[qmc] JOE: chiral-graviton NQS for Challenge 15

.github/workflows/test.yml

@@ -13,6 +13,7 @@ jobs:
       - uses: actions/setup-python@v5
         with:
           python-version: "3.12"
+      - uses: astral-sh/setup-uv@v5
       - run: pip install pytest pytest-cov numpy scipy matplotlib
       - run: make test
 

scripts/julia-daemon.sh

@@ -82,13 +82,20 @@ runner_path() {
   runner=$(cd "$(dirname "$0")" && pwd)/$(basename "$0")
 }
 
-install_shim() {
+validate_shim_path() {
   runner_path
-  shim_path="${1:-$HOME/.local/bin/julia}"
-  case "$shim_path" in /*) ;; *) shim_path="$PWD/$shim_path";; esac
-  if [ -e "$shim_path" ] && [ "$shim_path" -ef "$runner" ]; then
+  validated_shim_path="${1:-$HOME/.local/bin/julia}"
+  case "$validated_shim_path" in
+    /*) ;;
+    *) validated_shim_path="$PWD/$validated_shim_path";;
+  esac
+  if [ -e "$validated_shim_path" ] && [ "$validated_shim_path" -ef "$runner" ]; then
     fail "refusing to replace the daemon runner itself"
   fi
+}
+
+install_shim() {
+  shim_path="$1"
   resolve_real_julia
   if [ -e "$shim_path" ]; then
     first_line=$(IFS= read -r line <"$shim_path" && printf '%s' "$line")
@@ -388,8 +395,9 @@ case "$cmd" in
     ;;
   install-shim)
     [ $# -le 1 ] || fail "install-shim accepts at most one path"
+    validate_shim_path "${1:-}"
     install_daemonmode
-    install_shim "${1:-}"
+    install_shim "$validated_shim_path"
     ;;
   alias)
     [ $# -eq 0 ] || fail "alias takes no arguments"

tracks/qmc/solutions/JOE/README.md

@@ -0,0 +1,81 @@
+# JOE — Challenge 15
+
+## Team
+
+| | |
+|---|---|
+| **Team name** | JOE |
+| **Members** | Bei Qiao (乔北), Institute of Physics, Chinese Academy of Sciences (IOP) |
+
+## Challenge
+
+| Row | |
+|---|---|
+| **Challenge** | Construct an exchange-antisymmetric, SO(3)-equivariant neural quantum state for the ν = 1/3 chiral graviton and compute the gap Δ = E(L=2) − E(L=0). |
+| **Catalog issue** | Addresses #15. |
+| **Track** | `qmc` — Variational Monte Carlo / Neural Quantum States. |
+
+## Working repository
+
+- [Immutable reviewer-ready snapshot `a74b92a`](https://github.com/Joe-Nor/JOE-Harnessing-Quantum-2026/tree/a74b92a7da96f1eaceebd57f550e0405cf6c04f5)
+- [Concise Challenge 15 review](https://github.com/Joe-Nor/JOE-Harnessing-Quantum-2026/blob/a74b92a7da96f1eaceebd57f550e0405cf6c04f5/challenges/15-chiral-graviton/CHALLENGE15_REVIEW.md)
+- [Projected-CF Pfaffian derivation and implementation scope](https://github.com/Joe-Nor/JOE-Harnessing-Quantum-2026/blob/a74b92a7da96f1eaceebd57f550e0405cf6c04f5/challenges/15-chiral-graviton/PROJECTED_CF_PFAFFIAN.md)
+- [Reproduction guide](https://github.com/Joe-Nor/JOE-Harnessing-Quantum-2026/blob/a74b92a7da96f1eaceebd57f550e0405cf6c04f5/challenges/15-chiral-graviton/REPRODUCIBILITY.md)
+- [Machine-readable \(N=4\) Pfaffian oracle](https://github.com/Joe-Nor/JOE-Harnessing-Quantum-2026/blob/a74b92a7da96f1eaceebd57f550e0405cf6c04f5/challenges/15-chiral-graviton/results/projected_cf_pfaffian_N4_oracle.json)
+
+## Verified finite-size results
+
+For \(N=4\), the joint VMC calculation gives
+
+```math
+E_0/E_C=1.871850\pm0.001026,
+\qquad
+E_2/E_C=2.004406\pm0.003022,
+```
+
+and
+
+```math
+\Delta_4/E_C=0.132556\pm0.002941.
+```
+
+The corresponding exact-diagonalization value is
+
+```math
+\Delta_4^{\mathrm{ED}}/E_C=0.131856754927,
+```
+
+so the VMC and ED gaps differ by \(0.24\) Monte Carlo standard errors. This
+closes the small-system validation loop; it is not a thermodynamic-limit gap
+claim. The snapshot contains 84 project tests and 61 independent Stage-1
+Fock-oracle tests.
+
+The newer projected-CF route constructs the pure rank-two broken-pair tangent
+
+```math
+\Psi_{2,M}(X)
+=(-1)^P
+\left.
+\frac{d}{d\epsilon}
+\operatorname{Pf}\!\left(G+\epsilon B^M\right)
+\right|_{\epsilon=0}.
+```
+
+At \(N=4\), its two active channels span the complete
+\(\mu_2=2\) multiplicity space. The deterministic certificate reports
+
+```text
+|E₂(Pfaffian) − E₂(ED)| = 4.44×10⁻¹⁶
+maximum fivefold splitting = 4.44×10⁻¹⁶
+maximum coordinate-symmetry residual = 2.83×10⁻¹⁵
+maximum ‖(L²−6)ψ‖ residual = 1.31×10⁻¹³
+```
+
+This Pfaffian result is a deterministic small-system oracle, not a Monte Carlo
+error bar. It does not establish a large-\(N\) neural state, a thermodynamic
+gap, or chirality. Production coordinate evaluation is polynomial and does
+not enumerate Fock determinants, but the current large-\(N\) VMC extrapolation
+remains unfinished.
+
+Challenge 121 is intentionally absent from this branch and PR. It is submitted
+separately by team 格格巫.