Layer 1.5: architecture-general neural existence proof

Re-scopes Layer 2 into a cheaper, architecture-general neural collapse proof
before the LLM rung. Realises the same Wright–Fisher abstractions in real trained
generative models on a fully-synthetic sandbox with an exact oracle, reusing
knowledge.metrics/truth/seeding and the output contract so neural curves overlay
the Layer-1 analytic curves.

  - src/neural/: synthetic token-grammar sandbox (lossless identity + stochastic
    style), ExactOracle, HistogramModel bridge, generation loop, experiment runner
  - HARD GATE passed: histogram lineage reproduces Layer 1 exactly (neutral decay,
    exact H_eq, tracks run_lineage) — tests/test_neural_validation.py
  - torch models: autoregressive RNN + MLP (VAE implemented, not yet fidelity-
    passing); determinism seeding derived from the SeedSequence stream
  - N0 bridge (neural g*=0.047 ≈ Layer-1 0.048), N1 collapse-in-weights, N2 phase
    boundary, N5 architecture-generality (collapse + grounding-rescue in histogram
    + RNN + MLP). Manifests/configs committed; parquet gitignored, hashes tracked
  - additive backward-compatible save_artifacts extension; Makefile neural targets

Finding: neural smoothing partially resists H-collapse, so forward-KL and tail
survival are the sharp neural collapse metrics (H is smooth, per Layer 1).

92 tests green. Remaining (tasks/todo.md): N4 merge, N2 refine, N3/N6, VAE
fidelity, MNIST tier, figures. LLM/LoRA rung and C3 deferred.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/knowledge/experiment.py

@@ -211,11 +211,28 @@ def _content_hash(path: Path) -> str:
     return h.hexdigest()
 
 
-def save_artifacts(cfg: dict, df: pd.DataFrame, out_dir: Path) -> None:
+def save_artifacts(cfg: dict, df: pd.DataFrame, out_dir: Path,
+                   extra_libs: tuple[str, ...] = (),
+                   extra_manifest: dict | None = None,
+                   grid: list | None = None) -> None:
     """Write the reproducibility output contract (blueprint 2.7 / 4).
 
     Writes ``results.parquet``, ``resolved_config.yaml`` (the fully-expanded config), and
     ``manifest.json`` (library versions, master seed, git commit, content hash).
+
+    Args:
+        cfg (dict): The parsed experiment config.
+        df (pd.DataFrame): The long-form results.
+        out_dir (Path): Output directory.
+        extra_libs (tuple[str, ...]): Extra library names to record versions for (e.g.
+            ``torch``, ``torchvision`` for Layer 1.5). Missing libraries are skipped, so a
+            caller can pass optional deps unconditionally.
+        extra_manifest (dict | None): Extra key/value pairs to merge into the manifest
+            (e.g. model architecture, oracle checkpoint hash, determinism flags).
+        grid (list | None): Pre-expanded ``[{label, lineage_cfg}, ...]`` to record in the
+            resolved config. If None, it is computed via ``expand_sweeps`` for the Layer-1
+            ``lineage`` kind (a caller with a different schema, e.g. Layer 1.5, passes its
+            own expanded grid here).
     """
     out_dir.mkdir(parents=True, exist_ok=True)
     results_path = out_dir / "results.parquet"
@@ -227,24 +244,32 @@ def save_artifacts(cfg: dict, df: pd.DataFrame, out_dir: Path) -> None:
         "n_replicates": cfg["n_replicates"],
         "source_config": cfg,
     }
-    if cfg.get("kind", "lineage") == "lineage":
+    if grid is not None:
+        resolved["grid"] = grid
+    elif cfg.get("kind", "lineage") == "lineage":
         resolved["grid"] = [
             {"label": label, "lineage_cfg": lineage_cfg}
             for label, lineage_cfg in expand_sweeps(cfg)
         ]
     (out_dir / "resolved_config.yaml").write_text(yaml.safe_dump(resolved, sort_keys=False))
 
+    libraries: dict[str, str] = {}
+    for lib in ("numpy", "scipy", "pandas", "pyarrow") + tuple(extra_libs):
+        try:
+            libraries[lib] = version(lib)
+        except Exception:  # optional dep not installed -> omit rather than crash
+            pass
     manifest = {
         "experiment": cfg["experiment"],
         "master_seed": cfg["seed"],
         "git_commit": _git_commit(),
         "python": sys.version.split()[0],
-        "libraries": {
-            lib: version(lib) for lib in ("numpy", "scipy", "pandas", "pyarrow")
-        },
+        "libraries": libraries,
         "rows": int(len(df)),
         "results_sha256": _content_hash(results_path),
     }
+    if extra_manifest:
+        manifest.update(extra_manifest)
     (out_dir / "manifest.json").write_text(json.dumps(manifest, indent=2))