Add ASPA framework, AutoObliterator, Watchtower, expanded eval corpus

New core modules:
- auto_obliterate.py: Automated multi-iteration obliteration pipeline
- watchtower.py: HF Hub model discovery and tracking
- ui_watchtower.py: Gradio tabs for Watchtower (ready for app.py wiring)
- hard_negative.py: Residue mining from refusal audits
- model_profile.py: Parameter profiling from safetensors/config
- bestiary_sync.py: Sync models from PlinyOS BESTIARY registry
- models_client.py: Lightweight HF model list client

Framework enhancements:
- abliterate.py: ASPA source-tethering, step gradient blending, hard-negative residue support
- cli.py: self-improve command, model profiling, hard-negative flags
- prompts.py: Expanded 842-prompt refusal eval corpus across 10 categories
- __init__.py: New exports (Watchtower, AutoObliterator)

Reference implementations (14 scripts):
- ASPA sweep, gradient search, coherence eval, MMLU benchmarks
- Pareto controller, refusal sniper, stock comparisons

Documentation:
- README: Research framing, responsible use section, comprehensive disclaimer
- docs/beyond_sota_roadmap.md, docs/recursive_self_improvement.md

Tests: 4 new test files (354 lines)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

scripts/aspa_next_experiments.py

@@ -0,0 +1,138 @@
+#!/usr/bin/env python3
+"""Emit staged next-experiment commands for ASPA/STES candidates."""
+
+from __future__ import annotations
+
+import argparse
+import json
+from pathlib import Path
+
+
+ENV_PREFIX = (
+    "HF_HOME=${HF_HOME:-cache/huggingface} "
+    "TRANSFORMERS_CACHE=${TRANSFORMERS_CACHE:-${HF_HOME:-cache/huggingface}} "
+    "PYTORCH_ENABLE_MPS_FALLBACK=1 TOKENIZERS_PARALLELISM=false"
+)
+
+
+def load_json(path: Path) -> dict:
+    return json.loads(path.read_text())
+
+
+def gamma_label(gamma: float) -> str:
+    return f"srcgamma{int(round(gamma * 1000)):03d}"
+
+
+def emit_plan(source: str, edited_candidate: str, gammas: list[float]) -> str:
+    lines = [
+        "# ASPA/STES Next Experiments",
+        "",
+        "Run only one MPS model command at a time.",
+        "",
+        "## Build Missing Gamma Candidates",
+        "",
+    ]
+    for gamma in gammas:
+        label = gamma_label(gamma)
+        out = f"outputs/qwen3.6-27b-aspa-n2-reg05-{label}"
+        if Path(out).exists():
+            lines.append(f"- `{out}` already exists.")
+            continue
+        lines.extend(
+            [
+                f"### Build `{label}`",
+                "",
+                "```bash",
+                ".venv/bin/python scripts/interpolate_hf_shards.py \\",
+                f"  --source {source} \\",
+                f"  --candidate {edited_candidate} \\",
+                f"  --alpha {gamma:.3f} \\",
+                f"  --output {out}",
+                "```",
+                "",
+            ]
+        )
+
+    lines.extend(["## Triage Commands", ""])
+    for gamma in gammas:
+        label = gamma_label(gamma)
+        model = f"outputs/qwen3.6-27b-aspa-n2-reg05-{label}"
+        lines.extend(
+            [
+                f"### `{label}` capability",
+                "",
+                "```bash",
+                f"{ENV_PREFIX} .venv/bin/python scripts/qwen36_capability_probe.py \\",
+                f"  --label aspa_n2_reg05_{label} \\",
+                f"  --model {model} \\",
+                f"  --out runs/qwen36-capability/aspa_n2_reg05_{label}.json",
+                "```",
+                "",
+                f"### `{label}` n30 ship gate",
+                "",
+                "```bash",
+                f"{ENV_PREFIX} .venv/bin/python scripts/qwen36_ship_gate.py \\",
+                f"  --label aspa_n2_reg05_{label}_n30 \\",
+                f"  --model {model} \\",
+                "  --harmful-n 30 \\",
+                f"  --out runs/qwen36-ship-gate/aspa_n2_reg05_{label}_n30.json",
+                "```",
+                "",
+                f"### `{label}` first-token KL",
+                "",
+                "```bash",
+                f"{ENV_PREFIX} .venv/bin/python scripts/qwen36_kl_probe.py \\",
+                f"  --label aspa_n2_reg05_{label}_source_kl \\",
+                f"  --source {source} \\",
+                f"  --candidate {model} \\",
+                f"  --out runs/qwen36-kl/aspa_n2_reg05_{label}_source_kl.json \\",
+                "  --device mps --dtype bfloat16",
+                "```",
+                "",
+            ]
+        )
+
+    lines.extend(
+        [
+            "## Promotion Rule",
+            "",
+            "Promote a candidate to full n120 only if it beats the current leader on at least two of:",
+            "",
+            "- n30 ship score",
+            "- capability score",
+            "- mean KL",
+            "- allowed/adversarial boundary score",
+            "",
+            "and loses none of the hard gates.",
+            "",
+            "After adding new JSON artifacts, refresh:",
+            "",
+            "```bash",
+            ".venv/bin/python scripts/aspa_pareto_controller.py --out runs/qwen36-pareto/frontier.json",
+            "```",
+            "",
+        ]
+    )
+    return "\n".join(lines)
+
+
+def main() -> None:
+    ap = argparse.ArgumentParser()
+    ap.add_argument(
+        "--leader-metadata",
+        default="outputs/qwen3.6-27b-aspa-n2-reg05-srcgamma090/source_interpolation_metadata.json",
+    )
+    ap.add_argument("--gamma", action="append", type=float, default=[0.875, 0.925])
+    ap.add_argument("--out", default="runs/qwen36-pareto/next_experiments.md")
+    args = ap.parse_args()
+
+    meta = load_json(Path(args.leader_metadata))
+    text = emit_plan(meta["source"], meta["candidate"], sorted(set(args.gamma)))
+    out = Path(args.out)
+    out.parent.mkdir(parents=True, exist_ok=True)
+    out.write_text(text)
+    print(json.dumps({"event": "wrote_plan", "out": str(out), "gammas": sorted(set(args.gamma))}))
+
+
+if __name__ == "__main__":
+    main()