-bundle-g070-icvf061-3T.croissant.jsonld

@@ -5,7 +5,7 @@
   },
   "@type": "Dataset",
   "name": "cactus-axons/single-bundle-g070-icvf061-3T",
-  "description": "dmipy-sim replay pack: compressed Monte-Carlo master walk (lowrank K=96) with replay envelope {'gradient': True, 'bulk_relaxation': True, 'surface_relaxivity': True, 'field': False, 'magnetization_transfer': False, 'diffusivity_fixed': True, 'acquisition': {'b_max': 3000000000.0, 'ogse_periods': [1, 2, 3, 5], 'B0_list': [3.0, 7.0, 11.7], 'note': 'temporal band set by max OGSE period / min delta'}}.",
+  "description": "dmipy-sim replay pack: compressed Monte-Carlo master walk (lowrank K=64) with replay envelope {'gradient': True, 'bulk_relaxation': True, 'surface_relaxivity': True, 'field': True, 'magnetization_transfer': True, 'diffusivity_fixed': True, 'acquisition': {'b_max': 3000000000.0, 'ogse_periods': [1, 2, 3, 5], 'B0_list': [3.0, 7.0, 11.7], 'note': 'temporal band set by max OGSE period / min delta'}}.",
   "license": "CC-BY-4.0",
   "citeAs": "Fick RHJ, dmrai-lab (2026). CACTUS single-bundle Monte-Carlo master walk. Geometry generator: Villarreal-Haro et al., CACTUS (https://github.com/Juanitovh/CACTUS).",
   "cr:provenance": {
@@ -44,19 +44,19 @@
     "tiers_present": [
       "C0-gradient",
       "C1-bulk-relaxation",
-      "C2-surface-relaxivity"
+      "C2-surface-relaxivity",
+      "C3-field",
+      "C4-mt-two-pool-qMT"
     ],
-    "tiers_deferred": {
-      "C3_field": "rpk susceptibility channel is 2D translationally-invariant; CACTUS needs 3D field maps",
-      "C4_mt": "MT binding needs R/25 sub-step -> ~3.5M mesh reflections/walker on sub-micron features"
-    }
+    "c3_field": "3D myelin susceptibility field basis sampled along the walk (13-ch); any B0/orientation/chi",
+    "c4_mt": "geometry-derived two-pool qMT (f_b, k_f from myelin S/V); per-walker EMERGENT MT (binding coupled to diffusion) needs the intractable R/25 mesh walk"
   },
   "cr:replayEnvelope": {
     "gradient": true,
     "bulk_relaxation": true,
     "surface_relaxivity": true,
-    "field": false,
-    "magnetization_transfer": false,
+    "field": true,
+    "magnetization_transfer": true,
     "diffusivity_fixed": true,
     "acquisition": {
       "b_max": 3000000000.0,
@@ -76,37 +76,37 @@
   },
   "cr:fidelity": {
     "metric": "max_abs_complex_signal_error",
-    "err_max": 0.0005992180903185556,
-    "floor_max": 0.015136345027932882,
+    "err_max": 0.00246580676329679,
+    "floor_max": 0.01381479186198168,
     "noise_floor": 0.007071067811865475,
     "within_2x_floor": true,
     "per_family": {
       "OGSE1": {
-        "err_max": 4.110324120514571e-05,
-        "floor_max": 0.01101313849143775
+        "err_max": 5.2823911831579506e-05,
+        "floor_max": 0.011853765800666903
       },
       "OGSE2": {
-        "err_max": 3.0114713312470272e-05,
-        "floor_max": 0.007819370051004686
+        "err_max": 4.542523613621486e-05,
+        "floor_max": 0.007125620416962809
       },
       "OGSE3": {
-        "err_max": 0.00021818730268882545,
-        "floor_max": 0.008300622238353213
+        "err_max": 0.00017216395972078792,
+        "floor_max": 0.013183539653365321
       },
       "OGSE5": {
-        "err_max": 0.00032166529279630717,
-        "floor_max": 0.008387572597074308
+        "err_max": 0.00246580676329679,
+        "floor_max": 0.009112497233388646
       },
       "PGSE": {
-        "err_max": 5.3429294722594866e-05,
-        "floor_max": 0.008324886028105324
+        "err_max": 8.017915658644107e-05,
+        "floor_max": 0.007313176619329261
       },
       "SHORTD": {
-        "err_max": 0.0005992180903185556,
-        "floor_max": 0.015136345027932882
+        "err_max": 0.00110073216164879,
+        "floor_max": 0.01381479186198168
       }
     },
-    "target_floor": 0.001,
+    "target_floor": 0.005,
     "meets_target": false
   },
   "distribution": [

cactus-axons/single-bundle-g070-icvf061-3T.md

@@ -39,13 +39,8 @@ A **replay pack** (`.rpk`) — a compressed Monte-Carlo master walk on a *synthe
 | C0 | Gradient | ✅ | positions — any q/b, direction, waveform (PGSE/PGSTE/OGSE/…) |
 | C1 | Bulk relaxation | ✅ | per-compartment T2/T1 via the compartment map |
 | C2 | Surface relaxivity | ✅ | boundary-local-time channel (any ρ) |
-| C3 | Field / susceptibility | — | stored susceptibility field maps (any B0, χ, orientation) |
-| C4 | Magnetization transfer | — | bound-fraction channel (MT/ihMT) |
-
-**Deferred tiers (with reason):**
-
-- **C3_field** — rpk susceptibility channel is 2D translationally-invariant; CACTUS needs 3D field maps
-- **C4_mt** — MT binding needs R/25 sub-step -> ~3.5M mesh reflections/walker on sub-micron features
+| C3 | Field / susceptibility | ✅ | susceptibility field basis (any B0, direction, χ_iso, χ_aniso) |
+| C4 | Magnetization transfer | ✅ | MT saturation (Z-spectrum / MTR) |
 
 > Diffusivity is fixed at walk time (a geometry property, not a replay knob).
 
@@ -53,15 +48,26 @@ A **replay pack** (`.rpk`) — a compressed Monte-Carlo master walk on a *synthe
 
 | Metric | Value |
 |---|---|
-| Max replay error | 0.0005992 |
-| MC noise floor | 0.01514 |
+| Max replay error | 0.002466 |
+| MC noise floor | 0.01381 |
 | Within 2× floor | True |
-| Target floor σ* | 0.001 |
+| Target floor σ* | 0.005 |
 | Meets σ* | False |
 | Walkers | 20000 |
 | Saved time points | 81 |
 | T_max (s) | 0.04 |
-| Compression | lowrank (K=96) |
+| Compression | lowrank (K=64) |
+
+## Magnetization transfer (two-pool qMT)
+
+Geometry-derived pool (from the myelin surface-to-free-volume ratio); replay the Z-spectrum / MTR as a pool-level knob.
+
+| Parameter | Value |
+|---|---|
+| Bound fraction f_b | 0.08115 |
+| Forward rate k_f (1/s) | 88.32 |
+| Surface-to-volume (1/m) | 4.416e+06 |
+| T2 bound (s) | 1e-05 |
 
 ## Load & replay
 
@@ -71,10 +77,14 @@ from dmipy_sim import bank
 
 pack = bank.pull("cactus-axons/single-bundle-g070-icvf061-3T", repo_id="dmrai-lab/substrate-bank")  # cached + hash-verified
 
-# Build any waveform G on the pack's own time grid (shape (n_meas, pack.n_t, 3), T/m),
-# e.g. a PGSE perpendicular to the fibre axis, then replay it:
-S = pack.replay(waveform, relaxation=True)   # C1 relaxation; add rho=... for C2 surface
-# perpendicular vs parallel gradients recover the fibre anisotropy (S_perp > S_par).
+# C0/C1/C2 — build any waveform G on the pack's time grid (n_meas, pack.n_t, 3),
+# e.g. PGSE perpendicular vs parallel to the fibre axis (recovers the anisotropy):
+S = pack.replay(waveform, relaxation=True)          # C1; add rho=... for C2 surface
+# C3 field — any B0 / direction / susceptibility, off the stored field basis:
+S = pack.replay(waveform, B0=7.0, chi_aniso=-1e-7, b0_dir=[0,0,1], susc_refocus=False)
+# C4 MT — geometry-derived two-pool Z-spectrum / MTR:
+Z = pack.mt_zspectrum(np.linspace(-3e4, 3e4, 41), w1_hz=200.0, t_sat=0.5)
+mtr = pack.mtr(offset_hz=1e4, w1_hz=200.0, t_sat=0.5)
 ```
 
 ## Generator configuration

cactus-axons/single-bundle-g070-icvf061-3T.rpk

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a7dd9015b0a5cf5a4e0715c406c483404e7a10234572a76f0504d1142ada5d3
-size 5838772
+oid sha256:5685ea727ea9ae67fa08055905c0c3de9202b7752a1514c2d854e248b4162adf
+size 46648180

manifest.json

@@ -5,15 +5,15 @@
     {
       "id": "cactus-axons/single-bundle-g070-icvf061-3T",
       "file": "cactus-axons/single-bundle-g070-icvf061-3T.rpk",
-      "sha256": "9a7dd9015b0a5cf5a4e0715c406c483404e7a10234572a76f0504d1142ada5d3",
-      "bytes": 5838772,
+      "sha256": "5685ea727ea9ae67fa08055905c0c3de9202b7752a1514c2d854e248b4162adf",
+      "bytes": 46648180,
       "license": "CC-BY-4.0",
       "replay_envelope": {
         "gradient": true,
         "bulk_relaxation": true,
         "surface_relaxivity": true,
-        "field": false,
-        "magnetization_transfer": false,
+        "field": true,
+        "magnetization_transfer": true,
         "diffusivity_fixed": true,
         "acquisition": {
           "b_max": 3000000000.0,
@@ -31,10 +31,10 @@
           "note": "temporal band set by max OGSE period / min delta"
         }
       },
-      "fidelity": 0.0005992180903185556,
+      "fidelity": 0.00246580676329679,
       "compression": {
         "method": "lowrank",
-        "K": 96,
+        "K": 64,
         "walker_preserving": true,
         "channels": {
           "compartment": {
@@ -46,7 +46,7 @@
             "channel": "boundary_local_time",
             "mode": "dense",
             "n_t": 81,
-            "scale": 3.8564205169677734e-05,
+            "scale": 3.629922866821289e-05,
             "nlevels": 127
           }
         }