Add 3-stage CIF fallback: pymatgen → CifParser(occ_tol=2) → ASE

Fix GitHub link (MujeebOnawole/PoreGCN) + robust CIF fallback parser

app.py: remove inaccurate claim that Tb-MOF reproduces same Scenario pattern under both XAI modes

app.py: rename Full to Slow + plain-language Fast/Slow XAI explainer (predicted values are the same; only attribution method differs); add optional Goatcounter visitor-tracking pixel via GOATCOUNTER_CODE env var

app.py: drop the offline-pipeline bullet from the How-to-use list

app.py: per-property XAI in Fast mode (~35s, all rows correctly classified); Full stays target-only with Fast surrogate for other rows; About tab rewritten with honest how-to-use guidance

app.py: load property_means from per-ensemble JSON; classify_scenario uses manuscript dual criteria (ratio>=0.70 OR mean signed direction)

xai_engine.py: load property_means from per-ensemble JSON; classify_scenario uses manuscript dual criteria (ratio>=0.70 OR mean signed direction)

app.py: Fast/Full XAI toggle + skip +0.00 cavity labels + About Ensembles and XAI explainer

visualize.py: Fast/Full XAI toggle + skip +0.00 cavity labels + About Ensembles and XAI explainer

xai_engine.py: Fast/Full XAI toggle + skip +0.00 cavity labels + About Ensembles and XAI explainer

app.py: swap bundled CoRE MOF validation example to Tb-MOF (dual-method 5 of 7 Scenario A)

example_cifs: remove Zn-MOF-ChemSci2019 (replaced by dual-validated Tb-MOF)

example_cifs: bundle Tb-MOF (CrystEngComm 2023, dual-XAI-method validated CoRE MOF entry)

app.py: wire 4th example (Zn-MOF-ChemSci2019 -> core_mof) into KNOWN_MOFS + About Ensembles

example_cifs: bundle Zn-MOF (Chem. Sci. 2019, CoRE MOF entry) for trustworthiness validation

app.py: legend matches viewer (cornflower-blue metals + cavity attribution chips); slim footer with email

visualize.py: legend matches viewer (cornflower-blue metals + cavity attribution chips); slim footer with email

app.py: smaller cavity beads + per-bead attribution labels + wrap atoms to [0,1) + clip mesh to cell

visualize.py: smaller cavity beads + per-bead attribution labels + wrap atoms to [0,1) + clip mesh to cell

visualize.py: revert axis tip labels to crystallographic a/b/c

app.py: manual white wireframe + xyz labels; one example per ensemble; About Ensembles tab

visualize.py: manual white wireframe + xyz labels; one example per ensemble; About Ensembles tab

app.py: opaque mesh (0.55) + opaque beads + white wireframe + xyz axis labels

visualize.py: opaque mesh (0.55) + opaque beads + white wireframe + xyz axis labels

fix: align void mesh with 3Dmol standard CIF frame; switch viewer to black bg

visualize.py: harden void mesh (per-axis grid clamp, vertex cap, full failure fallback to atoms+beads)

requirements.txt: Mercury-style void isosurface (Path D, contact-surface mesh + attribution glyphs)

visualize.py: Mercury-style void isosurface (Path D, contact-surface mesh + attribution glyphs)

visualize.py: cavity-scale pore spheres (Lisensky and Yaghi 2022 idiom)

Pore markers: fixed 0.6A radius (color carries signal, not size)

Replace per-node occlusion with gradient x input attribution (O(1) backward, ~150x faster)

Cap per-pore occlusion XAI to 40 largest pores (free CPU compute budget)

Polyhedra-style aesthetic: indigo metal SBUs

Robust Zeo++: try -nt2 then -vornet, copy CIF first

Wire CSV download in tab 4

Hover shows attribution + add CSV export helper

Drop -r from Zeo++ command (incompatible with -vornet)

Add 3Dmol hover tooltip + element legend

Fix Zeo++ command (-vornet with explicit output path) + parser for both formats

Use bundled Zeo++ tarball as primary source

Bundle Zeo++ source tarball (HF outbound network corrupts download)

Add stick bonds via 3Dmol auto-bond detection

Use Jmol colorscheme (bwr gradient name was invalid in 3Dmol.js)

Fix 3D viewer: render all atoms as spheres (sticks need bonds CIF lacks)

Initial PoreGCN web tool deployment

initial commit