neural: real-MNIST external-validity tier (collapse + grounding)

Confirms model collapse and its arrest by grounding on REAL images, not
just the synthetic sandbox. A conv VAE (the canonical generative-collapse
model) is retrained each generation on its own generated digits, with a
fraction g of fresh real MNIST mixed in. Modes = digit class x stroke-
thickness bin (K=30, Zipf, ~18 tail modes); the oracle is a frozen CNN +
deterministic thickness at 98.5% mode accuracy (30x30 confusion matrix
recorded in the manifest as the measurement-noise floor).

Result (4 reps): dry (g=0) collapses to a single mode -- forward-KL
0.5->18, support 30->1, tail 1.0->0.06, H->0 -- while 10% grounding holds
all 30 modes (KL~0.6, full tail, H~0.9). Signs, not magnitudes (blueprint
3.5); the exact synthetic oracle stays the quantitative anchor. The VAE
needs ~10% grounding vs the synthetic histogram's ~5%, consistent with the
grounding finding that trained nets need more than the exact operator.

Plugs into the existing data-agnostic contract (metrics/grounding/output
reused verbatim): mnist_data (thickness bins, class x thickness bijection,
MnistSampler), mnist_oracle (ClassifierOracle + confusion matrix),
mnist_vae (ConvVAEGenerator), mnist_loop (run_mnist_lineage), kind=
mnist_lineage dispatch, MnistCfg/OracleCfg. Figures: plot_mnist (parquet-
only) + mnist_montage (eyeball diagnostic showing digits degenerate to one
blurry mode). make mnist / make env-mnist, kept out of the make neural
loop. 99 tests green (+5 torchvision-gated).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

figures/plot_mnist.py

@@ -0,0 +1,80 @@
+"""`mnist_collapse` figure — collapse and grounding-rescue on REAL MNIST images.
+
+External validity for the neural tier: a convolutional VAE retrained each generation on its own
+generated digits collapses — rare (class, thickness) modes die, forward-KL to the Zipf truth
+climbs, support shrinks — while a grounded arm (a fraction of fresh real MNIST images each
+generation) holds the tail. Modes are read by a frozen CNN oracle whose mode accuracy (the
+measurement-noise floor) is annotated from the run manifest. Signs, not magnitudes.
+
+Four panels, dry (g=0) vs grounded, mean ± 95% CI across replicates: (A) forward-KL trajectories;
+(B) support size (distinct modes alive); (C) tail truth-mass alive; (D) heterozygosity. Reads the
+committed bundle (parquet) + manifest.json only.
+
+Usage:  python figures/plot_mnist.py [results/mnist_collapse]
+"""
+
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+import matplotlib.pyplot as plt
+import numpy as np
+
+sys.path.insert(0, str(Path(__file__).parent))
+from _figlib import load_bundle, savefig  # noqa: E402
+
+sys.path.insert(0, str(Path(__file__).parents[1] / "src"))
+from knowledge.metrics import heterozygosity  # noqa: E402
+from neural.config import MnistCfg  # noqa: E402
+from neural.mnist_data import make_mnist_truth  # noqa: E402
+
+
+def _traj(df, g, col):
+    """Return (generations, mean, 95% half-width) of ``col`` for arm ``g`` across replicates."""
+    sub = df[df["g"] == g]
+    grp = sub.groupby("generation")[col]
+    gens = np.array(sorted(sub["generation"].unique()))
+    return gens, grp.mean().to_numpy(), 1.96 * grp.sem().to_numpy()
+
+
+def main(results_dir: str = "results/mnist_collapse") -> None:
+    df, cfg = load_bundle(results_dir)
+    syn = MnistCfg(**cfg["mnist"])
+    H_star = heterozygosity(make_mnist_truth(syn).p_star)
+    manifest = json.loads((Path(results_dir) / "manifest.json").read_text())
+    oracle_acc = manifest.get("oracle_mode_accuracy", float("nan"))
+
+    g_dry, g_wet = min(df["g"].unique()), max(df["g"].unique())
+    arms = [(g_dry, "#d62728", f"dry (g={g_dry:g})"), (g_wet, "#2ca02c", f"grounded (g={g_wet:g})")]
+
+    fig, axes = plt.subplots(2, 2, figsize=(13, 9))
+
+    def panel(ax, col, title, ylabel, hline=None):
+        for g, c, lab in arms:
+            gens, m, ci = _traj(df, g, col)
+            ax.plot(gens, m, "-o", color=c, ms=3, label=lab)
+            ax.fill_between(gens, m - ci, m + ci, color=c, alpha=0.2)
+        if hline is not None:
+            ax.axhline(hline[0], ls=":", color="gray", lw=1, label=hline[1])
+        ax.set(xlabel="generation", ylabel=ylabel, title=title)
+        ax.legend(frameon=False, fontsize=9)
+
+    panel(axes[0, 0], "forward_kl", "Collapse: dry forward-KL climbs, grounding holds it",
+          r"forward-KL $D(p^*\Vert\hat p)$")
+    panel(axes[0, 1], "support_size", f"Support collapses (of K={syn.K} modes)",
+          "distinct modes alive", hline=(syn.K, f"$K$={syn.K}"))
+    panel(axes[1, 0], "tail_truth_mass_alive", "Rare tail dies dry, held by grounding",
+          "tail truth-mass alive")
+    panel(axes[1, 1], "heterozygosity", "Diversity collapses dry, held by grounding",
+          "heterozygosity $H$", hline=(H_star, "$H^*$"))
+
+    fig.suptitle("mnist_collapse — model collapse and grounding-rescue on REAL MNIST images "
+                 f"(VAE; oracle mode acc {oracle_acc:.1%} = noise floor)", y=1.0, fontsize=13)
+    fig.tight_layout()
+    savefig(fig, results_dir, "mnist_collapse")
+
+
+if __name__ == "__main__":
+    main(*sys.argv[1:])