{
 "created": "2026-09-26T14:56:20",
 "code_commit": "30ad9b9a49f3e2905b396c4accf409bc26239815",
 "repo": "SubstrateCommons/imperial-rocks",
 "D0": 2.07e-09,
 "T2B": 3.1,
 "T": 3.0,
 "dt_save": 0.0002,
 "crop": "central 300^3",
 "rocks": {
  "LV60A": {
   "rho": 4.1e-05,
   "rho_note": "fitted to the LV60Y CPMG (thesis \u00a77.6.2)",
   "table": "Table 7-1 / 7-2",
   "figshare": "https://doi.org/10.6084/m9.figshare.1153795.v2",
   "porosity": {
    "ours": 0.3688355925925926,
    "talabi": 0.377
   },
   "s_over_v": {
    "ours": 59821.238561408434,
    "talabi": 57670
   },
   "T2lm_ms": {
    "replayed": 487.7,
    "direct": 487.6,
    "talabi_simulated": 512.0,
    "talabi_measured": 496.0
   },
   "direct_walk": {
    "recorded": true,
    "n_walkers": 200000,
    "dt_s": 0.0005,
    "source": "examples/validation/talabi_micro_ct_rocks.py, asserted by tests/validation/test_talabi_rocks.py"
   },
   "acceptance": {
    "T2lm_replayed_ms": 487.7,
    "T2lm_direct_ms": 487.6,
    "T2_talabi_sim_ms": 512.0,
    "rel_to_direct": 0.00029,
    "tol": 0.00842,
    "tolerance_terms": {
     "direct_floor": 0.00223606797749979,
     "pack_contact_floor": 0.004832018891264056,
     "dt_systematic": 0.0031,
     "tol": 0.00842432216958485
    },
    "inversion_sample_ms": 1.0,
    "n_points": 3000,
    "served_vs_channel": 1.1846357228506577e-12,
    "monotone": true,
    "envelope_worst_hz": 17.17,
    "pack_band_hz": 21.33,
    "envelope_in_band": true,
    "pass": true
   },
   "pack": {
    "n_walkers": 45000,
    "K": 128,
    "band_hz": 21.33,
    "certified": {
     "sigma_star": 0.005,
     "n_walkers": 45000,
     "peak_rss_gb": 57.8,
     "contact": {
      "floor": 0.004832018891264056,
      "err": 5.210667453070528e-10,
      "meets_target": true
     },
     "positions": {
      "floor": 0.008549848807884411,
      "err": 3.7322341429878124e-06,
      "meets_target": false
     },
     "note": "the contact tier is what this family is certified for; the gradient tier's floor is the split-half Monte-Carlo noise of the declared clinical envelope at this walker count, which the 60 GB build budget caps at 45000 walkers (measured: 1.27 GB per 1000 walkers on this window). 5e-3 on the gradient tier needs ~184,000 walkers and ~234 GB; rebuild with imperial-rocks/build.py --n-walkers to get it."
    }
   }
  },
  "F42A": {
   "rho": 4.1e-05,
   "rho_note": "fitted to the F42Y CPMG (thesis \u00a77.6.2)",
   "table": "Table 7-1 / 7-2",
   "figshare": "https://doi.org/10.6084/m9.figshare.1189259.v1",
   "porosity": {
    "ours": 0.33062044444444444,
    "talabi": 0.33
   },
   "s_over_v": {
    "ours": 44431.61992411219,
    "talabi": 43770
   },
   "T2lm_ms": {
    "replayed": 674.4,
    "direct": 679.0,
    "talabi_simulated": 677.0,
    "talabi_measured": 668.0
   },
   "direct_walk": {
    "recorded": true,
    "n_walkers": 200000,
    "dt_s": 0.0005,
    "source": "examples/validation/talabi_micro_ct_rocks.py, asserted by tests/validation/test_talabi_rocks.py"
   },
   "acceptance": {
    "T2lm_replayed_ms": 674.4,
    "T2lm_direct_ms": 679.0,
    "T2_talabi_sim_ms": 677.0,
    "rel_to_direct": 0.0068,
    "tol": 0.0094,
    "tolerance_terms": {
     "direct_floor": 0.00223606797749979,
     "pack_contact_floor": 0.005891480335269983,
     "dt_systematic": 0.0031,
     "tol": 0.009401550645743705
    },
    "inversion_sample_ms": 1.0,
    "n_points": 3000,
    "served_vs_channel": 2.3079316235907754e-12,
    "monotone": true,
    "envelope_worst_hz": 17.0,
    "pack_band_hz": 21.33,
    "envelope_in_band": true,
    "pass": true
   },
   "pack": {
    "n_walkers": 45000,
    "K": 128,
    "band_hz": 21.33,
    "certified": {
     "sigma_star": 0.005,
     "n_walkers": 45000,
     "peak_rss_gb": 57.8,
     "contact": {
      "floor": 0.005891480335269983,
      "err": 5.284309656516939e-10,
      "meets_target": false
     },
     "positions": {
      "floor": 0.00915603992321078,
      "err": 6.069027146376907e-06,
      "meets_target": false
     },
     "note": "the contact tier is what this family is certified for; the gradient tier's floor is the split-half Monte-Carlo noise of the declared clinical envelope at this walker count, which the 60 GB build budget caps at 45000 walkers (measured: 1.27 GB per 1000 walkers on this window). 5e-3 on the gradient tier needs ~184,000 walkers and ~234 GB; rebuild with imperial-rocks/build.py --n-walkers to get it."
    }
   }
  }
 },
 "caveats": {
  "LV60A_crop": "Talabi's 0.377 porosity is not reachable on the released LV60A image: over all 3,442,951 possible 300^3 crops the porosity spans [0.3555, 0.3713]. The offset he used is not recorded in the thesis, and a central cube is what his other six samples' porosities support. That is the whole of LV60A's porosity gap, and it is not the walk.",
  "rho": "rho = 41 um/s is HIS, fitted by him to his own LV60Y / F42Y CPMG (thesis \u00a77.6.2) and not to anything measured here. It is an effective value tied to his ~10 um voxel: the surface a segmentation gives is the Manhattan area at that resolution, which is what both walks relax against, so it does not transfer to another resolution.",
  "surface_rule": "His discrete surface rule is not this one: he kills a walker at an attempted move into a grain voxel with probability 2 rho s / (3 D), this walk weights it by exp(-2 (rho/D) d_perp) per specular reflection. They agree in the continuum limit.",
  "inversion": "The log-mean is the weakest link in this comparison, and the numbers above are all on a 1 ms sampling of the decay. Measured on F42A, which still holds 3.3 % of its signal at 3 s: 674.4 ms at 1 ms sampling, 668.0 at 10 ms, 309.7 at 20 ms -- a truncated decay lets the regularised solve split amplitude between a short and a very long component. LV60A, down to 1.0 %, is 487.7 / 487.6 / 487.4 at the same three. Over lambda in [0.003, 1] and five T2 grids the log-mean spans 3.1 % on LV60A and 4.2 % on Berea."
 }
}