https://app.pakericore.app/blog Technical notes on microstructure modeling, reduced-order network solvers, and structure workflows from the PKR Core team. en https://app.pakericore.app/blog/six-sections-for-the-solid-fifty-four-for-the-pore https://app.pakericore.app/blog/six-sections-for-the-solid-fifty-four-for-the-pore Wed, 07 Oct 2026 00:00:00 GMT A 64³ packing with a known 3D answer, cut into 64 one-voxel sections. Averaging them recovered every 3D number, but the number of sections that took ranged from six to fifty-four depending on which quantity you read off. https://app.pakericore.app/blog/copy-the-label-across-a-flip https://app.pakericore.app/blog/copy-the-label-across-a-flip Tue, 06 Oct 2026 00:00:00 GMT Flips and rotations returned structures with bit-identical porosity and interfacial area, yet solving them gave effective conductivities up to 34 % apart. Crop and coarsen moved the geometry too, so those are new samples rather than copies. https://app.pakericore.app/blog/one-voxel-of-coating https://app.pakericore.app/blog/one-voxel-of-coating Mon, 05 Oct 2026 00:00:00 GMT The same coating step was applied to three structures. What it cost each of them was set by their interfacial area, mean pore size moved in opposite directions, and the step that killed transport came without warning from porosity. https://app.pakericore.app/blog/one-gray-level-moves-porosity https://app.pakericore.app/blog/one-gray-level-moves-porosity Sun, 04 Oct 2026 00:00:00 GMT The threshold you pick when binarizing a CT or FIB-SEM volume has a cost, and it is not the same cost for every number. Porosity runs out of room after half a gray level. Effective conductivity absorbs nineteen. https://app.pakericore.app/blog/surface-mesh-comes-back-hollow https://app.pakericore.app/blog/surface-mesh-comes-back-hollow Sat, 03 Oct 2026 00:00:00 GMT A generated structure exports to the format your own solver reads. The volume formats arrive exact — the Abaqus mesh carried 118,159 elements against 118,159 solid voxels. Export the same packing as STL and voxelize it again, and the particles return as shells, with the two structures agreeing on only 37 % of their solid. https://app.pakericore.app/blog/zero-on-every-textbook-case https://app.pakericore.app/blog/zero-on-every-textbook-case Fri, 02 Oct 2026 00:00:00 GMT Conductivity offers three methods. Against five structures with known answers, two of them were right to within 0.02 % and the reduced method returned 0 every time. Against a random particle packing it inverts: above 24³ the reduced method is the only one that returns anything, and at 24³ the two usable methods are a factor of five apart. https://app.pakericore.app/blog/diffusion-adds-up-permeability-does-not https://app.pakericore.app/blog/diffusion-adds-up-permeability-does-not Wed, 30 Sep 2026 00:00:00 GMT Two layers of different porosity stacked into one structure with a single call. The series rule predicted the stacked diffusivity to within 2.9 %, but missed permeability by −11 % to +16 %, with the sign flipping between seeds. https://app.pakericore.app/blog/seven-descriptors-three-worth-keeping https://app.pakericore.app/blog/seven-descriptors-three-worth-keeping Mon, 28 Sep 2026 00:00:00 GMT Sixteen structures measured seven ways each. Two descriptors could not separate them from run-to-run scatter, three ranked them identically, and the three that survived recover the permeability ordering at a rank correlation of +0.91. https://app.pakericore.app/blog/press-harder-and-flow-pays-first https://app.pakericore.app/blog/press-harder-and-flow-pays-first Sun, 27 Sep 2026 00:00:00 GMT One packing compacted to six strains. Porosity fell by 1.6 times and permeability by 5.9, while effective conductivity barely moved — and two crops decide whether any of it is real. https://app.pakericore.app/blog/filler-stranded-outside-the-network https://app.pakericore.app/blog/filler-stranded-outside-the-network Sun, 20 Sep 2026 00:00:00 GMT A percolation threshold says when the filler starts to connect, not how much of it does. Two API calls per structure return the share sitting in clusters that reach nothing. https://app.pakericore.app/blog/what-the-sentence-actually-built https://app.pakericore.app/blog/what-the-sentence-actually-built Sat, 19 Sep 2026 00:00:00 GMT Five material specs written as sentences went through parameter assist, were built, and were measured in the same session. All five returned valid parameters and none matched the numbers in the sentence. Three of them produced the identical box. The response tells you which words were heard, and the measurement tells you which numbers were packed. https://app.pakericore.app/blog/how-much-filler-before-it-conducts https://app.pakericore.app/blog/how-much-filler-before-it-conducts Sun, 13 Sep 2026 00:00:00 GMT Spheres, fibers and platelets swept from 5 % to 40 % loading in the same 64³ box on five seeds each. Across the five seeds, median first-conduction loading along z landed at 10 % for fibers, 15 % for platelets, and 30 % for spheres, and 120 structures never reordered them. The catch is that a sphere at its threshold conducts almost nothing, and a fiber seed will not answer the same way twice. https://app.pakericore.app/blog/same-loading-narrower-pores https://app.pakericore.app/blog/same-loading-narrower-pores Sat, 12 Sep 2026 00:00:00 GMT Four particle blends held at the same 45 % solid, differing only in how that loading is split between 4 µm and 2 µm spheres. Mean pore chord fell from 7.24 to 4.24 µm and specific surface area rose by 74 %, in a ladder four seeds per blend never blurred. Permeability, on the same structures, did not follow. https://app.pakericore.app/blog/one-direction-reads-zero https://app.pakericore.app/blog/one-direction-reads-zero Fri, 11 Sep 2026 00:00:00 GMT Sweeping only how tightly fibers align, at a fixed 20 % loading. Along the fiber axis the effective conductivity ran from 3.8 to 14.6 W/mK; across it, twelve structures in a row returned 0. A cheap way to find out whether one measurement describes your material. https://app.pakericore.app/blog/one-call-per-candidate https://app.pakericore.app/blog/one-call-per-candidate Thu, 10 Sep 2026 00:00:00 GMT Screening candidate structures means running the same twelve-step pipeline over and over. One request runs all of it and returns whatever it managed, with the steps that refused named — so a sweep finishes instead of stopping at the first structure that says no. https://app.pakericore.app/blog/which-numbers-survive-half-the-resolution https://app.pakericore.app/blog/which-numbers-survive-half-the-resolution Thu, 10 Sep 2026 00:00:00 GMT One 64³ packing coarsened to 32³ and 16³, with the same five analyses on every rung. Tortuosity moved 0.05 %, porosity 7.9 %, conductivity 13.4 % — and permeability 381 %. The ordering tells you which results you can reuse on a coarser grid. https://app.pakericore.app/blog/how-big-a-box-before-the-number-settles https://app.pakericore.app/blog/how-big-a-box-before-the-number-settles Wed, 09 Sep 2026 00:00:00 GMT Changing only the random seed moved effective conductivity by 22 % at 48³ while porosity held to 0.1 %. Sweeping the box size instead shows where that scatter stops mattering — sixty structures, six requests, under twenty seconds. https://app.pakericore.app/blog/reproducing-a-run-two-days-later https://app.pakericore.app/blog/reproducing-a-run-two-days-later Sun, 06 Sep 2026 00:00:00 GMT The same recipes sent to the API two days apart, on a different deployed build, returned every effective property to the last published digit and the server-drawn figures byte for byte — so a study can be re-derived from its recipe rather than archived. https://app.pakericore.app/blog/checking-the-numbers https://app.pakericore.app/blog/checking-the-numbers Fri, 04 Sep 2026 00:00:00 GMT A homogeneous block, two slabs in series, and the same slabs in parallel have textbook answers. Running them through the API is how you find out what the solver is worth. https://app.pakericore.app/blog/three-structures-over-http https://app.pakericore.app/blog/three-structures-over-http Fri, 04 Sep 2026 00:00:00 GMT One run against the API turns three example recipes into structures, effective properties, cross-section images, and a sweep chart — with no browser and no local solver in the loop. https://app.pakericore.app/blog/import-ct-image-stacks https://app.pakericore.app/blog/import-ct-image-stacks Wed, 02 Sep 2026 00:00:00 GMT Sequential slice images become a voxel structure that enters the same viewer, inspection, and analysis pipeline as a generated packing. https://app.pakericore.app/blog/parametric-study-design-space https://app.pakericore.app/blog/parametric-study-design-space Wed, 02 Sep 2026 00:00:00 GMT One base recipe expands into up to 20 variants that run as a batch, so a parameter sweep becomes a single run you can compare and export rather than twenty manual rebuilds. https://app.pakericore.app/blog/particle-packing-tutorial https://app.pakericore.app/blog/particle-packing-tutorial Wed, 02 Sep 2026 00:00:00 GMT Pick a generator, set the packing parameters, look at the structure in 3D and in cross section, then run a property analysis on it — the core PKR Core loop in one pass. https://app.pakericore.app/blog/voxels-to-reduced-graph https://app.pakericore.app/blog/voxels-to-reduced-graph Wed, 02 Sep 2026 00:00:00 GMT A reduced graph keeps the connectivity that governs transport and throws away the rest, which is what makes a fast middle tier possible between an estimate and a full solver.